From a04c599c49465b67ea40a1f0de622706ab3da955 Mon Sep 17 00:00:00 2001
From: Dong1017 <55723162+Dong1017@users.noreply.github.com>
Date: Sat, 31 Oct 2026 11:54:31 +0800
Subject: [PATCH 01/11] docs: describe repository and framework skill
conditions
Preserve source provenance and evidence states while making the declared scope explicit.
Agent-assisted-by: Codex / gpt-6-astra [effort=xhigh] (author)
Signed-off-by: Dong1017 <55723162+Dong1017@users.noreply.github.com>
---
.../diffusers-invocation-state/metadata.json | 14 +++-
.../metadata.json | 14 +++-
.../dcp-async-save-performance/metadata.json | 15 ++++-
.../audit-afd-p2p-exchanges/metadata.json | 14 +++-
.../repositories/comfy-ui/metadata.json | 30 +++++++--
.../metadata.json | 13 +++-
.../metadata.json | 13 +++-
.../repositories/dwarfstar/metadata.json | 14 +++-
.../metadata.json | 14 +++-
.../audit-h3-runtime-residency/metadata.json | 14 +++-
.../metadata.json | 14 +++-
.../repositories/husky/metadata.json | 30 +++++++--
.../repositories/lily/metadata.json | 14 +++-
.../repositories/lithos-metal/metadata.json | 13 +++-
.../audit-llama-runtime-state/metadata.json | 31 +++++++--
.../lmcache/kv-tier-management/metadata.json | 14 +++-
.../repositories/local-dream/metadata.json | 13 +++-
.../repositories/mistral-rs/metadata.json | 13 +++-
.../repositories/mlc-llm/metadata.json | 13 +++-
.../generation-cache-lifecycle/metadata.json | 14 +++-
.../repositories/mlx-swift-lm/metadata.json | 13 +++-
.../repositories/project-maya/metadata.json | 14 +++-
.../metadata.json | 14 +++-
.../repositories/splash/metadata.json | 14 +++-
.../audit-sd-cpp-residency/metadata.json | 64 +++++++++++++++++--
.../repositories/vllm-metal/metadata.json | 13 +++-
.../metadata.json | 14 +++-
.../audit-vllm-kv-lifetimes/metadata.json | 15 ++++-
.../metadata.json | 15 ++++-
29 files changed, 433 insertions(+), 67 deletions(-)
diff --git a/src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/metadata.json b/src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/metadata.json
index 81602cf6d..4d74a60c1 100644
--- a/src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/metadata.json
+++ b/src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/metadata.json
@@ -32,7 +32,12 @@
}
],
"applicability": {
- "scope": "FluxPipeline and FlowMatchEulerDiscreteScheduler invocation ownership at this source pin; inspect the actual caller contract."
+ "scope": "FluxPipeline and FlowMatchEulerDiscreteScheduler invocation ownership at this source pin; inspect the actual caller contract.",
+ "prerequisites": "Access to the pinned ownership path and an identified caller admission/serialization contract; other pipeline or scheduler classes need their own contracts.",
+ "input_contract": "Diffusers/PyTorch revisions, pipeline/scheduler classes and object identities, construction/from_pipe path, actual components, scheduler config and resolved timesteps/sigmas, shape/dtype, callback/interrupt timing, output/cleanup path and overlapping intervals.",
+ "produced_artifact": "An object-identity/ownership map and call-to-schedule-reset/step/interrupt/output/cleanup trace, the caller admission contract, one next check and actual evidence status.",
+ "failure_modes": "Assuming from_pipe deep-copies scheduler state, passing an enumeration index instead of a scheduler timestep, or treating cooperative iteration skipping as kernel cancellation/exception cleanup can misidentify invocation ownership.",
+ "refusal_point": "Stop when the ownership question is resolved or a required fixture/backend is absent; do not infer service-wide race or serialization behavior from this readset."
},
"limitations": [
"Source-derived candidate; no installed API, native scheduler, model, concurrency or service validation.",
@@ -42,5 +47,10 @@
"provenance_readsets": [
"references/source-map.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Check wrapper and scheduler identities explicitly and follow the effective schedule reset and step-index progression.",
+ "Trace interruption through its loop-boundary consumer and normal decode/cleanup tail while preserving the caller's existing shared-state serialization.",
+ "Use a real scheduler CPU check only for its scheduler cycle; keep pipeline reuse/interruption and serving claims tied to matching fixtures and execution evidence."
+ ]
}
diff --git a/src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/metadata.json b/src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/metadata.json
index a3930b808..3b4b26ef7 100644
--- a/src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/metadata.json
+++ b/src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/metadata.json
@@ -42,7 +42,12 @@
}
],
"applicability": {
- "scope": "MLX evaluation/stream/buffer diagnosis at the pinned implementation; qualify the actual backend separately."
+ "scope": "MLX evaluation/stream/buffer diagnosis at the pinned implementation; qualify the actual backend separately.",
+ "prerequisites": "Access to the pinned MLX execution/lifetime contract and a bounded reproduction with an identified evaluation/read boundary; device-dependent conclusions need the actual selected backend.",
+ "input_contract": "MLX revision/build, backend/device, Python/C++ entry, array shapes/dtypes, primitive streams, evaluation/read boundary and transformation context; timing needs warmup/work boundaries, while memory needs live references, in-flight buffers, allocator cache and synchronized observations.",
+ "produced_artifact": "A producer/evaluation/dependency-wait/backend/availability-cleanup trace, ownership table, observed symptom and one discriminating check with build-bound actual evidence status.",
+ "failure_modes": "Timing graph construction or submission as execution, assuming evaluated means completed, synchronizing only the default stream, or treating graph detachment/Python deletion as reusable storage can misdiagnose time or memory.",
+ "refusal_point": "Stop when the trace answers the question; if the required backend is absent, retain the source-supported diagnosis and exact runtime gap rather than substituting CPU import success."
},
"limitations": [
"Source-derived candidate; no local native-backend, model, service or efficacy validation.",
@@ -51,5 +56,10 @@
"provenance_readsets": [
"references/source-map.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Separate graph construction, async_eval submission, eval and host reads, and record which stream the measurement actually synchronizes.",
+ "Preserve the dependency-wait path and selected backend buffer-retention/donation contract before changing streams or assuming storage can be reused.",
+ "Distinguish live, in-flight and allocator-cached memory; inspect the real error/cleanup path before proposing a leak fix or hiding the failure."
+ ]
}
diff --git a/src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/metadata.json b/src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/metadata.json
index 076ae8adc..ca690ae38 100644
--- a/src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/metadata.json
+++ b/src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/metadata.json
@@ -22,5 +22,18 @@
"provenance_readsets": [
"references/sources.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "applicability": {
+ "scope": "Source-supported diagnosis of training-visible PyTorch DCP save pauses, snapshot/staging/upload lifetimes and bounded CPU staging-buffer reuse; distributed recovery validity and backend durability need separate evidence.",
+ "prerequisites": "Known correspondence between the installed PyTorch API and source/version notes, identified checkpointed-state mutation boundaries and an initial one-outstanding-save policy before overlap is considered.",
+ "input_contract": "Installed PyTorch revision, stager, storage writer and thread/process executor, checkpoint contents, ranks per host and state-mutation boundaries; performance assessment needs caller/staging/upload waits and host/resource observations.",
+ "produced_artifact": "A completion/lifetime diagnosis and smallest justified staging/cache/backpressure lever, with separate caller-pause, staging, upload and step-time measurements proposed for authorized validation.",
+ "failure_modes": "Mutating source state before staging completes, overwriting cached snapshots before prior upload completes, ignoring caller-side state_dict work or treating async return as durability can corrupt the snapshot or misstate the pause.",
+ "refusal_point": "Reject staging overlap that races checkpointed-state mutation; do not infer backend durability from asynchronous return or writer completion."
+ },
+ "recommendations": [
+ "Place the staging wait before the earliest mutation of any checkpointed state, including forward-mutated buffers rather than only optimizer.step.",
+ "Wait for the previous upload before reusing a stager's CPU storage; observe the final upload before closing caller-owned staging resources.",
+ "Measure checkpoint and intervening steps with the same state/cadence, retaining first-use and cached-buffer costs plus per-host memory and I/O pressure."
+ ]
}
diff --git a/src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/metadata.json b/src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/metadata.json
index bd0e27521..7a0490096 100644
--- a/src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/metadata.json
+++ b/src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/metadata.json
@@ -41,7 +41,12 @@
}
],
"applicability": {
- "scope": "Source-based CUDA P2pNccl rank/payload audit; selected eager dense-FFN path with compute_gate_on_attention=false. Repository identity afd_plugin; runtime/backend compatibility unverified."
+ "scope": "Source-based CUDA P2pNccl rank/payload audit; selected eager dense-FFN path with compute_gate_on_attention=false. Repository identity afd_plugin; runtime/backend compatibility unverified.",
+ "prerequisites": "Access to the pinned source ledger and caller-level evidence for the selected eager dense FFN path with Attention-side gating disabled; any proposed cross-stream reuse needs an established communicator completion contract.",
+ "input_contract": "Plugin and dependency revisions, connector and role, A/F worker counts, DP/PCP/TP coordinates, worker/device ranks, subgroup roster, stage, per-peer token counts, payload shape/dtype/device, transfer context and returned slices.",
+ "produced_artifact": "A bounded rank-namespace and per-peer payload ownership ledger, the earliest evidenced mismatch or missing edge, and a proposed verification plan.",
+ "failure_modes": "Inferring data peers from control senders, accepting equal total row counts despite a peer permutation, or treating an even-split fallback or Python return as proof of ownership/completion can produce a wrong exchange diagnosis.",
+ "refusal_point": "Stop on A < F, invalid role ranks, duplicate/missing assignments or an unresolved rank namespace; return the conflicting callers and ledger entries without inventing ranks or attributing the fault to networking."
},
"limitations": [
"Source-derived candidate; native exchange, model quality, backend compatibility, performance and agent efficacy not_run.",
@@ -56,5 +61,10 @@
"../../../methods/inference-benchmark-design/SKILL.md",
"../../../methods/dense-performance-evidence/SKILL.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Record role, world and subgroup-local ranks separately; reconstruct the actual subgroup roster before interpreting send/receive peers.",
+ "Track distinct row identities and stage-specific peer lengths through concatenation and splitting; use same-total swapped lengths as a countercheck.",
+ "Establish the last communicator/stream completion obligation before reusing or releasing a receive buffer."
+ ]
}
diff --git a/src/inferrsi/data/skill_library/repositories/comfy-ui/metadata.json b/src/inferrsi/data/skill_library/repositories/comfy-ui/metadata.json
index 9399cc7d9..b37aeb6dc 100644
--- a/src/inferrsi/data/skill_library/repositories/comfy-ui/metadata.json
+++ b/src/inferrsi/data/skill_library/repositories/comfy-ui/metadata.json
@@ -2,11 +2,19 @@
"profile": "local-skill-v1",
"schema_version": 1,
"kind": "skill",
- "origin": ["external"],
+ "origin": [
+ "external"
+ ],
"status": "candidate",
"empirical_validation": "not_run",
"license": "GPL-3.0",
- "tags": ["comfyui", "workflow", "cache-invalidation", "model-residency", "state-ownership"],
+ "tags": [
+ "comfyui",
+ "workflow",
+ "cache-invalidation",
+ "model-residency",
+ "state-ownership"
+ ],
"sources": [
{
"type": "git",
@@ -40,13 +48,25 @@
}
],
"applicability": {
- "scope": "Source-grounded ComfyUI workflow-output cache and model-weight ownership audit; confirm the actual revision, node registry, cache caller and static/dynamic patcher before transfer."
+ "scope": "Source-grounded ComfyUI workflow-output cache and model-weight ownership audit; confirm the actual revision, node registry, cache caller and static/dynamic patcher before transfer.",
+ "prerequisites": "Access to the pinned provenance and the actual node registry/cache caller; residency questions additionally require the concrete patcher and effective dynamic/static mode. Native memory conclusions need the actual backend.",
+ "input_contract": "Exact revision, API-format prompt/output targets, relevant node classes/constants/links/hidden inputs and external state, selected cache caller/mode and patcher settings, plus available execution/load/memory/timing observations.",
+ "produced_artifact": "A state/owner/key/lifetime/consumer ledger with invalidation or release events, evidence and unresolved boundaries, followed by a discriminating check and its actual result.",
+ "failure_modes": "Treating identical JSON as proof of unchanged external inputs, confusing eviction with key validity, or equating workflow cache hits with device-resident model weights can explain the wrong state contract.",
+ "refusal_point": "Stop when the cause/action is supported or required environment, backend or model resources are unavailable; do not manufacture a hit, hardware result or performance threshold."
},
"limitations": [
"Source-derived candidate; no local executor-cache, device-residency or performance qualification.",
"Seven self-authored cases exercised only the unchanged pure upstream graph_utils module; they do not verify PromptExecutor, cache signatures or model behavior.",
"The dynamic native allocator and third-party custom-node behavior remain outside the verified scope."
],
- "provenance_readsets": ["references/repo-provenance.md"],
- "consumption_mode": "packaged"
+ "provenance_readsets": [
+ "references/repo-provenance.md"
+ ],
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Draw the affected dependency cone with declared, lazy and dynamically expanded links, then mark unfingerprinted external state unresolved.",
+ "Separate output/object cache validity and retention from patch identity, partial unload, full detach and allocator-visible release.",
+ "Change one test variable while fixing prompt, output target, model, backend, precision, seed and cache mode; check values and executed nodes before timing."
+ ]
}
diff --git a/src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/metadata.json b/src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/metadata.json
index 9227b28c2..806dac0f3 100644
--- a/src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/metadata.json
+++ b/src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/metadata.json
@@ -39,7 +39,11 @@
"applicability": {
"scope": "Decoders with an explicit producer/consumer boundary and a demonstrated dependency closure for compact query rows, including separately preserved historical state",
"prerequisites": "A derived dependency closure; a full-length KV write at the boundary; known rollback depth; control over metadata construction",
- "hardware": "Backend-independent retain arithmetic; window sizes, boundary index and ring lengths come from the model config and do not transfer"
+ "hardware": "Backend-independent retain arithmetic; window sizes, boundary index and ring lengths come from the model config and do not transfer",
+ "input_contract": "The actual producer/consumer split and transitive dependency closure, boundary index, window and rollback depth, consumer KV sharing/compression layout and separately preserved historical keys.",
+ "produced_artifact": "A dependency listing, enforced retain length, compact query/position/store-slot/visibility metadata, a separate historical-key retention statement and a diagnostic equivalence-reference mode.",
+ "failure_modes": "Applying window-plus-rollback without dependency closure, discarding full-layout state, clipping away the sampling row, or comparing across preemption can invalidate replay; compact queries do not imply historical keys were dropped.",
+ "refusal_point": "Stop if required consumer state is unavailable without a preserved full-length copy, the boundary representation cannot be captured, or comparison preemption moves the anchor. Do not use for clipped-region outputs or required bit-identical preemption replay."
},
"limitations": [
"Source-derived candidate; no equivalence run was performed.",
@@ -53,5 +57,10 @@
"references/source-notes.md",
"references/validation.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Derive every consumer dependency before computing keep; preserve the required full-layout representation separately from the compact query rows.",
+ "Use the visibility-only diagnostic mode to isolate equivalence from row gathering, and retain one sampling row for a fully clipped chunk.",
+ "Check metadata at chunk/store boundaries, actual prompt-logprob handling and zero preemption during equivalence comparison."
+ ]
}
diff --git a/src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/metadata.json b/src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/metadata.json
index 0d76d84cb..50d9822fa 100644
--- a/src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/metadata.json
+++ b/src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/metadata.json
@@ -37,7 +37,11 @@
"applicability": {
"scope": "Engines with multi-token speculative decoding whose verification forward cost grows with query rows",
"prerequisites": "A calibrated per-position acceptance signal; a measured (rows, step_time) curve; a handle on metadata construction",
- "hardware": "Backend-independent procedure; the reference implementation targets NVIDIA SM100 and Ascend 950 only"
+ "hardware": "Backend-independent procedure; the reference implementation targets NVIDIA SM100 and Ascend 950 only",
+ "input_contract": "Draft width and row representation, the defined individual/conditional versus cumulative acceptance semantics, verify rows/time curve, sampler cost and acceptance round actually available at schedule time.",
+ "produced_artifact": "A per-step verification row budget, per-request query lengths from one through draft_width plus one, cumulative acceptance vectors and predicted-versus-realized acceptance error logs.",
+ "failure_modes": "Confusing verification savings with reduced drafting, misreading confidence semantics or available rounds, allowing later-position confidence to select an earlier token, or generating non-prefix grants can invalidate the budget or sampling claim.",
+ "refusal_point": "Stop if unthrottled drafting or sampler cost dominates, the signal cannot be made cumulative, or no round supplies acceptance estimates; the procedure also does not apply without a per-step cost model."
},
"limitations": [
"Source-derived candidate; no local efficacy validation and no reproduced gain.",
@@ -52,5 +56,10 @@
"references/source-notes.md",
"references/validation.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Measure draft, verify and sampler costs separately and fit the verify curve for graph sizes that will actually replay.",
+ "Preserve the one-row floor and prefix grants, and audit CPU/GPU allocation timing against the per-position decision-independence contract.",
+ "Report verification rows, accepted tokens, total step time and throughput separately; use an independently specified sampling oracle rather than requiring unchanged accepted length."
+ ]
}
diff --git a/src/inferrsi/data/skill_library/repositories/dwarfstar/metadata.json b/src/inferrsi/data/skill_library/repositories/dwarfstar/metadata.json
index 1bebef173..a8f11e400 100644
--- a/src/inferrsi/data/skill_library/repositories/dwarfstar/metadata.json
+++ b/src/inferrsi/data/skill_library/repositories/dwarfstar/metadata.json
@@ -73,7 +73,12 @@
"method": "User-approved native C adaptation and partial drafting; manual stable-ID subskill orchestration and integrated create/verify stages. Required-backend blocks disable automatic admission."
},
"applicability": {
- "scope": "Pinned DwarfStar source investigation and experiment planning; not arbitrary GGUF support or cross-backend qualification."
+ "scope": "Pinned DwarfStar source investigation and experiment planning; not arbitrary GGUF support or cross-backend qualification.",
+ "prerequisites": "Access to the pinned native-source contract and relevant memory/state/qualification route; native verification needs a compatible build, matching hardware/weights and task-specific execution authority.",
+ "input_contract": "Engine revision/patches, actual model artifact/representation, selected backend/build and topology, memory/storage budget, context/chunk/session count, image inputs, sampling/speculation policy, cache state and quality/service requirements.",
+ "produced_artifact": "A bounded executing-tuple report with observed signal, source-backed mechanism, alternate explanation, proposed falsifier, actual check outcome and unresolved backend/model/service gap.",
+ "failure_modes": "Inferring model support from flags, aliases, file mapping or quantization names can select the wrong graph; treating CPU bookkeeping or parser checks as native backend/service qualification can overstate the result.",
+ "refusal_point": "Stop at unavailable hardware/weights, unsafe exposure or the authorized budget; retain memory-guard refusals rather than disabling the guard to claim fit."
},
"limitations": [
"Metal/CUDA/ROCm inference, actual model output, SSD runtime I/O, GPU recurrent rollback, physical distributed behavior, service quality/performance and skill efficacy remain unverified.",
@@ -85,5 +90,10 @@
"references/repo-provenance.md",
"references/source-identity.json"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Choose the memory, serving-state or qualification route by the observed symptom and identify the owner, validity key and release/rollback boundary.",
+ "Freeze comparison controls, test the smallest invariant first and retain fallback messages, cancellations, refusals and unavailable fixtures.",
+ "Base model support on the actual loaded artifact and graph; use native backend evidence for model, tensor-parallel or service conclusions."
+ ]
}
diff --git a/src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/metadata.json b/src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/metadata.json
index b8aaa2891..f433264e6 100644
--- a/src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/metadata.json
+++ b/src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/metadata.json
@@ -63,7 +63,12 @@
"method": "AREX scope/ground/construct/verify and repository evidence/usability methods adapted to InferRSI native owners and packaging; no DisCo runtime or managed import."
},
"applicability": {
- "scope": "Pinned Flare complete-session restore and admission audit for its NVIDIA SM120/SM121 research prototype. CUDA catalog scope is content classification, not device or backend qualification."
+ "scope": "Pinned Flare complete-session restore and admission audit for its NVIDIA SM120/SM121 research prototype. CUDA catalog scope is content classification, not device or backend qualification.",
+ "prerequisites": "Access to package/source identity and bounded restore traces, with preparation compatibility established before cache-state diagnosis. Running, freeing resources or changing configuration requires separate authority.",
+ "input_contract": "Execution ABI and prepared-contract version, ordinary/speculative configuration, model/layout identity, request IDs and consumed-token endpoints, GPU/host prefixes, continuation row, ticket/waiter identities, transfer events, fences and resource counts.",
+ "produced_artifact": "A checkpoint/ticket ownership table, five-outcome admission ledger and diagnosis identifying the earliest unsupported transition and smallest missing observation.",
+ "failure_modes": "A KV-only lookup hit can omit continuation state; admission can be mistaken for actual reuse; client cancellation can be mistaken for transfer completion; mixing absolute timestamps from client and native clocks can corrupt the diagnosis.",
+ "refusal_point": "Stop with an unresolved edge when package identity, layout, transfer completion or clock provenance is unknown."
},
"limitations": [
"Source-only candidate; native Flare/model/GPU/service and skill efficacy were not evaluated.",
@@ -76,5 +81,10 @@
"references/source-notes.md",
"references/source-ledger.json"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Pair the immutable KV parent chain with the complete continuation row at the same consumed-token endpoint before calling a checkpoint restorable.",
+ "Distinguish lookup outcomes, destination reservation, ticket creation, transfer completion, publication and actual reused-token evidence.",
+ "Follow scheduler-owned tickets through waiter cancellation and check write/use fences before allowing storage reuse."
+ ]
}
diff --git a/src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/metadata.json b/src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/metadata.json
index 3128ce37e..b026cebf4 100644
--- a/src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/metadata.json
+++ b/src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/metadata.json
@@ -105,7 +105,12 @@
"method": "Manual pinned-source distillation informed by the existing Repo-To-Skill ledger; no AREX/DisCo execution or behavioral efficacy claim."
},
"applicability": {
- "scope": "MiniMax-H3 joint audio/video C/Metal runtime audit: raw checkpoint compatibility, input encoders, schedule precompute, phased/cached residency, numerical controls and SSD streaming on Apple Silicon. Common catalog placement is a discovery classification, not cross-backend qualification."
+ "scope": "MiniMax-H3 joint audio/video C/Metal runtime audit: raw checkpoint compatibility, input encoders, schedule precompute, phased/cached residency, numerical controls and SSD streaming on Apple Silicon. Common catalog placement is a discovery classification, not cross-backend qualification.",
+ "prerequisites": "Access to the pinned source architecture and selected component/format inventory; native checks need matching model, fixture and backend resources within an authorized experiment budget.",
+ "input_contract": "Runtime revision/patches, macOS/SDK/GPU and memory budget, FL2VA/Ref2VA selection, component formats, prompt and ordered media, requested/aligned frames, canvases, steps/seed, layer/reuse/token/precision policies and preview/interactive-cache state.",
+ "produced_artifact": "A phase/component/owner ledger with resident weights, runtime/host buffers, overlap, release/reuse boundaries and observations/unknowns, followed by a discriminating check, actual result and remaining gap.",
+ "failure_modes": "Estimating fit from model-file bytes, confusing runtime int8 with incompatible MLX bundles, ignoring preview/cache overlap, or reducing DiT residency to address a VAE/host-output peak can produce a wrong capacity diagnosis.",
+ "refusal_point": "Stop at unavailable model/fixture/backend resources or the accepted experiment budget; missing target evidence remains unverified/INCONCLUSIVE."
},
"limitations": [
"Native h3.c/Metal execution, target peak memory, audio/video quality, performance, service behavior and skill efficacy were not run.",
@@ -119,5 +124,10 @@
"references/source-identity.json",
"references/minimax-h3-weight-inventory.json"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Resolve the actual conditioning and output components before counting residency; do not add alternative checkpoints or count shared files twice.",
+ "Trace one-shot, preview and interactive ownership separately, including AdaLN schedule dependencies and prepared-cache reset boundaries.",
+ "Classify approximation and precision controls before comparison, then measure the relevant phase with tensor, physical/compressed/swap and storage observations."
+ ]
}
diff --git a/src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/metadata.json b/src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/metadata.json
index 647cdc29d..2a83f4b35 100644
--- a/src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/metadata.json
+++ b/src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/metadata.json
@@ -77,7 +77,12 @@
"method": "Bounded fork-specific source audit, bundled selected-declaration helper and assertion-backed cases. Native Apple/Metal/model evidence remains blocked; no numerical/performance/skill-efficacy claim."
},
"applicability": {
- "scope": "h3-metal FastH3 requested/effective configuration, checkpoint transformation and schedule/RoPE semantics, VSA prefix/partial tiles and dispatch/fallback audit. Common catalog placement is discovery, not CPU inference or Apple qualification."
+ "scope": "h3-metal FastH3 requested/effective configuration, checkpoint transformation and schedule/RoPE semantics, VSA prefix/partial tiles and dispatch/fallback audit. Common catalog placement is discovery, not CPU inference or Apple qualification.",
+ "prerequisites": "Access to the fork-specific source and semantic contracts; native checks additionally need their matching Metal/model resources and separate execution authority.",
+ "input_contract": "Source revision/patches, requested parameters and explicitly captured environment controls, checkpoint path/key-family/config identity, requested/aligned frames, latent dimensions, packed prefix segments and sparse-pipeline availability or unknown.",
+ "produced_artifact": "A ledger of requested and normalized parameters, ambient overrides, source invariants, counterexamples, observed/unknown facts and the smallest discriminating next check.",
+ "failure_modes": "Ignoring ambient overrides, mixing checkpoint identity switches, using global rather than per-head Q/K/V concatenation, pooling padded boundary rows, or double-adding coarse attention can invalidate the semantic audit.",
+ "refusal_point": "Stop on conflicting configuration, mixed checkpoint identity, missing resources or the authorized boundary; name the missing native test/fixture and leave it blocked/not_run."
},
"limitations": [
"BLOCKED_REQUIRED_BACKEND: native Apple/Metal, actual FastH3 checkpoint/schedule/sparse parity, audio/video quality, target memory fit, performance and service evidence not_run.",
@@ -92,5 +97,10 @@
"references/source-identity.json",
"references/repo-provenance.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Record requested parameters, normalized values and later environment overrides separately, including requested versus aligned frame counts.",
+ "Audit checkpoint transformations and schedule consumers beyond tensor names/shapes; tiny permutation fixtures can falsify mappings but do not establish model parity.",
+ "Resolve sparse eligibility and the dense-fine-plus-gated-coarse fallback explicitly; preserve unknown pipeline availability."
+ ]
}
diff --git a/src/inferrsi/data/skill_library/repositories/husky/metadata.json b/src/inferrsi/data/skill_library/repositories/husky/metadata.json
index 62f34816d..378a5062b 100644
--- a/src/inferrsi/data/skill_library/repositories/husky/metadata.json
+++ b/src/inferrsi/data/skill_library/repositories/husky/metadata.json
@@ -1,11 +1,18 @@
{
"schema_version": 1,
"kind": "skill",
- "origin": ["external"],
+ "origin": [
+ "external"
+ ],
"status": "candidate",
"empirical_validation": "not_run",
"license": "NO_LICENSE",
- "tags": ["husky", "document-research", "evidence-qualification", "negative-results"],
+ "tags": [
+ "husky",
+ "document-research",
+ "evidence-qualification",
+ "negative-results"
+ ],
"sources": [
{
"type": "document",
@@ -37,7 +44,12 @@
"method": "Single coherent research workflow, bounded source summary, original verification questions and offline schema checks. Repository/environment/native stages remain unavailable."
},
"applicability": {
- "scope": "Review Husky document claims and specify missing evidence; common catalog scope is not inference backend compatibility."
+ "scope": "Review Husky document claims and specify missing evidence; common catalog scope is not inference backend compatibility.",
+ "prerequisites": "Access to the bounded official-document inventory and verification protocol; implementation and native reproduction remain unavailable until separately supplied and authenticated.",
+ "input_contract": "The exact claim, source/view date, workload and comparison intent, supporting evidence classes and available implementation/model/output/measurement artifacts; unknown identity, licensing and availability remain explicit.",
+ "produced_artifact": "A reviewable claim/evidence request recording source availability, mechanism hypothesis, counter-signal, missing identities/artifacts, proposed falsifying check and actual result or bounded INCONCLUSIVE conclusion.",
+ "failure_modes": "Treating a public-page reading as code inspection, proposed extensions as shipped behavior, summed independent ratios as one gain, or a structurally valid self-reported card as authentic evidence can overstate qualification.",
+ "refusal_point": "Stop at missing source/receipts, unsafe or private data, unsupported resources or the authorized budget; unavailable artifacts yield an unverified/INCONCLUSIVE research conclusion."
},
"limitations": [
"No canonical public repository, implementation revision, retained receipts or source-code license was identified; unknowns remain unknown.",
@@ -47,6 +59,14 @@
"NO_LICENSE is the unavailable repository-license resolver result, not a legal absence finding or grant to copy publisher material.",
"Final informed candidate admission is separate; no automatic import, native qualification or skill efficacy is claimed."
],
- "provenance_readsets": ["references/repo-provenance.md", "references/source-availability.json"],
- "consumption_mode": "packaged"
+ "provenance_readsets": [
+ "references/repo-provenance.md",
+ "references/source-availability.json"
+ ],
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Separate author reports, inspected implementation, executed experiments and new interpretations, including negative findings and future work.",
+ "Request immutable identities and artifacts, then state one mechanism, competing explanation and smallest ablation with frozen semantic requirements.",
+ "Use the offline card auditor only to identify structural gaps; it authenticates neither artifacts nor execution and leaves qualification unadjudicated."
+ ]
}
diff --git a/src/inferrsi/data/skill_library/repositories/lily/metadata.json b/src/inferrsi/data/skill_library/repositories/lily/metadata.json
index c4088431e..e986edc1e 100644
--- a/src/inferrsi/data/skill_library/repositories/lily/metadata.json
+++ b/src/inferrsi/data/skill_library/repositories/lily/metadata.json
@@ -90,7 +90,12 @@
"method": "User-approved native Rust/Metal and partial drafting adaptation; stable-ID lanes, integrated source graph and assertion/helper verification. Missing compiler/backend/checkpoint remains explicit."
},
"applicability": {
- "scope": "Exact Lily source/checkpoint/layout/API/cache/qualification investigation; common catalog classification is not CPU inference or cross-device support."
+ "scope": "Exact Lily source/checkpoint/layout/API/cache/qualification investigation; common catalog classification is not CPU inference or cross-device support.",
+ "prerequisites": "Access to the exact Lily source/model-layout/API contracts; native checks require the matching Rust/Metal/checkpoint environment and frozen comparison requirements.",
+ "input_contract": "Source revision, Cargo lock/toolchain, macOS/chip/GPU family, checkpoint and tensor payload identities, affine configuration/layout, tokenizer/template/stops, sequence/output budgets, greedy policy, input history, prefix-cache state and quality/measurement requirements.",
+ "produced_artifact": "A source/model/device-bound representation, serving-cache or qualification diagnosis with the observation, invariant, alternate cause, smallest falsifier, actual result and remaining native gap.",
+ "failure_modes": "Equal bit width, file size, API names or prompt_cache_key do not establish tensor/token/state equivalence; ignoring the extra lookahead slot can understate request capacity, and green CPU references do not replace matched-model Metal goldens.",
+ "refusal_point": "Stop at unsupported formats/APIs, missing compiler/backend/checkpoint, unfrozen gates or the authorized budget; do not weaken ignored/golden tests to obtain a green result."
},
"limitations": [
"Cargo/Rust1.92, supported M5/macOS26/Metal and model resources are unavailable; native evidence remains BLOCKED_REQUIRED_BACKEND.",
@@ -103,5 +108,10 @@
"references/repo-provenance.md",
"references/source-identity.json"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Resolve exact affine representation and strict API semantics, then trace token frontier, cache key and restore/release ownership.",
+ "Retain the source's extra request lookahead slot in capacity arithmetic rather than counting only prompt and visible output.",
+ "Separate reported benchmark version from the checked-out runner and distinguish offline screening, native compilation, Metal operators and matched-model goldens."
+ ]
}
diff --git a/src/inferrsi/data/skill_library/repositories/lithos-metal/metadata.json b/src/inferrsi/data/skill_library/repositories/lithos-metal/metadata.json
index 6136acb38..5c1dc9cf4 100644
--- a/src/inferrsi/data/skill_library/repositories/lithos-metal/metadata.json
+++ b/src/inferrsi/data/skill_library/repositories/lithos-metal/metadata.json
@@ -45,7 +45,11 @@
}
],
"applicability": {
- "scope": "Pinned Lithos Metal 0.1.2 source/API contracts; Apple silicon Metal runtime; model and recipe qualification required separately."
+ "scope": "Pinned Lithos Metal 0.1.2 source/API contracts; Apple silicon Metal runtime; model and recipe qualification required separately.",
+ "prerequisites": "Access to the pinned package/API and selected compiler, generation or serving contracts; revision/API drift needs a fresh bounded inspection, and efficacy claims need campaign-owned gates and matching native/model evidence.",
+ "input_contract": "Actual source/package version, backend/configuration, chip/core count and OS, target/draft checkpoint identities, context and prompt-token IDs, sampling/numerical options and the affected recipe/runtime/session/cache boundary.",
+ "produced_artifact": "A routed region/dependency, committed-state or cache-identity/lifecycle ledger, followed by the smallest supported finding, missing checks and next bounded action.",
+ "failure_modes": "Treating session adoption/cache hits as generation correctness, fewer dispatches as faster rounds, another chip recipe as the default, CPU cache checks as GPU rollback or health liveness as loaded/warm model state can yield unsupported conclusions."
},
"limitations": [
"Source-derived candidate; empirical efficacy validation not_run.",
@@ -60,5 +64,10 @@
"disable-model-invocation": true,
"metadata": {
"disco-role": "operating"
- }
+ },
+ "recommendations": [
+ "Follow session replacement across recipe geometry, committed generation state and serving adoption/cache identity before interpreting a cache hit.",
+ "Keep an unfused/native or target-only control with matching decoded weights, numerical mode, row bounds and sampling policy; verify dependencies and persistent state before timing.",
+ "Distinguish package/CPU checks, synthetic Metal execution, model continuation and HTTP workload evidence; chip recipes and target-compute projections do not establish another device or E2E rate."
+ ]
}
diff --git a/src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/metadata.json b/src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/metadata.json
index f102d8ecd..bd071221d 100644
--- a/src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/metadata.json
+++ b/src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/metadata.json
@@ -1,11 +1,20 @@
{
- "tags": ["llama.cpp", "context", "state", "graph-reuse", "cancellation", "correctness"],
+ "tags": [
+ "llama.cpp",
+ "context",
+ "state",
+ "graph-reuse",
+ "cancellation",
+ "correctness"
+ ],
"status": "candidate",
"empirical_validation": "not_run",
"profile": "local-skill-v1",
"schema_version": 1,
"kind": "skill",
- "origin": ["external"],
+ "origin": [
+ "external"
+ ],
"sources": [
{
"type": "git",
@@ -36,13 +45,25 @@
}
],
"applicability": {
- "scope": "Pinned llama.cpp context-memory, graph-reuse and server task-lifecycle diagnosis; concrete model/backend behavior requires native validation."
+ "scope": "Pinned llama.cpp context-memory, graph-reuse and server task-lifecycle diagnosis; concrete model/backend behavior requires native validation.",
+ "prerequisites": "Pinned llama.cpp source or a concrete runtime trace, access to the relevant memory/graph/server ownership evidence and an identified caller recovery policy that remains in force.",
+ "input_contract": "Source/build and model revisions, memory type, actual backend/device and execution mode, context/batch/ubatch parameters, sequence IDs/positions/output flags and return status; graph cases need dispatch evidence, server cases need correlated HTTP/task/slot/cache logs.",
+ "produced_artifact": "An owner/event ledger of ubatch progress, retained memory, graph-gate decisions, input readiness and slot release/retention, with the first supported broken obligation or smallest missing observation.",
+ "failure_modes": "Assuming whole-batch rollback after partial decode failure, treating token count as a complete graph-reuse key, or equating posted cancellation with released slots can hide retained state or misdiagnose correct deferred work.",
+ "refusal_point": "Stop before reset or repair if revision, memory type, actual backend, sequence ownership or partial progress is unknown; escalate unsupported rollback or unresolved in-flight work."
},
"limitations": [
"Source-derived candidate; native runtime and efficacy validation not run.",
"No cross-backend preemption, model-quality or performance claim.",
"KV, recurrent and hybrid state contracts must be checked independently."
],
- "provenance_readsets": ["references/source-notes.md"],
- "consumption_mode": "packaged"
+ "provenance_readsets": [
+ "references/source-notes.md"
+ ],
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Reconcile per-sequence retained positions with completed ubatches before retrying a failed logical batch.",
+ "Check both graph allow_reuse and input can_reuse gates, then establish copy readiness independently of graph identity.",
+ "Trace cancellation posting, normal task processing, slot release and prompt retention as separate events."
+ ]
}
diff --git a/src/inferrsi/data/skill_library/repositories/lmcache/kv-tier-management/metadata.json b/src/inferrsi/data/skill_library/repositories/lmcache/kv-tier-management/metadata.json
index 590c0f707..3f383464b 100644
--- a/src/inferrsi/data/skill_library/repositories/lmcache/kv-tier-management/metadata.json
+++ b/src/inferrsi/data/skill_library/repositories/lmcache/kv-tier-management/metadata.json
@@ -21,5 +21,17 @@
"Original typed source inventory is not normalized; retain source identities and limits in the procedure/readsets."
],
"provenance_readsets": [],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "applicability": {
+ "scope": "LMCache multiprocess runtime KV tier policy across engine GPU cache, host L1 and configured L2 adapters, covering store/retention, prefetch admission, eviction and RAM pressure; startup loading and checkpoint recovery have separate owners.",
+ "prerequisites": "The deployment uses the inspected MP-mode semantics rather than single-process knobs; selected GDS/DAX/raw-block paths need their own platform and adapter prerequisites. Lazy offload additionally requires vLLM prefix caching.",
+ "input_contract": "MP versus single-process mode, store/retention policy and rationale, L1 manager capacities and eviction settings, L2 adapter capacities/eviction, chunk-size distribution and prefetch in-flight setting, together with token-level hit and pressure counters.",
+ "produced_artifact": "A mode- and adapter-specific tier-policy choice with retention rationale, an ordered RAM-release plan and a verification record separating L1/L2 token hits, prefetch pressure, store lag and per-manager usage.",
+ "failure_modes": "Mixing MP and in-process controls, treating key-count eviction as a byte fraction for unequal chunks, omitting fs capacity management, or increasing count-based prefetch admission without tracking L1 pressure can make the policy or memory claim unusable."
+ },
+ "recommendations": [
+ "Choose store retention and prefetched-key retention separately; use skip_l1 with noop only for buffer-only operation and account for failed stores remaining in L1.",
+ "Tune the count-based prefetch in-flight limit against L1 allocation failures and in-flight bytes; eager or warm prefetch can raise pressure before useful reuse.",
+ "Record per-adapter capacity/eviction and actual chunk-size uniformity; report token-level L1/L2 hits and sum per-l1_tag bytes rather than ratios."
+ ]
}
diff --git a/src/inferrsi/data/skill_library/repositories/local-dream/metadata.json b/src/inferrsi/data/skill_library/repositories/local-dream/metadata.json
index 7cc15f18b..4f846892c 100644
--- a/src/inferrsi/data/skill_library/repositories/local-dream/metadata.json
+++ b/src/inferrsi/data/skill_library/repositories/local-dream/metadata.json
@@ -83,7 +83,11 @@
}
],
"applicability": {
- "scope": "Selected Android app/process and native diffusion boundary contracts; common is a discovery category, not CPU/device compatibility or native qualification."
+ "scope": "Selected Android app/process and native diffusion boundary contracts; common is a discovery category, not CPU/device compatibility or native qualification.",
+ "prerequisites": "Access to the pinned Android lifecycle/native/packaging contracts and identified installed artifacts; stronger conclusions require an authorized, frozen workload on the actual selected Android/ARM64 backend.",
+ "input_contract": "Source revision, installed APK/version/flavor/ABI, native executable and loaded DiT library, QNN/Hexagon/MNN builds, model/export/weights/tokenizer identities, requested and serving configuration, and lifecycle/readiness/generation/cache observations.",
+ "produced_artifact": "A bounded failed-boundary report with counter-signals, identities observed versus declared, the smallest missing discriminator and retained INCONCLUSIVE native scope.",
+ "failure_modes": "Reading UI flags or equal declared configurations as actual dispatch/readiness, treating cancellation/SSE progress as native completion, or interpreting model-directory caches as authenticated model-revision namespaces can misidentify stale state."
},
"limitations": [
"Source-derived candidate; required Android ARM64/MNN/QNN/Hexagon/model execution BLOCKED_REQUIRED_BACKEND / not_run; qualification INCONCLUSIVE.",
@@ -98,5 +102,10 @@
"references/repo-provenance.md",
"references/source-identity.json"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Compare requested and serving model/backend/dimensions/bind address through stop/reopen and health observations; never enable host mode as a diagnostic default.",
+ "Trace the actual MNN/QNN/Anima/DiT path and separate native library/ABI identity from UI runOnCpu flags and declared SDK labels.",
+ "Keep progress, cancellation, native completion, release and saved final output separate; check prompt-cache mode/version/dimensions and textual-inversion caller guards before proposing cache changes."
+ ]
}
diff --git a/src/inferrsi/data/skill_library/repositories/mistral-rs/metadata.json b/src/inferrsi/data/skill_library/repositories/mistral-rs/metadata.json
index 8da66c1d2..6eab1ae7f 100644
--- a/src/inferrsi/data/skill_library/repositories/mistral-rs/metadata.json
+++ b/src/inferrsi/data/skill_library/repositories/mistral-rs/metadata.json
@@ -75,7 +75,11 @@
"method": "Stable-ID bounded decode/prefill/qualification author lanes, whole-graph reconciliation, hard cases and offline helper/static/scaffold/wheel checks; compiler and actual Apple evidence not_run."
},
"applicability": {
- "scope": "Selected native Metal paged decode/prefill layout/metadata/workspace and preparation/evidence investigation. Common catalog categorization is not CPU inference or hardware certification."
+ "scope": "Selected native Metal paged decode/prefill layout/metadata/workspace and preparation/evidence investigation. Common catalog categorization is not CPU inference or hardware certification.",
+ "prerequisites": "Access to the pinned Rust/Candle/MSL decode, prefill and qualification contracts; native comparisons need the matched Rust/Apple/Candle/shader stack and separately authorized frozen references.",
+ "input_contract": "mistral.rs/Candle/native artifact revisions, Rust toolchain, Apple SDK/OS/device, actual model/cache dtype and representation, shape/strides/offsets, block table, context and execution limits.",
+ "produced_artifact": "A source/build-bound route and invariant report with actual versus declared facts, safe support result, counter-signal and smallest next native check; unavailable Metal/kernel/model evidence remains blocked.",
+ "failure_modes": "Equating element counts with physical alias/layout identity, transferring decode geometry to prefill, or treating prewarm/pipeline preparation as actual dispatch can misclassify eligibility and native correctness."
},
"limitations": [
"BLOCKED_REQUIRED_BACKEND: actual Rust/Candle/Apple Metal compilation, native tensor/cache/dispatch/parity/model/quality/service/performance evidence not_run.",
@@ -88,5 +92,10 @@
"references/repo-provenance.md",
"references/source-identity.json"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Select decode, prefill or qualification by the failed contract, then follow cache/metadata bytes from producer through view to native consumer.",
+ "Keep static predicates, pipeline preparation, kernel dispatch, numerical correctness and model/service outcomes as separate evidence steps; retain skipped or gated tests.",
+ "Use the matched Candle revision and native shader/build inputs; broad workspace tests can require model fixtures, tokens and downloads."
+ ]
}
diff --git a/src/inferrsi/data/skill_library/repositories/mlc-llm/metadata.json b/src/inferrsi/data/skill_library/repositories/mlc-llm/metadata.json
index 4c1955b50..bc539b634 100644
--- a/src/inferrsi/data/skill_library/repositories/mlc-llm/metadata.json
+++ b/src/inferrsi/data/skill_library/repositories/mlc-llm/metadata.json
@@ -79,7 +79,11 @@
}
],
"applicability": {
- "scope": "Selected artifact/Apple target/pinned TVM Metal consumer and KV configuration source contracts. Common catalog categorization is not CPU inference or Apple qualification."
+ "scope": "Selected artifact/Apple target/pinned TVM Metal consumer and KV configuration source contracts. Common catalog categorization is not CPU inference or Apple qualification.",
+ "prerequisites": "Access to the pinned MLC and exact TVM fork/gitlink contracts; native checks require separately prepared matching MLC/TVM/FFI and the applicable Apple SDK/device stack.",
+ "input_contract": "MLC revision, TVM gitlink/build, model/tokenizer/quantized-parameter identity, target/host/SDK, compiled artifact/format, runtime device and the actual compiled/runtime configuration sources.",
+ "produced_artifact": "An identity and actual-versus-declared boundary report for artifact, Metal runtime or KV-limit ownership, with a negative signal and smallest missing native discriminator.",
+ "failure_modes": "Treating a schema hash as payload authentication, an opt-in WebLLM artifact contract as a native-engine ABI, serializable Metal fallback as runnable, or automatic capacity estimates as free memory can create unsupported qualification claims."
},
"limitations": [
"Source-derived candidate; original compiled CPU/native tests, compiler/live API/CLI/model/Metal/fit/quality/service/performance execution not_run; required native BLOCKED_REQUIRED_BACKEND.",
@@ -93,5 +97,10 @@
"references/repo-provenance.md",
"references/source-identity.json"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Identify the interface's actual consumer before using a valid artifact manifest as evidence of native model/task support.",
+ "Separate codegen/export from runtime factory/load, pipeline creation, dispatch and stream completion; a serializable kind=metal fallback can still be non-runnable.",
+ "Trace runtime preference, compile-time bounds, nested workspace inputs and explicit overrides separately before interpreting a KV or working-set estimate."
+ ]
}
diff --git a/src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/metadata.json b/src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/metadata.json
index 1bbf2db98..e334dedd1 100644
--- a/src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/metadata.json
+++ b/src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/metadata.json
@@ -35,7 +35,12 @@
}
],
"applicability": {
- "scope": "MLX-LM generation, batch-row and prefix-cache diagnosis; model/backend/runtime compatibility requires separate qualification."
+ "scope": "MLX-LM generation, batch-row and prefix-cache diagnosis; model/backend/runtime compatibility requires separate qualification.",
+ "prerequisites": "Access to the independently pinned MLX-LM generation/cache contracts and the selected entry path; model, backend and runtime compatibility is separately established for any native claim.",
+ "input_contract": "MLX-LM/MLX revisions and builds, backend/device, model/tokenizer/adapter/draft identities, entry point, tokenized prompts, cache classes, batching/prefill limits, per-request sampling/processors/stops, seed and KV quantization; failure traces include UIDs, stages, lengths, terminal reason and next-request result.",
+ "produced_artifact": "An entry-path decision, UID/stage/cache-row table, prefix identity and cache-class assumptions, terminal/cleanup trace and actual evidence status with a targeted verification plan.",
+ "failure_modes": "Ignoring server batching gates, filtering only some per-request row state, reusing a text-only prefix key or treating a stop flag as immediate MLX work cancellation can corrupt lifecycle reasoning.",
+ "refusal_point": "Stop at the first unsupported contract or missing required fixture/backend; retain a targeted plan without inferring paged allocation, radix sharing or instantaneous prefill interruption."
},
"limitations": [
"Source-derived candidate; no local native-backend, model, service or efficacy validation.",
@@ -44,5 +49,10 @@
"provenance_readsets": [
"references/source-map.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Resolve the library versus server entry and its batching gates before transferring BatchGenerator behavior.",
+ "Keep UID, caches, tokens, samplers, processors, lengths, stop matchers and pending outputs aligned under the same removal selection, including prompt-stage requests.",
+ "Check exact token/model/adapter/draft identity and cache-type copy/trim semantics, then trace cooperative removal through survivor and next-request behavior."
+ ]
}
diff --git a/src/inferrsi/data/skill_library/repositories/mlx-swift-lm/metadata.json b/src/inferrsi/data/skill_library/repositories/mlx-swift-lm/metadata.json
index 20155fa4d..9c63f32f5 100644
--- a/src/inferrsi/data/skill_library/repositories/mlx-swift-lm/metadata.json
+++ b/src/inferrsi/data/skill_library/repositories/mlx-swift-lm/metadata.json
@@ -86,7 +86,11 @@
"method": "Stable-ID lifecycle and recurrent-kernel author lanes, integrated root/local references, hard cases and offline support/static/scaffold/package checks; native Swift/Xcode/MLX compiler/tests not_run."
},
"applicability": {
- "scope": "Selected Swift ownership/cancellation/settling and actual GatedDelta Metal/ops contracts. Common categorization is not CPU inference or Apple qualification."
+ "scope": "Selected Swift ownership/cancellation/settling and actual GatedDelta Metal/ops contracts. Common categorization is not CPU inference or Apple qualification.",
+ "prerequisites": "Access to the Swift lifecycle and selected GatedDelta contracts, with resolved MLX/native identities for a comparison; native checks require a prepared matching Swift/Xcode/Apple stack and separate authorization.",
+ "input_contract": "Exact source revision, resolved Swift/MLX dependencies, Xcode/SDK/OS/device, API family and native artifact identity; the selected question supplies lifecycle ownership/completion observations or recurrent producer shape/dtype/layout and actual consumer/fallback.",
+ "produced_artifact": "An identity and actual-versus-declared report for generation ownership or recurrent kernel invariants, with safe support result, counter-signal and smallest separately authorized native check.",
+ "failure_modes": "Conflating lexical isolation with exclusive access across await or completed task join, treating useKernel as observed dispatch, calling an ops fallback CPU execution, or assuming a manifest range is a resolved dependency can invalidate the conclusion."
},
"limitations": [
"BLOCKED_REQUIRED_BACKEND: native Swift/MLX compiler, host-control tests and required Apple Metal cancellation/backgrounding/kernel/cache/model/fit/quality/service/performance evidence not_run.",
@@ -100,5 +104,10 @@
"references/repo-provenance.md",
"references/source-identity.json"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Resolve mutable-context/cache ownership and task settlement separately from lexical isolation and queued GPU work before cache reuse or producer shutdown.",
+ "Follow recurrent shape/dtype/layout to the actual Metal or ops consumer; predicate eligibility and useKernel alone do not prove dispatch.",
+ "Record the actual dependency resolution and use the documented Xcode macOS test route; the source states swift test does not work for this package."
+ ]
}
diff --git a/src/inferrsi/data/skill_library/repositories/project-maya/metadata.json b/src/inferrsi/data/skill_library/repositories/project-maya/metadata.json
index 328798e0c..46142880b 100644
--- a/src/inferrsi/data/skill_library/repositories/project-maya/metadata.json
+++ b/src/inferrsi/data/skill_library/repositories/project-maya/metadata.json
@@ -132,7 +132,12 @@
"method": "Stable-ID isolated author lanes, integrated runtime graph, assertion/helper/static and selected direct-compiler CPU verification; required accelerator/model blocks retained."
},
"applicability": {
- "scope": "Model-specific dispatch/residency, API/native state and qualification investigation. Common catalog placement is not CPU inference or cross-device certification."
+ "scope": "Model-specific dispatch/residency, API/native state and qualification investigation. Common catalog placement is not CPU inference or cross-device certification.",
+ "prerequisites": "Access to the selected model-family runtime/state/qualification contract; native changes need frozen correctness/comparison requirements and the matching authorized CUDA/HIP/model resources.",
+ "input_contract": "Full source revision, actual executable/model family, model and asset identities, representation, compiler/runtime/device, rendered workload, sampling policy, single/parallel route, exact tokens/images and relevant completion observations.",
+ "produced_artifact": "A route/identity/evidence-class report with invariant, proposed check, actual result, counter-signal, missing resource and next authorized action; missing native requirements remain INCONCLUSIVE.",
+ "failure_modes": "Transferring coverage through a common facade or quantization name, confusing API history with native KV/private state, double-counting unified RAM/VRAM, or repairing a slot header to appear compatible can invalidate diagnosis.",
+ "refusal_point": "When required CUDA/HIP/model evidence is unavailable, retain BLOCKED_REQUIRED_BACKEND and an unverified/INCONCLUSIVE conclusion; exit-zero skips or CPU simulations cannot clear it."
},
"limitations": [
"BLOCKED_REQUIRED_BACKEND: no actual CUDA/HIP/model/device, quantized payload, vision, parallel/state/restore or cancellation-cleanup evidence.",
@@ -145,5 +150,10 @@
"references/repo-provenance.md",
"references/source-identity.json"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Select the family-specific route and preserve the explicitly shallow Qwen coverage rather than transferring deeper GLM/DeepSeek/MiniMax evidence.",
+ "Separate shared weights, private conversation state, API history and native KV; count unified physical memory once.",
+ "Preserve exact token/image identities and completion boundaries before reuse, then test one falsifiable hypothesis without modifying slot headers to bypass compatibility."
+ ]
}
diff --git a/src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/metadata.json b/src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/metadata.json
index 25375769a..0b7873ee8 100644
--- a/src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/metadata.json
+++ b/src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/metadata.json
@@ -50,7 +50,12 @@
}
],
"applicability": {
- "scope": "Bounded monolithic multimodal_gen composed-pipeline state and lifecycle diagnosis; actual backend/model behavior requires native validation."
+ "scope": "Bounded monolithic multimodal_gen composed-pipeline state and lifecycle diagnosis; actual backend/model behavior requires native validation.",
+ "prerequisites": "Pinned source or a concrete trace of the selected composed-pipeline executor and installed residency manager, with the applicable monolithic route identified.",
+ "input_contract": "Source/build, backend and executor/pipeline class, model revision, modality/shape, steps/scheduler/seed/output count, grouped/serial route, ranks, placement, warmup/save settings and symptom; timeout cases need correlated request/job and client/scheduler/worker events.",
+ "produced_artifact": "A version-bound owner/event ledger, first unresolved boundary, causal hypothesis, smallest real discriminating check, exact results and missing dependencies.",
+ "failure_modes": "Assuming all executors share the SyncExecutor contract, treating closed component use as eviction, assuming grouped schedulers are isolated, or reading socket close/job deletion as stopped device compute can misdiagnose ownership.",
+ "refusal_point": "Stop performance claims when the executing path, cleanup or output quality is unresolved; retain the failure and a known baseline rather than inferring a blanket leak or safety conclusion."
},
"limitations": [
"Source-derived candidate; upstream runtime and efficacy validation not run.",
@@ -60,5 +65,10 @@
"provenance_readsets": [
"references/source-notes.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Trace the actual executor and separate request, stage, component-use, module, scheduler, finalizer and allocator-cache lifetimes.",
+ "Check scheduler aliases and declared output-copy policies before introducing overlap; compare real serial/grouped outputs with fixed timesteps and seeds.",
+ "Correlate worker, finalizer and device completion after client timeout or deletion; preserve an unknown execution outcome until that evidence exists."
+ ]
}
diff --git a/src/inferrsi/data/skill_library/repositories/splash/metadata.json b/src/inferrsi/data/skill_library/repositories/splash/metadata.json
index c851c0bdd..81cb03e26 100644
--- a/src/inferrsi/data/skill_library/repositories/splash/metadata.json
+++ b/src/inferrsi/data/skill_library/repositories/splash/metadata.json
@@ -102,7 +102,12 @@
"method": "User-approved native Metal/C++ and isolated source-Python adaptation; stable-ID lanes, integrated graph, assertion cases and create/verify stages. Required-backend block disables automatic admission."
},
"applicability": {
- "scope": "Pinned Splash representation, planning, state/serving and verification investigation; common discovery does not qualify a different Mac/model/backend."
+ "scope": "Pinned Splash representation, planning, state/serving and verification investigation; common discovery does not qualify a different Mac/model/backend.",
+ "prerequisites": "Access to the pinned representation/planning, serving-cache or qualification route; native checks need matching Apple/backend/model resources, an authorized workload and frozen semantic/measurement controls.",
+ "input_contract": "Source/build, macOS/device, target and compatible draft artifacts/formats, tokenizer/template/vision selection, context/cache/memory controls, KV/recurrent precision, ANE and verifier-row policy, concurrency/priority/sampling, quality controls and cache state.",
+ "produced_artifact": "An execution-identity report with observation, source-backed invariant, alternate explanation, smallest discriminating check, actual result and unresolved Metal/ANE/model/disk-state gap.",
+ "failure_modes": "Treating an API alias as new weights, equal bit width as loader/kernel compatibility, or a fixed seed as fixed cache/batch/adaptive-verification scheduling can invalidate the comparison; CPU cache/codec checks do not prove GPU state.",
+ "refusal_point": "Stop at unsupported formats, unavailable backend/model resources, unsafe service exposure or the authorized budget."
},
"limitations": [
"Required Metal/ANE/model/actual device and disk-state execution remains BLOCKED_REQUIRED_BACKEND; Python policy/codec checks do not clear it.",
@@ -114,5 +119,10 @@
"references/repo-provenance.md",
"references/source-identity.json"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Resolve representation, memory/cache owner, validity key and reclaim/rollback boundary before tuning a numeric control.",
+ "Freeze semantic and workload controls beyond the seed, and retain actual-path fallbacks, admission waits, evictions, cancellations and errors.",
+ "Inspect options without starting the normal serve path, which can download target/draft weights; installation and serving require separate task authority."
+ ]
}
diff --git a/src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/metadata.json b/src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/metadata.json
index 0e177ce5a..4fa47651f 100644
--- a/src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/metadata.json
+++ b/src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/metadata.json
@@ -2,22 +2,66 @@
"profile": "local-skill-v1",
"schema_version": 1,
"kind": "skill",
- "origin": ["external"],
+ "origin": [
+ "external"
+ ],
"status": "candidate",
"empirical_validation": "not_run",
"license": "MIT",
- "tags": ["stable-diffusion-cpp", "diffusion", "model-residency", "workspace", "prefetch", "state-ownership", "weight-loading", "storage-io"],
+ "tags": [
+ "stable-diffusion-cpp",
+ "diffusion",
+ "model-residency",
+ "workspace",
+ "prefetch",
+ "state-ownership",
+ "weight-loading",
+ "storage-io"
+ ],
"sources": [
{
"type": "git",
"url": "https://github.com/leejet/stable-diffusion.cpp",
"commit": "b13a48219d5cdf408edcb93da6cd69b612ed03d3",
- "paths": ["src/stable-diffusion.cpp", "src/pipeline/image.cpp", "src/pipeline/diffusion_engine.h", "src/pipeline/diffusion_engine.cpp", "src/core/ggml_runner.cpp", "src/core/segment_weight_pipeline.cpp", "src/core/compute_workspace.cpp", "src/model_manager.cpp", "src/model_manager_prefetch.cpp", "src/device_residency_manager.h", ".github/workflows/build.yml", ".gitmodules", "cmake/ggml.cmake", "CMakeLists.txt", "examples/cli/main.cpp", "include/stable-diffusion.h", "LICENSE", "src/model_manager.h", "src/model_loader.h", "src/model_loader_files.cpp", "src/model_loader.cpp", "src/model_manager_files.cpp", "src/model.h", "src/core/util.h", "src/core/util.cpp", "src/model_io/tensor_storage.h", "src/model_io/safetensors_io.cpp"],
+ "paths": [
+ "src/stable-diffusion.cpp",
+ "src/pipeline/image.cpp",
+ "src/pipeline/diffusion_engine.h",
+ "src/pipeline/diffusion_engine.cpp",
+ "src/core/ggml_runner.cpp",
+ "src/core/segment_weight_pipeline.cpp",
+ "src/core/compute_workspace.cpp",
+ "src/model_manager.cpp",
+ "src/model_manager_prefetch.cpp",
+ "src/device_residency_manager.h",
+ ".github/workflows/build.yml",
+ ".gitmodules",
+ "cmake/ggml.cmake",
+ "CMakeLists.txt",
+ "examples/cli/main.cpp",
+ "include/stable-diffusion.h",
+ "LICENSE",
+ "src/model_manager.h",
+ "src/model_loader.h",
+ "src/model_loader_files.cpp",
+ "src/model_loader.cpp",
+ "src/model_manager_files.cpp",
+ "src/model.h",
+ "src/core/util.h",
+ "src/core/util.cpp",
+ "src/model_io/tensor_storage.h",
+ "src/model_io/safetensors_io.cpp"
+ ],
"readset": "references/source-notes.md"
}
],
"applicability": {
- "scope": "Source-grounded audit of native image-generation weight/runtime residency, optional segment prefetch and bounded non-ZIP safetensors catalog/read/staging ownership; confirm the actual source, GGML, backend, model and executing path independently."
+ "scope": "Source-grounded audit of native image-generation weight/runtime residency, optional segment prefetch and bounded non-ZIP safetensors catalog/read/staging ownership; confirm the actual source, GGML, backend, model and executing path independently.",
+ "prerequisites": "Access to the pinned source/storage boundaries and evidence of the actual image runner/backend dispatch; native capacity or prefetch checks require matching GGML, model and backend resources.",
+ "input_contract": "Runtime/GGML revisions and selection, compiled/loaded backends and devices, model/components and numerical image settings, effective memory/prefetch/placement settings and actual dispatch; retain available phase allocations, cache/workspace, transfer events and completed-image counts.",
+ "produced_artifact": "A phase-by-runner/backend/device ownership table for stored and resident weights, retained runtime, pending allocations, pin/prefetch ownership and release/eviction, with one discriminating check and its actual status.",
+ "failure_modes": "Treating max_vram as proof of segmentation, unpin as physical freeing, or configured prefetch as completed useful overlap can misidentify the capacity cause; missing observations cannot be read as zero.",
+ "refusal_point": "Stop when the cause is resolved or matching GGML/model/backend resources are unavailable; invoking this audit does not authorize dependency installation, model downloads, policy changes or device provisioning."
},
"limitations": [
"Source-derived candidate; native inference, allocator behavior, image quality and performance were not run.",
@@ -28,6 +72,14 @@
"Video/audio pipelines, detailed model kernels, approximate denoising caches and full multi-device execution remain outside this readset.",
"Loader rules select ordinary non-ZIP safetensors; full GGUF/Torch/ZIP parsers, special quantization/conversion correctness and native failure injection remain unqualified."
],
- "provenance_readsets": ["references/source-notes.md", "references/loader-storage-boundaries.md"],
- "consumption_mode": "packaged"
+ "provenance_readsets": [
+ "references/source-notes.md",
+ "references/loader-storage-boundaries.md"
+ ],
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Separate parameter storage, compute-resident weights, retained workspace/output caches and pending allocations for each runner and phase.",
+ "Verify the actual monolithic, segmented or multi-device path before interpreting a memory budget or prefetch setting.",
+ "For a policy comparison, hold model, inputs, numerical settings and device fixed; check completed images and quality before latency, and separate startup, first and warm requests."
+ ]
}
diff --git a/src/inferrsi/data/skill_library/repositories/vllm-metal/metadata.json b/src/inferrsi/data/skill_library/repositories/vllm-metal/metadata.json
index e9be25b3f..5df0e0f7e 100644
--- a/src/inferrsi/data/skill_library/repositories/vllm-metal/metadata.json
+++ b/src/inferrsi/data/skill_library/repositories/vllm-metal/metadata.json
@@ -89,7 +89,11 @@
"method": "Stable-ID manual author sequencing plus bounded cache author lane; integrated graph followed by assertion/helper/static and selected pure Python source tests; Apple native blocks retained."
},
"applicability": {
- "scope": "Selected plugin preparation, KV/offload ownership and speculative/deferred/dispatch evidence investigation. Common catalog scope does not certify CPU inference or cross-backend qualification."
+ "scope": "Selected plugin preparation, KV/offload ownership and speculative/deferred/dispatch evidence investigation. Common catalog scope does not certify CPU inference or cross-backend qualification.",
+ "prerequisites": "Access to the plugin-specific platform/ABI, cache-lifecycle and qualification contracts; actual engine changes require separate authority, resource ownership, frozen gates and cleanup, with a supported artifact pairing.",
+ "input_contract": "Source/package/runtime/native artifact identities, Python/OS/chip/device, backend/executor, checkpoint/tokenizer/draft identities, format/dtype, effective feature flags and workload.",
+ "produced_artifact": "An identity/route/invariant report with actual safe-check result, counter-signal and next authorized native check; partial CPU evidence stays separate from unverified transfers, sleep/wake, model and HTTP outcomes.",
+ "failure_modes": "Using lazy import success as native registration or ABI evidence, treating a feature switch as dispatch, or conflating cache capacity with physical residency and kernel time with HTTP time can overstate plugin behavior."
},
"limitations": [
"BLOCKED_REQUIRED_BACKEND: actual Apple MLX/Metal and MPS kernels/model loading, cache transfers/sleep/wake and real worker dispatch are not verified.",
@@ -102,5 +106,10 @@
"references/repo-provenance.md",
"references/source-identity.json"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Resolve platform registration, backend/runner and native ABI pairing before capacity tuning; a package visibility check establishes only its reported software facts.",
+ "Keep MLX and MPS, full-attention and recurrent state, logical cache capacity and physical residency as separate contracts.",
+ "Preserve failed paths and distinguish parser/API checks from real transfers, sleep/wake, model output and HTTP performance evidence."
+ ]
}
diff --git a/src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/metadata.json b/src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/metadata.json
index ca178bbed..2d8c751ed 100644
--- a/src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/metadata.json
+++ b/src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/metadata.json
@@ -54,7 +54,12 @@
}
],
"applicability": {
- "scope": "Pinned vLLM-RLT native Ouro synchronous scheduling with host KVMetadata: LAST_EXITED propagation versus SHARED, later depth-specific Torch-reference attention and normal release/wait/free order. No speculation, connector, preemption, prefix sharing, split/packed prefill or graph/static-buffer path qualification."
+ "scope": "Pinned vLLM-RLT native Ouro synchronous scheduling with host KVMetadata: LAST_EXITED propagation versus SHARED, later depth-specific Torch-reference attention and normal release/wait/free order. No speculation, connector, preemption, prefix sharing, split/packed prefill or graph/static-buffer path qualification.",
+ "prerequisites": "The selected native Ouro synchronous path uses host KVMetadata, no speculation/connector/preemption, atomic prefill and disabled graph/static-buffer/packed-prefill paths; changed execution conditions need a new bounded check.",
+ "input_contract": "Repository revision, model/configuration, recurrent-layer set, loop count, cache layout and block size, dtype/device, attention backend and effective flags, plus the traced token/depth/layer and its next consumer.",
+ "produced_artifact": "A bounded token/finalization/consumer/retirement trace separating source call order, observed data and written validity, completion evidence and remaining native model/backend/performance gaps.",
+ "failure_modes": "Substituting mark_finalized for K/V propagation, accepting allocated pages as a complete written prefix, applying LAST_EXITED copy expectations to SHARED, or treating event recording as completion can invalidate the ownership diagnosis.",
+ "refusal_point": "On a failed prerequisite or complete-prefix check, retain the first failing token/depth/layer and trace the missing operation. Stop the conclusion when flags or metadata ownership leave the selected path."
},
"limitations": [
"Source-derived candidate; native CPU/CUDA, model correctness/quality, performance and agent efficacy are not validated.",
@@ -65,5 +70,10 @@
"references/source-notes.md",
"references/source-identity.json"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Distinguish copied K/V values, logical written positions and allocated capacity; follow the later depth-specific attention consumer.",
+ "Preserve executed shallower values during LAST_EXITED propagation and use SHARED as a separate one-plane control.",
+ "Follow the latest request event through runner release before scheduler cache release; page reuse is not evidence of erased old bytes."
+ ]
}
diff --git a/src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/metadata.json b/src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/metadata.json
index 58235b46a..2240a2356 100644
--- a/src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/metadata.json
+++ b/src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/metadata.json
@@ -31,5 +31,18 @@
"provenance_readsets": [
"references/source-notes.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "applicability": {
+ "scope": "Pinned vLLM V1 scheduler/connector ownership audit for asynchronous KV load/save completion, retained blocks, preemption and failed receives; concrete connector conformance is not established by the interface.",
+ "prerequisites": "A concrete connector and worker/model-runner call trace at identified revisions, with request/block events sufficient to compare actual callers against the pinned scheduler/interface contracts.",
+ "input_contract": "Engine/connector revision and class, producer/consumer roles, KV-group count/block namespace, failure policy, concurrent-batch and graph modes; correlate bounded request IDs with allocation, matched tokens, load/save events, receive failures, preemption, finish and pool return.",
+ "produced_artifact": "An ownership/event ledger and bounded contract-conformance report naming the earliest evidenced broken obligation, owner, affected requests/blocks and smallest missing observation, with a proposed targeted regression check.",
+ "failure_modes": "Equating reserved tokens with usable KV, finished responses with reusable blocks, or send completion with scheduler write-fence completion can hide unsafe reuse; missing failed-receive completion can leave waits unreleased.",
+ "refusal_point": "Escalate absent concrete caller evidence, ambiguous block namespaces, unsupported layouts or revision drift; stop before inferring a backend fix, freeing retained blocks or claiming a reproduced bug."
+ },
+ "recommendations": [
+ "Trace lookup, allocation, worker execution and completion propagation before changing request state; repeated or undecided lookup and ordinary admission waits can be valid.",
+ "Check delayed-free ownership and scheduler in-flight-write fences separately, including empty-step save/metadata ordering in the real worker.",
+ "Keep receive failure and finished_recving correlated, and use request-level failure evidence for multi-group layouts while preserving the existing recovery policy."
+ ]
}
diff --git a/src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/metadata.json b/src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/metadata.json
index c6f7a6a10..e4470c935 100644
--- a/src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/metadata.json
+++ b/src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/metadata.json
@@ -33,5 +33,18 @@
"references/papers-20261015/references/orca-readset.md",
"references/source-notes.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "applicability": {
+ "scope": "Pinned vLLM V1 branch-based diagnosis of queued requests, low scheduled-token counts and KV-pressure preemption, separating admission/token/input/allocation/connector-wait constraints.",
+ "prerequisites": "The deployed scheduler must be mapped to the pinned branches, with effective configuration and a bounded request/step trace; another revision requires re-establishing that correspondence.",
+ "input_contract": "Deployed revision, scheduler class/configuration, connector role, KV layout, policy and speculative/encoder/concurrent-batch features; per-step queues, tokens, both budgets, allocation results, victims, transfer/free state and timing.",
+ "produced_artifact": "A per-step accounting ledger, observed limiting branch and competing explanation, missing instrumentation and one conditional configuration/progress hypothesis with predicted effects and trade-offs.",
+ "failure_modes": "Inferring memory exhaustion from a queue or high KV usage, conflating token/input budgets, ignoring pending frees or reading async token accounting as load completion can attribute admission stalls to the wrong branch.",
+ "refusal_point": "When required state is missing, leave the diagnosis unresolved and request the smallest trace; escalate allocation or connector behavior outside the readset and stop before live mutation or performance claims."
+ },
+ "recommendations": [
+ "Trace the first actual blocking or skip branch and record running versus admission caps and token versus input budgets separately.",
+ "Inspect deferred frees and request-free eligibility before blaming preemption victim selection, then reconcile restored budgets for an already-scheduled victim.",
+ "Reconstruct waiting admission independently, including queue precedence and async prefix states; change no setting without a branch-specific hypothesis."
+ ]
}
From 57bc805db9bb56995e2ae988830e1c52a14f85d3 Mon Sep 17 00:00:00 2001
From: Dong1017 <55723162+Dong1017@users.noreply.github.com>
Date: Sat, 31 Oct 2026 11:58:49 +0800
Subject: [PATCH 02/11] docs: regenerate skill projections from published
source history
Keep the canonical source snapshot reachable as a preceding commit.
Agent-assisted-by: Codex / gpt-6-astra [effort=xhigh] (author)
Signed-off-by: Dong1017 <55723162+Dong1017@users.noreply.github.com>
---
docs/skill-coverage.md | 62 +++++++++----------
.../repo_packs/inferrsi/knowledge/README.md | 4 +-
.../ascend-aclgraph-capture-replay.json | 4 +-
.../ascend/skills/ascend-device-health.json | 4 +-
.../skills/ascend-distributed-parallel.json | 4 +-
.../ascend/skills/ascend-graph-mode.json | 4 +-
.../skills/ascend-model-adaptation.json | 4 +-
.../skills/ascend-operator-profiling.json | 4 +-
.../ascend/skills/ascend-operator-tuning.json | 4 +-
.../skills/ascend-optimization-casebook.json | 4 +-
.../skills/ascend-performance-analysis.json | 4 +-
.../ascend-video-diffusion-optimization.json | 4 +-
.../ascend/skills/ascendc-tiling-design.json | 4 +-
.../skills/cann-attribution-overlap.json | 4 +-
.../torchnpu-device-memory-lifecycle.json | 4 +-
...torchnpu-pinned-host-memory-lifecycle.json | 4 +-
.../skills/assess-serving-preemption.json | 4 +-
.../assess-split-backward-schedule.json | 4 +-
.../skills/audit-fixed-workload-claim.json | 4 +-
.../audit-h3-metal-fasth3-semantics.json | 32 +++++++---
.../skills/audit-h3-runtime-residency.json | 32 +++++++---
.../skills/audit-llama-runtime-state.json | 32 +++++++---
.../audit-recurrent-kv-finalization.json | 32 +++++++---
.../common/skills/audit-sd-cpp-residency.json | 32 +++++++---
.../audit-sglang-diffusion-lifetimes.json | 32 +++++++---
.../skills/audit-vllm-kv-lifetimes.json | 35 ++++++++---
.../bound-windowed-attention-replay.json | 30 ++++++---
.../budget-speculative-verification.json | 30 ++++++---
.../skills/calibrate-serving-simulation.json | 4 +-
.../skills/cgroup-resource-controls.json | 4 +-
.../checkpoint-recovery-consistency.json | 4 +-
.../skills/choose-parallel-training-plan.json | 4 +-
.../skills/cite-check-downstream-pr.json | 4 +-
.../common/skills/claim-contract.json | 4 +-
.../knowledge/common/skills/comfy-ui.json | 32 +++++++---
.../common/skills/competitor-scan.json | 4 +-
.../skills/dcp-async-save-performance.json | 35 ++++++++---
.../skills/dense-performance-evidence.json | 4 +-
.../diagnose-vllm-admission-pressure.json | 35 ++++++++---
.../skills/diffusers-invocation-state.json | 32 +++++++---
.../common/skills/draft-acceptance-issue.json | 4 +-
.../knowledge/common/skills/dwarfstar.json | 32 +++++++---
.../evaluate-elastic-serving-placement.json | 4 +-
.../skills/generation-cache-lifecycle.json | 32 +++++++---
.../common/skills/host-memory-management.json | 4 +-
.../skills/host-overhead-optimization.json | 4 +-
.../knowledge/common/skills/husky.json | 32 +++++++---
.../skills/inference-benchmark-design.json | 4 +-
.../inference-optimization-casebook.json | 4 +-
.../common/skills/kernel-writing-guide.json | 4 +-
.../common/skills/kv-tier-management.json | 33 +++++++---
.../lazy-evaluation-stream-lifetimes.json | 32 +++++++---
.../knowledge/common/skills/lily.json | 32 +++++++---
.../knowledge/common/skills/lithos-metal.json | 30 ++++++---
.../knowledge/common/skills/local-dream.json | 30 ++++++---
.../knowledge/common/skills/mistral-rs.json | 30 ++++++---
.../knowledge/common/skills/mlc-llm.json | 30 ++++++---
.../knowledge/common/skills/mlx-swift-lm.json | 30 ++++++---
.../common/skills/npu-skill-factory.json | 4 +-
.../skills/optimization-search-policy.json | 4 +-
.../common/skills/os-process-placement.json | 4 +-
.../skills/performance-ceiling-analysis.json | 4 +-
.../skills/plan-activation-residency.json | 4 +-
.../knowledge/common/skills/project-maya.json | 32 +++++++---
.../skills/repository-evolution-analysis.json | 4 +-
.../knowledge/common/skills/rsi-evidence.json | 4 +-
.../skills/skill-evolution-playbook.json | 4 +-
.../knowledge/common/skills/splash.json | 32 +++++++---
.../common/skills/storage-io-measurement.json | 4 +-
.../skills/training-input-pipeline.json | 4 +-
.../skills/training-resource-budget.json | 4 +-
.../skills/video-diffusion-optimization.json | 4 +-
.../knowledge/common/skills/vllm-metal.json | 30 ++++++---
.../skills/weight-loading-optimization.json | 4 +-
.../cuda/skills/audit-afd-p2p-exchanges.json | 32 +++++++---
.../skills/audit-flare-session-lifecycle.json | 32 +++++++---
.../diagnose-attention-bottlenecks.json | 4 +-
.../cuda/skills/host-overhead-profiling.json | 4 +-
.../cuda/skills/reason-hopper-kernels.json | 4 +-
79 files changed, 823 insertions(+), 357 deletions(-)
diff --git a/docs/skill-coverage.md b/docs/skill-coverage.md
index 0fbd111e4..8c08db27c 100644
--- a/docs/skill-coverage.md
+++ b/docs/skill-coverage.md
@@ -4,7 +4,7 @@ Generated from the registry, the skill sidecars and the two maintained
matrices. **Do not edit by hand**; regenerate after any registry, skill,
matrix or frozen-plan change.
-- Input digest: `sha256:b6f5cec72e5e52fcc4113bf50c68612417688b40c54cc79cbd24a611db574f3c`
+- Input digest: `sha256:e3d541405e1c345056e3873a283604e04d74fac3cc1f1e76e56389e971624f0e`
- Regenerate: `python tools/build_skill_coverage.py`
- Verify without writing: `python tools/build_skill_coverage.py --check`
@@ -79,40 +79,40 @@ optimization is fully covered or that any procedure has measured benefit.
| Skill | Backend | Declared task | Applicability | Conditions | SearchSpace layer / stage | Reference Matrix | State | Gaps |
|---|---|---|---|---|---|---|---|---|
-| [dcp-async-save-performance](../src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/SKILL.md) | common | Diagnose training-visible checkpoint pauses using PyTorch Distributed Checkpoint asynchronous saving and staging ... | **unknown** | — | 1 Accelerator · Diagnose / Configure / Optimize
2 RAM/Host · Observe / Profile / Configure / Optimize / Verify
3 SSD · Observe / Profile / Configure / Optimize / Verify | 1 · Systems sources (C1 / DCP async-save-only approval — pytorch/pytorch @ ...) | candidate · not_run · 1 readset(s) | no-applicability, no-eval |
-| [diffusers-invocation-state](../src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/SKILL.md) | common | Trace Diffusers pipeline-wrapper and scheduler invocation state when repeated calls, from_pipe reuse or interruption ... | FluxPipeline and FlowMatchEulerDiscreteScheduler invocation ownership at this source pin ... | — | 2 RAM/Host †guidance: Diffusers invocation guidance (source-derived) | Diffusers invocation source extraction (huggingface/diffusers @ `aea7b70be0f594bcdc15a6f2677ee2ca85e49e8e` ...) | candidate · not_run · 1 readset(s) | no-eval |
-| [lazy-evaluation-stream-lifetimes](../src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/SKILL.md) | common | Diagnose MLX lazy evaluation, cross-stream ordering and array-buffer lifetimes when timing, memory use or asynchronous ... | MLX evaluation/stream/buffer diagnosis at the pinned implementation; qualify the actual ... | — | 1 Accelerator †guidance: MLX framework guidance (source-derived) | MLX framework source extraction (ml-explore/mlx @ `bff10220c742907f0c68e55ff174aaea56f74e3d` (MIT)) | candidate · not_run · 1 readset(s) | no-eval |
+| [dcp-async-save-performance](../src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/SKILL.md) | common | Diagnose training-visible checkpoint pauses using PyTorch Distributed Checkpoint asynchronous saving and staging ... | Source-supported diagnosis of training-visible PyTorch DCP save pauses ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator · Diagnose / Configure / Optimize
2 RAM/Host · Observe / Profile / Configure / Optimize / Verify
3 SSD · Observe / Profile / Configure / Optimize / Verify | 1 · Systems sources (C1 / DCP async-save-only approval — pytorch/pytorch @ ...) | candidate · not_run · 1 readset(s) | no-eval |
+| [diffusers-invocation-state](../src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/SKILL.md) | common | Trace Diffusers pipeline-wrapper and scheduler invocation state when repeated calls, from_pipe reuse or interruption ... | FluxPipeline and FlowMatchEulerDiscreteScheduler invocation ownership at this source pin ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 2 RAM/Host †guidance: Diffusers invocation guidance (source-derived) | Diffusers invocation source extraction (huggingface/diffusers @ `aea7b70be0f594bcdc15a6f2677ee2ca85e49e8e` ...) | candidate · not_run · 1 readset(s) | no-eval |
+| [lazy-evaluation-stream-lifetimes](../src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/SKILL.md) | common | Diagnose MLX lazy evaluation, cross-stream ordering and array-buffer lifetimes when timing, memory use or asynchronous ... | MLX evaluation/stream/buffer diagnosis at the pinned implementation; qualify the actual ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator †guidance: MLX framework guidance (source-derived) | MLX framework source extraction (ml-explore/mlx @ `bff10220c742907f0c68e55ff174aaea56f74e3d` (MIT)) | candidate · not_run · 1 readset(s) | no-eval |
## Registered: Repositories (26)
| Skill | Backend | Declared task | Applicability | Conditions | SearchSpace layer / stage | Reference Matrix | State | Gaps |
|---|---|---|---|---|---|---|---|---|
-| [audit-afd-p2p-exchanges](../src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/SKILL.md) | cuda | Audit CUDA P2pNccl Attention/FFN rank namespaces and per-stage, per-peer hidden-state ownership when uneven worker ... | Source-based CUDA P2pNccl rank/payload audit; selected eager dense-FFN path with ... | — | 1 Accelerator · Diagnose / Configure / Verify | 1 · Systems sources (vllm-project/afd-plugin @ `df8d6a0c403c3ca1b4ed9cf1687e7204908bebec` ...) | candidate · not_run · 1 readset(s) | no-eval |
-| [audit-flare-session-lifecycle](../src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/SKILL.md) | cuda | Audit Flare complete prefix/session restoration, inbound transfer tickets, publication/use/retirement fences and ... | Pinned Flare complete-session restore and admission audit for its NVIDIA SM120/SM121 ... | — | 1 Accelerator †guidance: Flare preparation and complete-session source contracts
2 RAM/Host †guidance: Flare preparation and complete-session source contracts
3 SSD †guidance: Flare preparation and complete-session source contracts | Flare preparation and complete-session source knowledge (NVIDIA/flare-local-inference → [NVIDIA/local-inference](https://github ...) | candidate · not_run · 2 readset(s) | no-eval |
-| [audit-h3-metal-fasth3-semantics](../src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/SKILL.md) | common | Audit h3-metal FastH3 effective configuration, checkpoint tensor and schedule semantics, partial VSA tiles, sparse ... | h3-metal FastH3 requested/effective configuration, checkpoint transformation and ... | — | 1 Accelerator · Observe / Diagnose / Verify | 1 · Systems sources (zuiho-kai/h3-metal @ `4ba47a3eabdbebf801ac6664b31f8ec168bc57a8` ...) | candidate · not_run · 3 readset(s) | no-eval |
-| [audit-h3-runtime-residency](../src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/SKILL.md) | common | Audit MiniMax-H3 generation in h3.c on Apple Silicon by tracing conditioning, AdaLN precompute, DiT residency, Metal ... | MiniMax-H3 joint audio/video C/Metal runtime audit: raw checkpoint compatibility, input ... | — | 1 Accelerator · Observe / Diagnose / Verify
2 RAM/Host · Observe / Diagnose / Verify
3 SSD · Diagnose / Configure / Verify | 1 · Systems sources (antirez/h3.c @ `8974cc055ea9c02fcd14cc27dfda3e1027c05153` (MIT ...) | candidate · not_run · 3 readset(s) | no-eval |
-| [audit-llama-runtime-state](../src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/SKILL.md) | common | Audit llama.cpp context memory, partial decode progress, graph reuse and server cancellation when a retry, stale ... | Pinned llama.cpp context-memory, graph-reuse and server task-lifecycle diagnosis ... | — | 1 Accelerator · Diagnose / Verify
2 RAM/Host · Diagnose / Verify | 1 · Systems sources (ggml-org/llama.cpp @ `707616feb5f379e99ff7442680a55b4a1704c7c9` (MIT ...) | candidate · not_run · 1 readset(s) | no-eval |
-| [audit-recurrent-kv-finalization](../src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/SKILL.md) | common | Audit vLLM-RLT synchronous early-exit KV propagation, later depth-specific attention reads, and request storage ... | Pinned vLLM-RLT native Ouro synchronous scheduling with host KVMetadata: LAST_EXITED ... | — | — | 1 · Systems sources (InferMatrixLab/vllm-rlt (legacy ThinkFlowLab/vllm-rlt) @ ...) | candidate · not_run · 2 readset(s) | no-SearchSpace-layer, no-eval |
-| [audit-sd-cpp-residency](../src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/SKILL.md) | common | Audit stable-diffusion.cpp image-generation memory pressure by separating catalog freshness and weight-byte delivery ... | Source-grounded audit of native image-generation weight/runtime residency, optional ... | — | 1 Accelerator · Observe / Diagnose / Verify
2 RAM/Host · Observe / Diagnose / Verify
3 SSD · Diagnose / Verify | 1 · Systems sources (leejet/stable-diffusion.cpp @ `b13a48219d5cdf408edcb93da6cd69b612ed03d ...) | candidate · not_run · 2 readset(s) | no-eval |
-| [audit-sglang-diffusion-lifetimes](../src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/SKILL.md) | common | Diagnose SGLang multimodal_gen stage state, component residency and request timeout/deletion when grouped outputs ... | Bounded monolithic multimodal_gen composed-pipeline state and lifecycle diagnosis; actual ... | — | 1 Accelerator · Diagnose / Verify
2 RAM/Host · Diagnose / Verify | 1 · Systems sources (sgl-project/sglang @ `df13d5dce31abfeff0e701a62e8c39a136758e6c` ...) | candidate · not_run · 1 readset(s) | no-eval |
-| [audit-vllm-kv-lifetimes](../src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/SKILL.md) | common | Audit vLLM V1 asynchronous KV load/save completion, block retention, preemption, and receive-failure contracts when ... | **unknown** | — | 1 Accelerator · Observe / Diagnose / Verify | 1 · Systems sources (vllm-project/vllm @ `7f730b4cea9a66520a9200265928ebe016b48a39` ...) | candidate · not_run · 1 readset(s) | no-applicability, no-eval |
-| [bound-windowed-attention-replay](../src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/SKILL.md) | common | Bound replay or prefill query rows across a producer/consumer boundary whose retained inputs and separately preserved ... | Decoders with an explicit producer/consumer boundary and a demonstrated dependency ... | — | 1 Accelerator †guidance: DeepInfer bounded source guidance
2 RAM/Host †guidance: DeepInfer bounded source guidance | DeepInfer source extraction (deepseek-ai/DeepInfer @ `8b16149741a4935e1dec8cea27853f273b374c5b` ...) | candidate · not_run · 2 readset(s) | no-eval |
-| [budget-speculative-verification](../src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/SKILL.md) | common | Decide how many draft tokens to verify per decode step when acceptance varies by request and position. Use when an ... | Engines with multi-token speculative decoding whose verification forward cost grows with ... | — | 1 Accelerator †guidance: DeepInfer bounded source guidance | DeepInfer source extraction (deepseek-ai/DeepInfer @ `8b16149741a4935e1dec8cea27853f273b374c5b` ...) | candidate · not_run · 2 readset(s) | no-eval |
-| [comfy-ui](../src/inferrsi/data/skill_library/repositories/comfy-ui/SKILL.md) | common | Audit ComfyUI workflow-output cache validity, lazy dependencies and model-weight residency as separate state contracts ... | Source-grounded ComfyUI workflow-output cache and model-weight ownership audit; confirm ... | — | 1 Accelerator · Diagnose / Configure / Verify
2 RAM/Host · Diagnose / Configure / Verify | 1 · Systems sources (Comfy-Org/ComfyUI @ `b1a247c9cf9dbfda4b2c0a0c1d50950667eeaa0c` (GNU ...) | candidate · not_run · 1 readset(s) | no-eval |
-| [diagnose-vllm-admission-pressure](../src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/SKILL.md) | common | Diagnose vLLM V1 admission stalls and KV-pressure preemption by separating token, input, active-request ... | **unknown** | — | 1 Accelerator · Observe / Diagnose / Configure / Optimize / Verify | 1 · Systems sources (vllm-project/vllm @ `7f730b4cea9a66520a9200265928ebe016b48a39` ...) | candidate · not_run · 14 readset(s) | no-applicability, no-eval |
-| [dwarfstar](../src/inferrsi/data/skill_library/repositories/dwarfstar/SKILL.md) | common | Investigate DwarfStar model fit, expert SSD streaming, session and vision cache correctness, speculative sampling and ... | Pinned DwarfStar source investigation and experiment planning; not arbitrary GGUF support ... | — | — | DwarfStar portable AREX extraction (antirez/ds4 @ `ee7e8405232c8971000cc7ba2b945b00bf6d20d3`; AREX-Skill ...) | candidate · partial_cpu_only · 2 readset(s) | no-SearchSpace-layer, no-eval |
-| [generation-cache-lifecycle](../src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/SKILL.md) | common | Trace MLX-LM prefill, generation batching, prompt-cache reuse and cooperative stopping when request state, cache rows ... | MLX-LM generation, batch-row and prefix-cache diagnosis; model/backend/runtime ... | — | 2 RAM/Host †guidance: MLX-LM application guidance (source-derived) | MLX-LM application source extraction (ml-explore/mlx-lm @ `ab7f7048dc341ced2d84b1887eeb17d60b0d1b02` (MIT)) | candidate · not_run · 1 readset(s) | no-eval |
-| [husky](../src/inferrsi/data/skill_library/repositories/husky/SKILL.md) | common | Review Husky's official-document claims as research hypotheses, separating author reports, implementation evidence ... | Review Husky document claims and specify missing evidence; common catalog scope is not ... | — | — | Husky document-only research adaptation (Official https://husky.underdog.ai/; displayed2026-09-20 ...) | candidate · not_run · 2 readset(s) | no-SearchSpace-layer, no-eval |
-| [kv-tier-management](../src/inferrsi/data/skill_library/repositories/lmcache/kv-tier-management/SKILL.md) | common | Manage runtime KV cache tiers (GPU → CPU RAM → local disk) for LLM serving: decide when KV is stored, retained ... | **unknown** | — | 2 RAM/Host · Configure / Optimize
3 SSD · Optimize | 1 · Systems sources (LMCache/LMCache @ c3b875a) | candidate · not_run · 0 readset(s) | no-applicability, no-eval |
-| [lily](../src/inferrsi/data/skill_library/repositories/lily/SKILL.md) | common | Investigate Lily's exact Qwen3.6-35B-A3B MLX-affine-Q4 contract, Metal state and kernels, strict greedy text API ... | Exact Lily source/checkpoint/layout/API/cache/qualification investigation; common catalog ... | — | — | Lily portable AREX extraction (perplexityai/pplx-garden `lily/` @ `1ed972ed3f0bd5616c997c9507c25616c6 ...) | candidate · not_run · 2 readset(s) | no-SearchSpace-layer, no-eval |
-| [lithos-metal](../src/inferrsi/data/skill_library/repositories/lithos-metal/SKILL.md) | common | Inspect Lithos Metal compilation and fusion, GPU-resident generation and DSpark state, or local serving and ... | Pinned Lithos Metal 0.1.2 source/API contracts; Apple silicon Metal runtime; model and ... | — | 1 Accelerator · Diagnose / Configure / Optimize / Verify | 1 · Systems sources (lithos-ai/lithos-metal @ `6653b7df5f2c4392ec7b038960f2796b1499923f` ...) | candidate · not_run · 1 readset(s) | no-eval |
-| [local-dream](../src/inferrsi/data/skill_library/repositories/local-dream/SKILL.md) | common | Audit Local Dream Android desired-versus-serving process state, remote exposure, native backend and DiT packaging/ABI ... | Selected Android app/process and native diffusion boundary contracts; common is a ... | — | 2 RAM/Host †guidance: Local Dream Android diffusion source contracts | Local Dream Android diffusion boundary source (xororz/local-dream @ `db162df4d9bb6efc6595250423d8df26241f4c27` ...) | candidate · not_run · 2 readset(s) | no-eval |
-| [mistral-rs](../src/inferrsi/data/skill_library/repositories/mistral-rs/SKILL.md) | common | Investigate mistral.rs Metal paged-attention decode layouts, prefill coordinates and workspace, actual-model prewarm ... | Selected native Metal paged decode/prefill layout/metadata/workspace and ... | — | 1 Accelerator †guidance: mistral.rs Metal paged-attention source contracts
2 RAM/Host †guidance: mistral.rs Metal paged-attention source contracts | mistral.rs Metal paged-attention extraction (EricLBuehler/mistral.rs @ `9459891e7697238225885518119682ac8209a612` ...) | candidate · not_run · 2 readset(s) | no-eval |
-| [mlc-llm](../src/inferrsi/data/skill_library/repositories/mlc-llm/SKILL.md) | common | Investigate MLC LLM compiled Metal artifact/schema, Apple target and pinned TVM module/storage/stream contracts, and ... | Selected artifact/Apple target/pinned TVM Metal consumer and KV configuration source ... | — | 1 Accelerator †guidance: MLC LLM artifact, target and runtime source contracts
2 RAM/Host †guidance: MLC LLM artifact, target and runtime source contracts | MLC LLM compiled Metal and runtime contracts (mlc-ai/mlc-llm @ `8978ea9aec626d64096aaa92fbc89bda2e3ea802`; TVM ...) | candidate · not_run · 2 readset(s) | no-eval |
-| [mlx-swift-lm](../src/inferrsi/data/skill_library/repositories/mlx-swift-lm/SKILL.md) | common | Investigate MLX Swift LM generation ownership, cancellation and settled task completion, and recurrent GatedDelta Metal ... | Selected Swift ownership/cancellation/settling and actual GatedDelta Metal/ops contracts ... | — | 1 Accelerator †guidance: MLX Swift LM ownership and recurrent Metal source contracts
2 RAM/Host †guidance: MLX Swift LM ownership and recurrent Metal source contracts | MLX Swift LM lifecycle and recurrent Metal extraction (ml-explore/mlx-swift-lm @ `3d066d3a0ffdc7e70c476a3d31488ea62dfdd0ac` ...) | candidate · not_run · 2 readset(s) | no-eval |
-| [project-maya](../src/inferrsi/data/skill_library/repositories/project-maya/SKILL.md) | common | Investigate Project Maya model-family dispatch, representation and memory tiers, API versus native state, speculative ... | Model-specific dispatch/residency, API/native state and qualification investigation ... | — | 1 Accelerator †guidance: Project Maya model-specific source contracts
2 RAM/Host †guidance: Project Maya model-specific source contracts
3 SSD †guidance: Project Maya model-specific source contracts | Project Maya source and CPU-slice extraction (mw00/project-maya @ `249f9a87d1a9d72bec2b684d704b0aa9296b1c9c` ...) | candidate · partial_cpu_only · 2 readset(s) | no-eval |
-| [splash](../src/inferrsi/data/skill_library/repositories/splash/SKILL.md) | common | Investigate Splash model-format compatibility, automatic memory planning, Metal and ANE execution, DFlash2 ... | Pinned Splash representation, planning, state/serving and verification investigation ... | — | — | Splash portable AREX extraction (incoai/splash @ `39a97385f8560cf8a93972cec9b0a963c9195250`; AREX ...) | candidate · partial_cpu_only · 2 readset(s) | no-SearchSpace-layer, no-eval |
-| [vllm-metal](../src/inferrsi/data/skill_library/repositories/vllm-metal/SKILL.md) | common | Investigate vllm-metal Apple inference through platform and native-ABI selection, paged KV/offload ownership ... | Selected plugin preparation, KV/offload ownership and speculative/deferred/dispatch ... | — | 1 Accelerator †guidance: vLLM Metal selected source contracts
2 RAM/Host †guidance: vLLM Metal selected source contracts
3 SSD †guidance: vLLM Metal selected source contracts | vLLM Metal portable AREX source and CPU extraction (vllm-project/vllm-metal @ `e7df7a6ff215f99c7842552d230fdbc709835934` ...) | candidate · partial_cpu_only · 2 readset(s) | no-eval |
+| [audit-afd-p2p-exchanges](../src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/SKILL.md) | cuda | Audit CUDA P2pNccl Attention/FFN rank namespaces and per-stage, per-peer hidden-state ownership when uneven worker ... | Source-based CUDA P2pNccl rank/payload audit; selected eager dense-FFN path with ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator · Diagnose / Configure / Verify | 1 · Systems sources (vllm-project/afd-plugin @ `df8d6a0c403c3ca1b4ed9cf1687e7204908bebec` ...) | candidate · not_run · 1 readset(s) | no-eval |
+| [audit-flare-session-lifecycle](../src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/SKILL.md) | cuda | Audit Flare complete prefix/session restoration, inbound transfer tickets, publication/use/retirement fences and ... | Pinned Flare complete-session restore and admission audit for its NVIDIA SM120/SM121 ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator †guidance: Flare preparation and complete-session source contracts
2 RAM/Host †guidance: Flare preparation and complete-session source contracts
3 SSD †guidance: Flare preparation and complete-session source contracts | Flare preparation and complete-session source knowledge (NVIDIA/flare-local-inference → [NVIDIA/local-inference](https://github ...) | candidate · not_run · 2 readset(s) | no-eval |
+| [audit-h3-metal-fasth3-semantics](../src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/SKILL.md) | common | Audit h3-metal FastH3 effective configuration, checkpoint tensor and schedule semantics, partial VSA tiles, sparse ... | h3-metal FastH3 requested/effective configuration, checkpoint transformation and ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator · Observe / Diagnose / Verify | 1 · Systems sources (zuiho-kai/h3-metal @ `4ba47a3eabdbebf801ac6664b31f8ec168bc57a8` ...) | candidate · not_run · 3 readset(s) | no-eval |
+| [audit-h3-runtime-residency](../src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/SKILL.md) | common | Audit MiniMax-H3 generation in h3.c on Apple Silicon by tracing conditioning, AdaLN precompute, DiT residency, Metal ... | MiniMax-H3 joint audio/video C/Metal runtime audit: raw checkpoint compatibility, input ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator · Observe / Diagnose / Verify
2 RAM/Host · Observe / Diagnose / Verify
3 SSD · Diagnose / Configure / Verify | 1 · Systems sources (antirez/h3.c @ `8974cc055ea9c02fcd14cc27dfda3e1027c05153` (MIT ...) | candidate · not_run · 3 readset(s) | no-eval |
+| [audit-llama-runtime-state](../src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/SKILL.md) | common | Audit llama.cpp context memory, partial decode progress, graph reuse and server cancellation when a retry, stale ... | Pinned llama.cpp context-memory, graph-reuse and server task-lifecycle diagnosis ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator · Diagnose / Verify
2 RAM/Host · Diagnose / Verify | 1 · Systems sources (ggml-org/llama.cpp @ `707616feb5f379e99ff7442680a55b4a1704c7c9` (MIT ...) | candidate · not_run · 1 readset(s) | no-eval |
+| [audit-recurrent-kv-finalization](../src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/SKILL.md) | common | Audit vLLM-RLT synchronous early-exit KV propagation, later depth-specific attention reads, and request storage ... | Pinned vLLM-RLT native Ouro synchronous scheduling with host KVMetadata: LAST_EXITED ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | — | 1 · Systems sources (InferMatrixLab/vllm-rlt (legacy ThinkFlowLab/vllm-rlt) @ ...) | candidate · not_run · 2 readset(s) | no-SearchSpace-layer, no-eval |
+| [audit-sd-cpp-residency](../src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/SKILL.md) | common | Audit stable-diffusion.cpp image-generation memory pressure by separating catalog freshness and weight-byte delivery ... | Source-grounded audit of native image-generation weight/runtime residency, optional ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator · Observe / Diagnose / Verify
2 RAM/Host · Observe / Diagnose / Verify
3 SSD · Diagnose / Verify | 1 · Systems sources (leejet/stable-diffusion.cpp @ `b13a48219d5cdf408edcb93da6cd69b612ed03d ...) | candidate · not_run · 2 readset(s) | no-eval |
+| [audit-sglang-diffusion-lifetimes](../src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/SKILL.md) | common | Diagnose SGLang multimodal_gen stage state, component residency and request timeout/deletion when grouped outputs ... | Bounded monolithic multimodal_gen composed-pipeline state and lifecycle diagnosis; actual ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator · Diagnose / Verify
2 RAM/Host · Diagnose / Verify | 1 · Systems sources (sgl-project/sglang @ `df13d5dce31abfeff0e701a62e8c39a136758e6c` ...) | candidate · not_run · 1 readset(s) | no-eval |
+| [audit-vllm-kv-lifetimes](../src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/SKILL.md) | common | Audit vLLM V1 asynchronous KV load/save completion, block retention, preemption, and receive-failure contracts when ... | Pinned vLLM V1 scheduler/connector ownership audit for asynchronous KV load/save ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator · Observe / Diagnose / Verify | 1 · Systems sources (vllm-project/vllm @ `7f730b4cea9a66520a9200265928ebe016b48a39` ...) | candidate · not_run · 1 readset(s) | no-eval |
+| [bound-windowed-attention-replay](../src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/SKILL.md) | common | Bound replay or prefill query rows across a producer/consumer boundary whose retained inputs and separately preserved ... | Decoders with an explicit producer/consumer boundary and a demonstrated dependency ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator †guidance: DeepInfer bounded source guidance
2 RAM/Host †guidance: DeepInfer bounded source guidance | DeepInfer source extraction (deepseek-ai/DeepInfer @ `8b16149741a4935e1dec8cea27853f273b374c5b` ...) | candidate · not_run · 2 readset(s) | no-eval |
+| [budget-speculative-verification](../src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/SKILL.md) | common | Decide how many draft tokens to verify per decode step when acceptance varies by request and position. Use when an ... | Engines with multi-token speculative decoding whose verification forward cost grows with ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator †guidance: DeepInfer bounded source guidance | DeepInfer source extraction (deepseek-ai/DeepInfer @ `8b16149741a4935e1dec8cea27853f273b374c5b` ...) | candidate · not_run · 2 readset(s) | no-eval |
+| [comfy-ui](../src/inferrsi/data/skill_library/repositories/comfy-ui/SKILL.md) | common | Audit ComfyUI workflow-output cache validity, lazy dependencies and model-weight residency as separate state contracts ... | Source-grounded ComfyUI workflow-output cache and model-weight ownership audit; confirm ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator · Diagnose / Configure / Verify
2 RAM/Host · Diagnose / Configure / Verify | 1 · Systems sources (Comfy-Org/ComfyUI @ `b1a247c9cf9dbfda4b2c0a0c1d50950667eeaa0c` (GNU ...) | candidate · not_run · 1 readset(s) | no-eval |
+| [diagnose-vllm-admission-pressure](../src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/SKILL.md) | common | Diagnose vLLM V1 admission stalls and KV-pressure preemption by separating token, input, active-request ... | Pinned vLLM V1 branch-based diagnosis of queued requests, low scheduled-token counts and ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator · Observe / Diagnose / Configure / Optimize / Verify | 1 · Systems sources (vllm-project/vllm @ `7f730b4cea9a66520a9200265928ebe016b48a39` ...) | candidate · not_run · 14 readset(s) | no-eval |
+| [dwarfstar](../src/inferrsi/data/skill_library/repositories/dwarfstar/SKILL.md) | common | Investigate DwarfStar model fit, expert SSD streaming, session and vision cache correctness, speculative sampling and ... | Pinned DwarfStar source investigation and experiment planning; not arbitrary GGUF support ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | — | DwarfStar portable AREX extraction (antirez/ds4 @ `ee7e8405232c8971000cc7ba2b945b00bf6d20d3`; AREX-Skill ...) | candidate · partial_cpu_only · 2 readset(s) | no-SearchSpace-layer, no-eval |
+| [generation-cache-lifecycle](../src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/SKILL.md) | common | Trace MLX-LM prefill, generation batching, prompt-cache reuse and cooperative stopping when request state, cache rows ... | MLX-LM generation, batch-row and prefix-cache diagnosis; model/backend/runtime ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 2 RAM/Host †guidance: MLX-LM application guidance (source-derived) | MLX-LM application source extraction (ml-explore/mlx-lm @ `ab7f7048dc341ced2d84b1887eeb17d60b0d1b02` (MIT)) | candidate · not_run · 1 readset(s) | no-eval |
+| [husky](../src/inferrsi/data/skill_library/repositories/husky/SKILL.md) | common | Review Husky's official-document claims as research hypotheses, separating author reports, implementation evidence ... | Review Husky document claims and specify missing evidence; common catalog scope is not ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | — | Husky document-only research adaptation (Official https://husky.underdog.ai/; displayed2026-09-20 ...) | candidate · not_run · 2 readset(s) | no-SearchSpace-layer, no-eval |
+| [kv-tier-management](../src/inferrsi/data/skill_library/repositories/lmcache/kv-tier-management/SKILL.md) | common | Manage runtime KV cache tiers (GPU → CPU RAM → local disk) for LLM serving: decide when KV is stored, retained ... | LMCache multiprocess runtime KV tier policy across engine GPU cache, host L1 and ... | failure_modes, input_contract, produced_artifact
3 advice | 2 RAM/Host · Configure / Optimize
3 SSD · Optimize | 1 · Systems sources (LMCache/LMCache @ c3b875a) | candidate · not_run · 0 readset(s) | no-eval |
+| [lily](../src/inferrsi/data/skill_library/repositories/lily/SKILL.md) | common | Investigate Lily's exact Qwen3.6-35B-A3B MLX-affine-Q4 contract, Metal state and kernels, strict greedy text API ... | Exact Lily source/checkpoint/layout/API/cache/qualification investigation; common catalog ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | — | Lily portable AREX extraction (perplexityai/pplx-garden `lily/` @ `1ed972ed3f0bd5616c997c9507c25616c6 ...) | candidate · not_run · 2 readset(s) | no-SearchSpace-layer, no-eval |
+| [lithos-metal](../src/inferrsi/data/skill_library/repositories/lithos-metal/SKILL.md) | common | Inspect Lithos Metal compilation and fusion, GPU-resident generation and DSpark state, or local serving and ... | Pinned Lithos Metal 0.1.2 source/API contracts; Apple silicon Metal runtime; model and ... | failure_modes, input_contract, produced_artifact
3 advice | 1 Accelerator · Diagnose / Configure / Optimize / Verify | 1 · Systems sources (lithos-ai/lithos-metal @ `6653b7df5f2c4392ec7b038960f2796b1499923f` ...) | candidate · not_run · 1 readset(s) | no-eval |
+| [local-dream](../src/inferrsi/data/skill_library/repositories/local-dream/SKILL.md) | common | Audit Local Dream Android desired-versus-serving process state, remote exposure, native backend and DiT packaging/ABI ... | Selected Android app/process and native diffusion boundary contracts; common is a ... | failure_modes, input_contract, produced_artifact
3 advice | 2 RAM/Host †guidance: Local Dream Android diffusion source contracts | Local Dream Android diffusion boundary source (xororz/local-dream @ `db162df4d9bb6efc6595250423d8df26241f4c27` ...) | candidate · not_run · 2 readset(s) | no-eval |
+| [mistral-rs](../src/inferrsi/data/skill_library/repositories/mistral-rs/SKILL.md) | common | Investigate mistral.rs Metal paged-attention decode layouts, prefill coordinates and workspace, actual-model prewarm ... | Selected native Metal paged decode/prefill layout/metadata/workspace and ... | failure_modes, input_contract, produced_artifact
3 advice | 1 Accelerator †guidance: mistral.rs Metal paged-attention source contracts
2 RAM/Host †guidance: mistral.rs Metal paged-attention source contracts | mistral.rs Metal paged-attention extraction (EricLBuehler/mistral.rs @ `9459891e7697238225885518119682ac8209a612` ...) | candidate · not_run · 2 readset(s) | no-eval |
+| [mlc-llm](../src/inferrsi/data/skill_library/repositories/mlc-llm/SKILL.md) | common | Investigate MLC LLM compiled Metal artifact/schema, Apple target and pinned TVM module/storage/stream contracts, and ... | Selected artifact/Apple target/pinned TVM Metal consumer and KV configuration source ... | failure_modes, input_contract, produced_artifact
3 advice | 1 Accelerator †guidance: MLC LLM artifact, target and runtime source contracts
2 RAM/Host †guidance: MLC LLM artifact, target and runtime source contracts | MLC LLM compiled Metal and runtime contracts (mlc-ai/mlc-llm @ `8978ea9aec626d64096aaa92fbc89bda2e3ea802`; TVM ...) | candidate · not_run · 2 readset(s) | no-eval |
+| [mlx-swift-lm](../src/inferrsi/data/skill_library/repositories/mlx-swift-lm/SKILL.md) | common | Investigate MLX Swift LM generation ownership, cancellation and settled task completion, and recurrent GatedDelta Metal ... | Selected Swift ownership/cancellation/settling and actual GatedDelta Metal/ops contracts ... | failure_modes, input_contract, produced_artifact
3 advice | 1 Accelerator †guidance: MLX Swift LM ownership and recurrent Metal source contracts
2 RAM/Host †guidance: MLX Swift LM ownership and recurrent Metal source contracts | MLX Swift LM lifecycle and recurrent Metal extraction (ml-explore/mlx-swift-lm @ `3d066d3a0ffdc7e70c476a3d31488ea62dfdd0ac` ...) | candidate · not_run · 2 readset(s) | no-eval |
+| [project-maya](../src/inferrsi/data/skill_library/repositories/project-maya/SKILL.md) | common | Investigate Project Maya model-family dispatch, representation and memory tiers, API versus native state, speculative ... | Model-specific dispatch/residency, API/native state and qualification investigation ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator †guidance: Project Maya model-specific source contracts
2 RAM/Host †guidance: Project Maya model-specific source contracts
3 SSD †guidance: Project Maya model-specific source contracts | Project Maya source and CPU-slice extraction (mw00/project-maya @ `249f9a87d1a9d72bec2b684d704b0aa9296b1c9c` ...) | candidate · partial_cpu_only · 2 readset(s) | no-eval |
+| [splash](../src/inferrsi/data/skill_library/repositories/splash/SKILL.md) | common | Investigate Splash model-format compatibility, automatic memory planning, Metal and ANE execution, DFlash2 ... | Pinned Splash representation, planning, state/serving and verification investigation ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | — | Splash portable AREX extraction (incoai/splash @ `39a97385f8560cf8a93972cec9b0a963c9195250`; AREX ...) | candidate · partial_cpu_only · 2 readset(s) | no-SearchSpace-layer, no-eval |
+| [vllm-metal](../src/inferrsi/data/skill_library/repositories/vllm-metal/SKILL.md) | common | Investigate vllm-metal Apple inference through platform and native-ABI selection, paged KV/offload ownership ... | Selected plugin preparation, KV/offload ownership and speculative/deferred/dispatch ... | failure_modes, input_contract, produced_artifact
3 advice | 1 Accelerator †guidance: vLLM Metal selected source contracts
2 RAM/Host †guidance: vLLM Metal selected source contracts
3 SSD †guidance: vLLM Metal selected source contracts | vLLM Metal portable AREX source and CPU extraction (vllm-project/vllm-metal @ `e7df7a6ff215f99c7842552d230fdbc709835934` ...) | candidate · partial_cpu_only · 2 readset(s) | no-eval |
## Registered: Methods (20)
@@ -221,7 +221,7 @@ an entrypoint routes to procedures below it rather than adding one.
| `no-SearchSpace-layer` | 15: ascend-video-diffusion-optimization, audit-fixed-workload-claim, audit-recurrent-kv-finalization, cite-check-downstream-pr, competitor-scan, draft-acceptance-issue, dwarfstar, husky, inference-benchmark-design, lily, npu-skill-factory, repository-evolution-analysis, skill-evolution-playbook, splash, video-diffusion-optimization |
| `no-Reference-Matrix-row` | 8: ascend-video-diffusion-optimization, audit-fixed-workload-claim, cite-check-downstream-pr, claim-contract, competitor-scan, draft-acceptance-issue, rsi-evidence, video-diffusion-optimization |
| `no-sidecar` | 0: — |
-| `no-applicability` | 41: as listed in the Gaps column above |
+| `no-applicability` | 37: as listed in the Gaps column above |
| `no-eval` | 75: as listed in the Gaps column above |
These are inventory gaps, not defects. A workflow skill may legitimately
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/README.md b/src/inferrsi/data/repo_packs/inferrsi/knowledge/README.md
index e2b0617d4..ee78084b8 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/README.md
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/README.md
@@ -15,7 +15,7 @@ These are knowledge abstracts (`--kind knowledge`), with namespaced package
IDs distinct from explicitly selected `checkout_ref` skills. Ascend/CUDA
procedures use their corresponding backend; shared methods use `common`.
-Source snapshot: [bde70d82bb65cc2e5dcd0f9d33cda5d38091120b](https://github.com/InferMatrixLab/InferRSI/tree/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library).
+Source snapshot: [a04c599c49465b67ea40a1f0de622706ab3da955](https://github.com/InferMatrixLab/InferRSI/tree/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library).
```bash
inferrsi knowledge search 'profiling' --repo-type inferrsi --limit 50
@@ -29,7 +29,7 @@ contain the current input paths and bytes (Git checkout filters are respected).
```bash
python tools/build_skill_index.py --source-revision
-python tools/build_skill_index.py --source-revision bde70d82bb65cc2e5dcd0f9d33cda5d38091120b --check
+python tools/build_skill_index.py --source-revision a04c599c49465b67ea40a1f0de622706ab3da955 --check
```
`--check` detects missing, stale and unexpected generated records without writing.
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-aclgraph-capture-replay.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-aclgraph-capture-replay.json
index d1841646a..8a85b8189 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-aclgraph-capture-replay.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-aclgraph-capture-replay.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/hardware/ascend/ascend-aclgraph-capture-replay/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/hardware/ascend/ascend-aclgraph-capture-replay/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-aclgraph-capture-replay/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-aclgraph-capture-replay/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-device-health.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-device-health.json
index 23c6a195c..ec2ae6a15 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-device-health.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-device-health.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/hardware/ascend/ascend-device-health/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/hardware/ascend/ascend-device-health/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-device-health/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-device-health/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-distributed-parallel.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-distributed-parallel.json
index 4d6cf119e..bc0817b22 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-distributed-parallel.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-distributed-parallel.json
@@ -27,8 +27,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/hardware/ascend/ascend-distributed-parallel/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/hardware/ascend/ascend-distributed-parallel/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-distributed-parallel/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-distributed-parallel/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-graph-mode.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-graph-mode.json
index 3baf8d822..4ceadaf5b 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-graph-mode.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-graph-mode.json
@@ -26,8 +26,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/hardware/ascend/ascend-graph-mode/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/hardware/ascend/ascend-graph-mode/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-graph-mode/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-graph-mode/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-model-adaptation.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-model-adaptation.json
index a4d9f4c2d..25915c60e 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-model-adaptation.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-model-adaptation.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/hardware/ascend/ascend-model-adaptation/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/hardware/ascend/ascend-model-adaptation/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-model-adaptation/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-model-adaptation/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-profiling.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-profiling.json
index 4513309da..cb7621f19 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-profiling.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-profiling.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-profiling/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-profiling/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-profiling/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-profiling/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-tuning.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-tuning.json
index 72287107b..5d8e65c24 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-tuning.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-tuning.json
@@ -24,8 +24,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-tuning/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-tuning/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-tuning/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-tuning/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-optimization-casebook.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-optimization-casebook.json
index 3bc45c815..f84d9b26e 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-optimization-casebook.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-optimization-casebook.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/hardware/ascend/ascend-optimization-casebook/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/hardware/ascend/ascend-optimization-casebook/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-optimization-casebook/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-optimization-casebook/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-performance-analysis.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-performance-analysis.json
index c8689ef9f..025cfae9b 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-performance-analysis.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-performance-analysis.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/hardware/ascend/ascend-performance-analysis/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/hardware/ascend/ascend-performance-analysis/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-performance-analysis/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-performance-analysis/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-video-diffusion-optimization.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-video-diffusion-optimization.json
index 3b0b531c8..b42a3d960 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-video-diffusion-optimization.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-video-diffusion-optimization.json
@@ -26,8 +26,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/hardware/ascend/ascend-video-diffusion-optimization/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/hardware/ascend/ascend-video-diffusion-optimization/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-video-diffusion-optimization/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-video-diffusion-optimization/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascendc-tiling-design.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascendc-tiling-design.json
index 100d6c8d2..9a6139301 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascendc-tiling-design.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascendc-tiling-design.json
@@ -24,8 +24,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/hardware/ascend/ascendc-tiling-design/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/hardware/ascend/ascendc-tiling-design/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascendc-tiling-design/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascendc-tiling-design/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/cann-attribution-overlap.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/cann-attribution-overlap.json
index 78df33c6a..5173819a6 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/cann-attribution-overlap.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/cann-attribution-overlap.json
@@ -24,8 +24,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/hardware/ascend/cann-attribution-overlap/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/hardware/ascend/cann-attribution-overlap/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/cann-attribution-overlap/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/cann-attribution-overlap/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-device-memory-lifecycle.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-device-memory-lifecycle.json
index 3b974ef7f..6402c71c1 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-device-memory-lifecycle.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-device-memory-lifecycle.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/hardware/ascend/torchnpu-device-memory-lifecycle/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/hardware/ascend/torchnpu-device-memory-lifecycle/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/torchnpu-device-memory-lifecycle/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/torchnpu-device-memory-lifecycle/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-pinned-host-memory-lifecycle.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-pinned-host-memory-lifecycle.json
index bb789a376..e3cfdc5ac 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-pinned-host-memory-lifecycle.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-pinned-host-memory-lifecycle.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/hardware/ascend/torchnpu-pinned-host-memory-lifecycle/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/hardware/ascend/torchnpu-pinned-host-memory-lifecycle/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/torchnpu-pinned-host-memory-lifecycle/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/torchnpu-pinned-host-memory-lifecycle/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-serving-preemption.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-serving-preemption.json
index 9bfdef399..31fe88251 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-serving-preemption.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-serving-preemption.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/methods/assess-serving-preemption/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/methods/assess-serving-preemption/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/methods/assess-serving-preemption/SKILL.md",
"src/inferrsi/data/skill_library/methods/assess-serving-preemption/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-split-backward-schedule.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-split-backward-schedule.json
index d944db3aa..139a8a812 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-split-backward-schedule.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-split-backward-schedule.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/methods/assess-split-backward-schedule/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/methods/assess-split-backward-schedule/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/methods/assess-split-backward-schedule/SKILL.md",
"src/inferrsi/data/skill_library/methods/assess-split-backward-schedule/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-fixed-workload-claim.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-fixed-workload-claim.json
index fbb2c1a4c..f4235e6e7 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-fixed-workload-claim.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-fixed-workload-claim.json
@@ -20,8 +20,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/methods/audit-fixed-workload-claim/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/methods/audit-fixed-workload-claim/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/methods/audit-fixed-workload-claim/SKILL.md",
"src/inferrsi/data/skill_library/methods/audit-fixed-workload-claim/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-metal-fasth3-semantics.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-metal-fasth3-semantics.json
index 6e2fcbedc..26a05b4ef 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-metal-fasth3-semantics.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-metal-fasth3-semantics.json
@@ -29,8 +29,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/LICENSE",
"src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/SKILL.md",
@@ -56,7 +56,7 @@
"source_files": {
"src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/LICENSE": "663e51edcd9a3a8f858018ce6924ad2895a24c269c26c39ad75325bcd0f7e5bc",
"src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/SKILL.md": "e5205de6eba52b3d1693666af755640697717e52ee7834e1ae30e4071b208d54",
- "src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/metadata.json": "86a67bb56110c3a9e191e0ff576194162e960cbddca0a99c593ea7aef8df8a4e",
+ "src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/metadata.json": "f1681981db4f9077ee27ff71571dceecd69daf25136bd374b7d9ad3b59762a6a",
"src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/references/THIRD_PARTY_NOTICES.md": "a0d77db7d2aca0eda811ab3c386cc32b3143f2c2609a042cf4642954f0f6fb09",
"src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/references/preflight.md": "5cf745005914c74b18a09f127d1d02ce0cb64f6cd7899501ce3cb40938c42a4c",
"src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/references/repo-provenance.md": "6a4c14207223dcc2584f7133bd14ff4590c2258fc2bd44f3721b6bf63e3e35cf",
@@ -148,7 +148,12 @@
"method": "Bounded fork-specific source audit, bundled selected-declaration helper and assertion-backed cases. Native Apple/Metal/model evidence remains blocked; no numerical/performance/skill-efficacy claim."
},
"applicability": {
- "scope": "h3-metal FastH3 requested/effective configuration, checkpoint transformation and schedule/RoPE semantics, VSA prefix/partial tiles and dispatch/fallback audit. Common catalog placement is discovery, not CPU inference or Apple qualification."
+ "scope": "h3-metal FastH3 requested/effective configuration, checkpoint transformation and schedule/RoPE semantics, VSA prefix/partial tiles and dispatch/fallback audit. Common catalog placement is discovery, not CPU inference or Apple qualification.",
+ "prerequisites": "Access to the fork-specific source and semantic contracts; native checks additionally need their matching Metal/model resources and separate execution authority.",
+ "input_contract": "Source revision/patches, requested parameters and explicitly captured environment controls, checkpoint path/key-family/config identity, requested/aligned frames, latent dimensions, packed prefix segments and sparse-pipeline availability or unknown.",
+ "produced_artifact": "A ledger of requested and normalized parameters, ambient overrides, source invariants, counterexamples, observed/unknown facts and the smallest discriminating next check.",
+ "failure_modes": "Ignoring ambient overrides, mixing checkpoint identity switches, using global rather than per-head Q/K/V concatenation, pooling padded boundary rows, or double-adding coarse attention can invalidate the semantic audit.",
+ "refusal_point": "Stop on conflicting configuration, mixed checkpoint identity, missing resources or the authorized boundary; name the missing native test/fixture and leave it blocked/not_run."
},
"limitations": [
"BLOCKED_REQUIRED_BACKEND: native Apple/Metal, actual FastH3 checkpoint/schedule/sparse parity, audio/video quality, target memory fit, performance and service evidence not_run.",
@@ -163,14 +168,25 @@
"references/source-identity.json",
"references/repo-provenance.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Record requested parameters, normalized values and later environment overrides separately, including requested versus aligned frame counts.",
+ "Audit checkpoint transformations and schedule consumers beyond tensor names/shapes; tiny permutation fixtures can falsify mappings but do not establish model parity.",
+ "Resolve sparse eligibility and the dense-fine-plus-gated-coarse fallback explicitly; preserve unknown pipeline availability."
+ ]
},
"applicability": {
- "scope": "h3-metal FastH3 requested/effective configuration, checkpoint transformation and schedule/RoPE semantics, VSA prefix/partial tiles and dispatch/fallback audit. Common catalog placement is discovery, not CPU inference or Apple qualification."
+ "scope": "h3-metal FastH3 requested/effective configuration, checkpoint transformation and schedule/RoPE semantics, VSA prefix/partial tiles and dispatch/fallback audit. Common catalog placement is discovery, not CPU inference or Apple qualification.",
+ "prerequisites": "Access to the fork-specific source and semantic contracts; native checks additionally need their matching Metal/model resources and separate execution authority.",
+ "input_contract": "Source revision/patches, requested parameters and explicitly captured environment controls, checkpoint path/key-family/config identity, requested/aligned frames, latent dimensions, packed prefix segments and sparse-pipeline availability or unknown.",
+ "produced_artifact": "A ledger of requested and normalized parameters, ambient overrides, source invariants, counterexamples, observed/unknown facts and the smallest discriminating next check.",
+ "failure_modes": "Ignoring ambient overrides, mixing checkpoint identity switches, using global rather than per-head Q/K/V concatenation, pooling padded boundary rows, or double-adding coarse attention can invalidate the semantic audit.",
+ "refusal_point": "Stop on conflicting configuration, mixed checkpoint identity, missing resources or the authorized boundary; name the missing native test/fixture and leave it blocked/not_run."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Record requested parameters, normalized values and later environment overrides separately, including requested versus aligned frame counts.",
+ "Audit checkpoint transformations and schedule consumers beyond tensor names/shapes; tiny permutation fixtures can falsify mappings but do not establish model parity.",
+ "Resolve sparse eligibility and the dense-fine-plus-gated-coarse fallback explicitly; preserve unknown pipeline availability."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-runtime-residency.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-runtime-residency.json
index 6fe9c3504..619b00c34 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-runtime-residency.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-runtime-residency.json
@@ -30,8 +30,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/LICENSE",
"src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/SKILL.md",
@@ -57,7 +57,7 @@
"source_files": {
"src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/LICENSE": "663e51edcd9a3a8f858018ce6924ad2895a24c269c26c39ad75325bcd0f7e5bc",
"src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/SKILL.md": "fdb4d9483c6efa836f9e59617da605fa9c2186e11c2041b05d5c81850ba0475d",
- "src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/metadata.json": "fe626a77af3f084f36fd73ffdb6f67976e7dceeafb94ba22078ef03c3a96cfae",
+ "src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/metadata.json": "952bd2947b68e56d249f10798e028d191a8ffbdb2f123775be66946d3e2ef77e",
"src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/references/THIRD_PARTY_NOTICES.md": "a0d77db7d2aca0eda811ab3c386cc32b3143f2c2609a042cf4642954f0f6fb09",
"src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/references/minimax-h3-apple.md": "57ce01754ba96c378f13a91305ef3d418b926f99b2753f0af99315572bce063a",
"src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/references/minimax-h3-weight-inventory.json": "fbaacaeec340d3ce893556ee864405a0746af4d13530a38cb3d5aa9d60929271",
@@ -177,7 +177,12 @@
"method": "Manual pinned-source distillation informed by the existing Repo-To-Skill ledger; no AREX/DisCo execution or behavioral efficacy claim."
},
"applicability": {
- "scope": "MiniMax-H3 joint audio/video C/Metal runtime audit: raw checkpoint compatibility, input encoders, schedule precompute, phased/cached residency, numerical controls and SSD streaming on Apple Silicon. Common catalog placement is a discovery classification, not cross-backend qualification."
+ "scope": "MiniMax-H3 joint audio/video C/Metal runtime audit: raw checkpoint compatibility, input encoders, schedule precompute, phased/cached residency, numerical controls and SSD streaming on Apple Silicon. Common catalog placement is a discovery classification, not cross-backend qualification.",
+ "prerequisites": "Access to the pinned source architecture and selected component/format inventory; native checks need matching model, fixture and backend resources within an authorized experiment budget.",
+ "input_contract": "Runtime revision/patches, macOS/SDK/GPU and memory budget, FL2VA/Ref2VA selection, component formats, prompt and ordered media, requested/aligned frames, canvases, steps/seed, layer/reuse/token/precision policies and preview/interactive-cache state.",
+ "produced_artifact": "A phase/component/owner ledger with resident weights, runtime/host buffers, overlap, release/reuse boundaries and observations/unknowns, followed by a discriminating check, actual result and remaining gap.",
+ "failure_modes": "Estimating fit from model-file bytes, confusing runtime int8 with incompatible MLX bundles, ignoring preview/cache overlap, or reducing DiT residency to address a VAE/host-output peak can produce a wrong capacity diagnosis.",
+ "refusal_point": "Stop at unavailable model/fixture/backend resources or the accepted experiment budget; missing target evidence remains unverified/INCONCLUSIVE."
},
"limitations": [
"Native h3.c/Metal execution, target peak memory, audio/video quality, performance, service behavior and skill efficacy were not run.",
@@ -191,14 +196,25 @@
"references/source-identity.json",
"references/minimax-h3-weight-inventory.json"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Resolve the actual conditioning and output components before counting residency; do not add alternative checkpoints or count shared files twice.",
+ "Trace one-shot, preview and interactive ownership separately, including AdaLN schedule dependencies and prepared-cache reset boundaries.",
+ "Classify approximation and precision controls before comparison, then measure the relevant phase with tensor, physical/compressed/swap and storage observations."
+ ]
},
"applicability": {
- "scope": "MiniMax-H3 joint audio/video C/Metal runtime audit: raw checkpoint compatibility, input encoders, schedule precompute, phased/cached residency, numerical controls and SSD streaming on Apple Silicon. Common catalog placement is a discovery classification, not cross-backend qualification."
+ "scope": "MiniMax-H3 joint audio/video C/Metal runtime audit: raw checkpoint compatibility, input encoders, schedule precompute, phased/cached residency, numerical controls and SSD streaming on Apple Silicon. Common catalog placement is a discovery classification, not cross-backend qualification.",
+ "prerequisites": "Access to the pinned source architecture and selected component/format inventory; native checks need matching model, fixture and backend resources within an authorized experiment budget.",
+ "input_contract": "Runtime revision/patches, macOS/SDK/GPU and memory budget, FL2VA/Ref2VA selection, component formats, prompt and ordered media, requested/aligned frames, canvases, steps/seed, layer/reuse/token/precision policies and preview/interactive-cache state.",
+ "produced_artifact": "A phase/component/owner ledger with resident weights, runtime/host buffers, overlap, release/reuse boundaries and observations/unknowns, followed by a discriminating check, actual result and remaining gap.",
+ "failure_modes": "Estimating fit from model-file bytes, confusing runtime int8 with incompatible MLX bundles, ignoring preview/cache overlap, or reducing DiT residency to address a VAE/host-output peak can produce a wrong capacity diagnosis.",
+ "refusal_point": "Stop at unavailable model/fixture/backend resources or the accepted experiment budget; missing target evidence remains unverified/INCONCLUSIVE."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Resolve the actual conditioning and output components before counting residency; do not add alternative checkpoints or count shared files twice.",
+ "Trace one-shot, preview and interactive ownership separately, including AdaLN schedule dependencies and prepared-cache reset boundaries.",
+ "Classify approximation and precision controls before comparison, then measure the relevant phase with tensor, physical/compressed/swap and storage observations."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-llama-runtime-state.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-llama-runtime-state.json
index a690d3eb5..e6cbc9046 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-llama-runtime-state.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-llama-runtime-state.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/SKILL.md",
"src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/metadata.json",
@@ -41,7 +41,7 @@
"checkout_ref": "checkout:audit-llama-runtime-state",
"source_files": {
"src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/SKILL.md": "95e81b6bc096a954fa73139ea5bc001d1efa6c99747161da62f012b494e5c013",
- "src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/metadata.json": "1d30b31dc8352fc6a8d7b8abbef8118c840e47e5b90795dae0c758bf9717278c",
+ "src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/metadata.json": "22f1d01b48ea6e2b2ef2deeb20c1404ce8b586eb18442b5adb1390310b4b7503",
"src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/references/source-notes.md": "3d12a401e9decb55a2f4f3b2bfa0d360ef5110f63a0469dfe285654e78054190",
"src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
},
@@ -94,7 +94,12 @@
}
],
"applicability": {
- "scope": "Pinned llama.cpp context-memory, graph-reuse and server task-lifecycle diagnosis; concrete model/backend behavior requires native validation."
+ "scope": "Pinned llama.cpp context-memory, graph-reuse and server task-lifecycle diagnosis; concrete model/backend behavior requires native validation.",
+ "prerequisites": "Pinned llama.cpp source or a concrete runtime trace, access to the relevant memory/graph/server ownership evidence and an identified caller recovery policy that remains in force.",
+ "input_contract": "Source/build and model revisions, memory type, actual backend/device and execution mode, context/batch/ubatch parameters, sequence IDs/positions/output flags and return status; graph cases need dispatch evidence, server cases need correlated HTTP/task/slot/cache logs.",
+ "produced_artifact": "An owner/event ledger of ubatch progress, retained memory, graph-gate decisions, input readiness and slot release/retention, with the first supported broken obligation or smallest missing observation.",
+ "failure_modes": "Assuming whole-batch rollback after partial decode failure, treating token count as a complete graph-reuse key, or equating posted cancellation with released slots can hide retained state or misdiagnose correct deferred work.",
+ "refusal_point": "Stop before reset or repair if revision, memory type, actual backend, sequence ownership or partial progress is unknown; escalate unsupported rollback or unresolved in-flight work."
},
"limitations": [
"Source-derived candidate; native runtime and efficacy validation not run.",
@@ -104,14 +109,25 @@
"provenance_readsets": [
"references/source-notes.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Reconcile per-sequence retained positions with completed ubatches before retrying a failed logical batch.",
+ "Check both graph allow_reuse and input can_reuse gates, then establish copy readiness independently of graph identity.",
+ "Trace cancellation posting, normal task processing, slot release and prompt retention as separate events."
+ ]
},
"applicability": {
- "scope": "Pinned llama.cpp context-memory, graph-reuse and server task-lifecycle diagnosis; concrete model/backend behavior requires native validation."
+ "scope": "Pinned llama.cpp context-memory, graph-reuse and server task-lifecycle diagnosis; concrete model/backend behavior requires native validation.",
+ "prerequisites": "Pinned llama.cpp source or a concrete runtime trace, access to the relevant memory/graph/server ownership evidence and an identified caller recovery policy that remains in force.",
+ "input_contract": "Source/build and model revisions, memory type, actual backend/device and execution mode, context/batch/ubatch parameters, sequence IDs/positions/output flags and return status; graph cases need dispatch evidence, server cases need correlated HTTP/task/slot/cache logs.",
+ "produced_artifact": "An owner/event ledger of ubatch progress, retained memory, graph-gate decisions, input readiness and slot release/retention, with the first supported broken obligation or smallest missing observation.",
+ "failure_modes": "Assuming whole-batch rollback after partial decode failure, treating token count as a complete graph-reuse key, or equating posted cancellation with released slots can hide retained state or misdiagnose correct deferred work.",
+ "refusal_point": "Stop before reset or repair if revision, memory type, actual backend, sequence ownership or partial progress is unknown; escalate unsupported rollback or unresolved in-flight work."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Reconcile per-sequence retained positions with completed ubatches before retrying a failed logical batch.",
+ "Check both graph allow_reuse and input can_reuse gates, then establish copy readiness independently of graph identity.",
+ "Trace cancellation posting, normal task processing, slot release and prompt retention as separate events."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-recurrent-kv-finalization.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-recurrent-kv-finalization.json
index 8fd55bb5f..bf0d5a267 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-recurrent-kv-finalization.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-recurrent-kv-finalization.json
@@ -26,8 +26,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/LICENSE",
"src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/SKILL.md",
@@ -48,7 +48,7 @@
"source_files": {
"src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/LICENSE": "cfc7749b96f63bd31c3c42b5c471bf756814053e847c10f3eb003417bc523d30",
"src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/SKILL.md": "a468d75f25e77b96e6fb5376692f7a89c0464cc00d4df035f1dc6cc923f713cc",
- "src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/metadata.json": "db0b4593d7a6fbb898fdf11a1710b529abbade847ea57319843669b375b68fc8",
+ "src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/metadata.json": "dfd605f8f31740565063fad0785dea3e7751b8deea4d43e1a66ca30d92ae901f",
"src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/references/NOTICE.md": "cca794656897ecdc37f983f7f23c4cd12306dbe3c9cd2f54fece6340753ae4cf",
"src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/references/source-identity.json": "5e52db0a0bda4673b9aac46a3fc035b824780def72921fee6796edb2ed9fca65",
"src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/references/source-notes.md": "0956f59cd9a1fac27da53228dd60c51a07b87639a43c09bfab296c4aeeceaab9",
@@ -112,7 +112,12 @@
}
],
"applicability": {
- "scope": "Pinned vLLM-RLT native Ouro synchronous scheduling with host KVMetadata: LAST_EXITED propagation versus SHARED, later depth-specific Torch-reference attention and normal release/wait/free order. No speculation, connector, preemption, prefix sharing, split/packed prefill or graph/static-buffer path qualification."
+ "scope": "Pinned vLLM-RLT native Ouro synchronous scheduling with host KVMetadata: LAST_EXITED propagation versus SHARED, later depth-specific Torch-reference attention and normal release/wait/free order. No speculation, connector, preemption, prefix sharing, split/packed prefill or graph/static-buffer path qualification.",
+ "prerequisites": "The selected native Ouro synchronous path uses host KVMetadata, no speculation/connector/preemption, atomic prefill and disabled graph/static-buffer/packed-prefill paths; changed execution conditions need a new bounded check.",
+ "input_contract": "Repository revision, model/configuration, recurrent-layer set, loop count, cache layout and block size, dtype/device, attention backend and effective flags, plus the traced token/depth/layer and its next consumer.",
+ "produced_artifact": "A bounded token/finalization/consumer/retirement trace separating source call order, observed data and written validity, completion evidence and remaining native model/backend/performance gaps.",
+ "failure_modes": "Substituting mark_finalized for K/V propagation, accepting allocated pages as a complete written prefix, applying LAST_EXITED copy expectations to SHARED, or treating event recording as completion can invalidate the ownership diagnosis.",
+ "refusal_point": "On a failed prerequisite or complete-prefix check, retain the first failing token/depth/layer and trace the missing operation. Stop the conclusion when flags or metadata ownership leave the selected path."
},
"limitations": [
"Source-derived candidate; native CPU/CUDA, model correctness/quality, performance and agent efficacy are not validated.",
@@ -123,14 +128,25 @@
"references/source-notes.md",
"references/source-identity.json"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Distinguish copied K/V values, logical written positions and allocated capacity; follow the later depth-specific attention consumer.",
+ "Preserve executed shallower values during LAST_EXITED propagation and use SHARED as a separate one-plane control.",
+ "Follow the latest request event through runner release before scheduler cache release; page reuse is not evidence of erased old bytes."
+ ]
},
"applicability": {
- "scope": "Pinned vLLM-RLT native Ouro synchronous scheduling with host KVMetadata: LAST_EXITED propagation versus SHARED, later depth-specific Torch-reference attention and normal release/wait/free order. No speculation, connector, preemption, prefix sharing, split/packed prefill or graph/static-buffer path qualification."
+ "scope": "Pinned vLLM-RLT native Ouro synchronous scheduling with host KVMetadata: LAST_EXITED propagation versus SHARED, later depth-specific Torch-reference attention and normal release/wait/free order. No speculation, connector, preemption, prefix sharing, split/packed prefill or graph/static-buffer path qualification.",
+ "prerequisites": "The selected native Ouro synchronous path uses host KVMetadata, no speculation/connector/preemption, atomic prefill and disabled graph/static-buffer/packed-prefill paths; changed execution conditions need a new bounded check.",
+ "input_contract": "Repository revision, model/configuration, recurrent-layer set, loop count, cache layout and block size, dtype/device, attention backend and effective flags, plus the traced token/depth/layer and its next consumer.",
+ "produced_artifact": "A bounded token/finalization/consumer/retirement trace separating source call order, observed data and written validity, completion evidence and remaining native model/backend/performance gaps.",
+ "failure_modes": "Substituting mark_finalized for K/V propagation, accepting allocated pages as a complete written prefix, applying LAST_EXITED copy expectations to SHARED, or treating event recording as completion can invalidate the ownership diagnosis.",
+ "refusal_point": "On a failed prerequisite or complete-prefix check, retain the first failing token/depth/layer and trace the missing operation. Stop the conclusion when flags or metadata ownership leave the selected path."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Distinguish copied K/V values, logical written positions and allocated capacity; follow the later depth-specific attention consumer.",
+ "Preserve executed shallower values during LAST_EXITED propagation and use SHARED as a separate one-plane control.",
+ "Follow the latest request event through runner release before scheduler cache release; page reuse is not evidence of erased old bytes."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sd-cpp-residency.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sd-cpp-residency.json
index 0821a7d0e..7b31e3fad 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sd-cpp-residency.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sd-cpp-residency.json
@@ -25,8 +25,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/LICENSE",
"src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/SKILL.md",
@@ -49,7 +49,7 @@
"source_files": {
"src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/LICENSE": "b53fa08f515cb5a6fff7b9fd8fcd0961a4b80df29d74372d25e1f9171aa042ee",
"src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/SKILL.md": "4c04d39812fb566e3ecb5807c170d5c32848eb2b93a1f0fedcf66ba9a9c92860",
- "src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/metadata.json": "13ab6d874c4e3135be80ef7c3cb6bfd28fb9121012e0536831ec3705ccb8c5eb",
+ "src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/metadata.json": "412c649b16526038ee0299f8b2a610bb7f120eaab25cd83575809fe03b604b2e",
"src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/references/loader-storage-boundaries.md": "c1ec047aa37fca6b374f9bfd750e8f30d45af1d48fe728db23873c0b84ad94e8",
"src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/references/phase-residency.md": "7b537cb972347dd68e303960f142892346d8b19b884e4d08016ea36a23bdaa7a",
"src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/references/source-identity.json": "af4830370b6eddc2d176a9ae14da4ab56a90b0fe048a01a89ee051df4365d8e9",
@@ -117,7 +117,12 @@
}
],
"applicability": {
- "scope": "Source-grounded audit of native image-generation weight/runtime residency, optional segment prefetch and bounded non-ZIP safetensors catalog/read/staging ownership; confirm the actual source, GGML, backend, model and executing path independently."
+ "scope": "Source-grounded audit of native image-generation weight/runtime residency, optional segment prefetch and bounded non-ZIP safetensors catalog/read/staging ownership; confirm the actual source, GGML, backend, model and executing path independently.",
+ "prerequisites": "Access to the pinned source/storage boundaries and evidence of the actual image runner/backend dispatch; native capacity or prefetch checks require matching GGML, model and backend resources.",
+ "input_contract": "Runtime/GGML revisions and selection, compiled/loaded backends and devices, model/components and numerical image settings, effective memory/prefetch/placement settings and actual dispatch; retain available phase allocations, cache/workspace, transfer events and completed-image counts.",
+ "produced_artifact": "A phase-by-runner/backend/device ownership table for stored and resident weights, retained runtime, pending allocations, pin/prefetch ownership and release/eviction, with one discriminating check and its actual status.",
+ "failure_modes": "Treating max_vram as proof of segmentation, unpin as physical freeing, or configured prefetch as completed useful overlap can misidentify the capacity cause; missing observations cannot be read as zero.",
+ "refusal_point": "Stop when the cause is resolved or matching GGML/model/backend resources are unavailable; invoking this audit does not authorize dependency installation, model downloads, policy changes or device provisioning."
},
"limitations": [
"Source-derived candidate; native inference, allocator behavior, image quality and performance were not run.",
@@ -132,14 +137,25 @@
"references/source-notes.md",
"references/loader-storage-boundaries.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Separate parameter storage, compute-resident weights, retained workspace/output caches and pending allocations for each runner and phase.",
+ "Verify the actual monolithic, segmented or multi-device path before interpreting a memory budget or prefetch setting.",
+ "For a policy comparison, hold model, inputs, numerical settings and device fixed; check completed images and quality before latency, and separate startup, first and warm requests."
+ ]
},
"applicability": {
- "scope": "Source-grounded audit of native image-generation weight/runtime residency, optional segment prefetch and bounded non-ZIP safetensors catalog/read/staging ownership; confirm the actual source, GGML, backend, model and executing path independently."
+ "scope": "Source-grounded audit of native image-generation weight/runtime residency, optional segment prefetch and bounded non-ZIP safetensors catalog/read/staging ownership; confirm the actual source, GGML, backend, model and executing path independently.",
+ "prerequisites": "Access to the pinned source/storage boundaries and evidence of the actual image runner/backend dispatch; native capacity or prefetch checks require matching GGML, model and backend resources.",
+ "input_contract": "Runtime/GGML revisions and selection, compiled/loaded backends and devices, model/components and numerical image settings, effective memory/prefetch/placement settings and actual dispatch; retain available phase allocations, cache/workspace, transfer events and completed-image counts.",
+ "produced_artifact": "A phase-by-runner/backend/device ownership table for stored and resident weights, retained runtime, pending allocations, pin/prefetch ownership and release/eviction, with one discriminating check and its actual status.",
+ "failure_modes": "Treating max_vram as proof of segmentation, unpin as physical freeing, or configured prefetch as completed useful overlap can misidentify the capacity cause; missing observations cannot be read as zero.",
+ "refusal_point": "Stop when the cause is resolved or matching GGML/model/backend resources are unavailable; invoking this audit does not authorize dependency installation, model downloads, policy changes or device provisioning."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Separate parameter storage, compute-resident weights, retained workspace/output caches and pending allocations for each runner and phase.",
+ "Verify the actual monolithic, segmented or multi-device path before interpreting a memory budget or prefetch setting.",
+ "For a policy comparison, hold model, inputs, numerical settings and device fixed; check completed images and quality before latency, and separate startup, first and warm requests."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sglang-diffusion-lifetimes.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sglang-diffusion-lifetimes.json
index cd18a9077..58d37a273 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sglang-diffusion-lifetimes.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sglang-diffusion-lifetimes.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/LICENSE",
"src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/SKILL.md",
@@ -43,7 +43,7 @@
"source_files": {
"src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/LICENSE": "1495e1e757ef4d0925a2350563cf5754bb23c51701a8ec4fb3c5cdcbedae6747",
"src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/SKILL.md": "b0a4ba4751087126143ef86983798deb0ed00e24b6ff06bf6ef76a9df6e274f2",
- "src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/metadata.json": "6ed57dfb4f25ecb34b81f2188b055e05d172ba2581e6a3b605b814b2f733a4d6",
+ "src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/metadata.json": "e2cb137978960d29b3644967a29dab930c768d3453bce8aff2aad96fd90a56d9",
"src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/references/source-notes.md": "d0fcfd1177ae2f351041c5b2abba7f5121cec133006d9e4dca8f74ddb567332d",
"src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
},
@@ -101,7 +101,12 @@
}
],
"applicability": {
- "scope": "Bounded monolithic multimodal_gen composed-pipeline state and lifecycle diagnosis; actual backend/model behavior requires native validation."
+ "scope": "Bounded monolithic multimodal_gen composed-pipeline state and lifecycle diagnosis; actual backend/model behavior requires native validation.",
+ "prerequisites": "Pinned source or a concrete trace of the selected composed-pipeline executor and installed residency manager, with the applicable monolithic route identified.",
+ "input_contract": "Source/build, backend and executor/pipeline class, model revision, modality/shape, steps/scheduler/seed/output count, grouped/serial route, ranks, placement, warmup/save settings and symptom; timeout cases need correlated request/job and client/scheduler/worker events.",
+ "produced_artifact": "A version-bound owner/event ledger, first unresolved boundary, causal hypothesis, smallest real discriminating check, exact results and missing dependencies.",
+ "failure_modes": "Assuming all executors share the SyncExecutor contract, treating closed component use as eviction, assuming grouped schedulers are isolated, or reading socket close/job deletion as stopped device compute can misdiagnose ownership.",
+ "refusal_point": "Stop performance claims when the executing path, cleanup or output quality is unresolved; retain the failure and a known baseline rather than inferring a blanket leak or safety conclusion."
},
"limitations": [
"Source-derived candidate; upstream runtime and efficacy validation not run.",
@@ -111,14 +116,25 @@
"provenance_readsets": [
"references/source-notes.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Trace the actual executor and separate request, stage, component-use, module, scheduler, finalizer and allocator-cache lifetimes.",
+ "Check scheduler aliases and declared output-copy policies before introducing overlap; compare real serial/grouped outputs with fixed timesteps and seeds.",
+ "Correlate worker, finalizer and device completion after client timeout or deletion; preserve an unknown execution outcome until that evidence exists."
+ ]
},
"applicability": {
- "scope": "Bounded monolithic multimodal_gen composed-pipeline state and lifecycle diagnosis; actual backend/model behavior requires native validation."
+ "scope": "Bounded monolithic multimodal_gen composed-pipeline state and lifecycle diagnosis; actual backend/model behavior requires native validation.",
+ "prerequisites": "Pinned source or a concrete trace of the selected composed-pipeline executor and installed residency manager, with the applicable monolithic route identified.",
+ "input_contract": "Source/build, backend and executor/pipeline class, model revision, modality/shape, steps/scheduler/seed/output count, grouped/serial route, ranks, placement, warmup/save settings and symptom; timeout cases need correlated request/job and client/scheduler/worker events.",
+ "produced_artifact": "A version-bound owner/event ledger, first unresolved boundary, causal hypothesis, smallest real discriminating check, exact results and missing dependencies.",
+ "failure_modes": "Assuming all executors share the SyncExecutor contract, treating closed component use as eviction, assuming grouped schedulers are isolated, or reading socket close/job deletion as stopped device compute can misdiagnose ownership.",
+ "refusal_point": "Stop performance claims when the executing path, cleanup or output quality is unresolved; retain the failure and a known baseline rather than inferring a blanket leak or safety conclusion."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Trace the actual executor and separate request, stage, component-use, module, scheduler, finalizer and allocator-cache lifetimes.",
+ "Check scheduler aliases and declared output-copy policies before introducing overlap; compare real serial/grouped outputs with fixed timesteps and seeds.",
+ "Correlate worker, finalizer and device completion after client timeout or deletion; preserve an unknown execution outcome until that evidence exists."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-vllm-kv-lifetimes.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-vllm-kv-lifetimes.json
index f8959efe9..86266f045 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-vllm-kv-lifetimes.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-vllm-kv-lifetimes.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/SKILL.md",
"src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/metadata.json",
@@ -39,7 +39,7 @@
"checkout_ref": "checkout:audit-vllm-kv-lifetimes",
"source_files": {
"src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/SKILL.md": "4c3f27459df239f9f4fc971c559d4c9ddfbfebd70507a8a93be8eae59a30b69d",
- "src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/metadata.json": "b43fc3a7ccae59c664fcdf9ff77962edcc07c2a3df8b59324db551642bd23e23",
+ "src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/metadata.json": "8ded6e60c562131616dc8cdcf0d7efc04aef1884885da0b30fb16f42d7433b7c",
"src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/references/source-notes.md": "3665b034594fc8cc767c408002f4e16f8803369242405027c7f37ce6c9ecebe9",
"src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
},
@@ -78,12 +78,33 @@
"provenance_readsets": [
"references/source-notes.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "applicability": {
+ "scope": "Pinned vLLM V1 scheduler/connector ownership audit for asynchronous KV load/save completion, retained blocks, preemption and failed receives; concrete connector conformance is not established by the interface.",
+ "prerequisites": "A concrete connector and worker/model-runner call trace at identified revisions, with request/block events sufficient to compare actual callers against the pinned scheduler/interface contracts.",
+ "input_contract": "Engine/connector revision and class, producer/consumer roles, KV-group count/block namespace, failure policy, concurrent-batch and graph modes; correlate bounded request IDs with allocation, matched tokens, load/save events, receive failures, preemption, finish and pool return.",
+ "produced_artifact": "An ownership/event ledger and bounded contract-conformance report naming the earliest evidenced broken obligation, owner, affected requests/blocks and smallest missing observation, with a proposed targeted regression check.",
+ "failure_modes": "Equating reserved tokens with usable KV, finished responses with reusable blocks, or send completion with scheduler write-fence completion can hide unsafe reuse; missing failed-receive completion can leave waits unreleased.",
+ "refusal_point": "Escalate absent concrete caller evidence, ambiguous block namespaces, unsupported layouts or revision drift; stop before inferring a backend fix, freeing retained blocks or claiming a reproduced bug."
+ },
+ "recommendations": [
+ "Trace lookup, allocation, worker execution and completion propagation before changing request state; repeated or undecided lookup and ordinary admission waits can be valid.",
+ "Check delayed-free ownership and scheduler in-flight-write fences separately, including empty-step save/metadata ordering in the real worker.",
+ "Keep receive failure and finished_recving correlated, and use request-level failure evidence for multi-group layouts while preserving the existing recovery policy."
+ ]
+ },
+ "applicability": {
+ "scope": "Pinned vLLM V1 scheduler/connector ownership audit for asynchronous KV load/save completion, retained blocks, preemption and failed receives; concrete connector conformance is not established by the interface.",
+ "prerequisites": "A concrete connector and worker/model-runner call trace at identified revisions, with request/block events sufficient to compare actual callers against the pinned scheduler/interface contracts.",
+ "input_contract": "Engine/connector revision and class, producer/consumer roles, KV-group count/block namespace, failure policy, concurrent-batch and graph modes; correlate bounded request IDs with allocation, matched tokens, load/save events, receive failures, preemption, finish and pool return.",
+ "produced_artifact": "An ownership/event ledger and bounded contract-conformance report naming the earliest evidenced broken obligation, owner, affected requests/blocks and smallest missing observation, with a proposed targeted regression check.",
+ "failure_modes": "Equating reserved tokens with usable KV, finished responses with reusable blocks, or send completion with scheduler write-fence completion can hide unsafe reuse; missing failed-receive completion can leave waits unreleased.",
+ "refusal_point": "Escalate absent concrete caller evidence, ambiguous block namespaces, unsupported layouts or revision drift; stop before inferring a backend fix, freeing retained blocks or claiming a reproduced bug."
},
- "applicability": {},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Trace lookup, allocation, worker execution and completion propagation before changing request state; repeated or undecided lookup and ordinary admission waits can be valid.",
+ "Check delayed-free ownership and scheduler in-flight-write fences separately, including empty-step save/metadata ordering in the real worker.",
+ "Keep receive failure and finished_recving correlated, and use request-level failure evidence for multi-group layouts while preserving the existing recovery policy."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/bound-windowed-attention-replay.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/bound-windowed-attention-replay.json
index 7ca1325d7..1e2bf9c3b 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/bound-windowed-attention-replay.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/bound-windowed-attention-replay.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/SKILL.md",
"src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/metadata.json",
@@ -42,7 +42,7 @@
"checkout_ref": "checkout:bound-windowed-attention-replay",
"source_files": {
"src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/SKILL.md": "4703beb6e5194ef2ff2c0b5ec69f64e35417a4bc0dfb50a10c80a920b77ef3b5",
- "src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/metadata.json": "014be24d0e669f4acbc0717dc323eec784026cf185186619d5a19d6d84496131",
+ "src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/metadata.json": "c68254b1a96a8834ae46c7a1d9035856fa4783315b219aa95e31461bd8ac58c9",
"src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/references/source-notes.md": "fd917bd5ac80bee9c9f4525f8e0300f6d67dc144e2a16632af42a67f3d53d098",
"src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/references/validation.md": "60e71e70c5519bcdf6c171eec2be2f342e8801bedd746d6005f11adc646dfaff",
"src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
@@ -90,7 +90,11 @@
"applicability": {
"scope": "Decoders with an explicit producer/consumer boundary and a demonstrated dependency closure for compact query rows, including separately preserved historical state",
"prerequisites": "A derived dependency closure; a full-length KV write at the boundary; known rollback depth; control over metadata construction",
- "hardware": "Backend-independent retain arithmetic; window sizes, boundary index and ring lengths come from the model config and do not transfer"
+ "hardware": "Backend-independent retain arithmetic; window sizes, boundary index and ring lengths come from the model config and do not transfer",
+ "input_contract": "The actual producer/consumer split and transitive dependency closure, boundary index, window and rollback depth, consumer KV sharing/compression layout and separately preserved historical keys.",
+ "produced_artifact": "A dependency listing, enforced retain length, compact query/position/store-slot/visibility metadata, a separate historical-key retention statement and a diagnostic equivalence-reference mode.",
+ "failure_modes": "Applying window-plus-rollback without dependency closure, discarding full-layout state, clipping away the sampling row, or comparing across preemption can invalidate replay; compact queries do not imply historical keys were dropped.",
+ "refusal_point": "Stop if required consumer state is unavailable without a preserved full-length copy, the boundary representation cannot be captured, or comparison preemption moves the anchor. Do not use for clipped-region outputs or required bit-identical preemption replay."
},
"limitations": [
"Source-derived candidate; no equivalence run was performed.",
@@ -104,16 +108,26 @@
"references/source-notes.md",
"references/validation.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Derive every consumer dependency before computing keep; preserve the required full-layout representation separately from the compact query rows.",
+ "Use the visibility-only diagnostic mode to isolate equivalence from row gathering, and retain one sampling row for a fully clipped chunk.",
+ "Check metadata at chunk/store boundaries, actual prompt-logprob handling and zero preemption during equivalence comparison."
+ ]
},
"applicability": {
"scope": "Decoders with an explicit producer/consumer boundary and a demonstrated dependency closure for compact query rows, including separately preserved historical state",
"prerequisites": "A derived dependency closure; a full-length KV write at the boundary; known rollback depth; control over metadata construction",
- "hardware": "Backend-independent retain arithmetic; window sizes, boundary index and ring lengths come from the model config and do not transfer"
+ "hardware": "Backend-independent retain arithmetic; window sizes, boundary index and ring lengths come from the model config and do not transfer",
+ "input_contract": "The actual producer/consumer split and transitive dependency closure, boundary index, window and rollback depth, consumer KV sharing/compression layout and separately preserved historical keys.",
+ "produced_artifact": "A dependency listing, enforced retain length, compact query/position/store-slot/visibility metadata, a separate historical-key retention statement and a diagnostic equivalence-reference mode.",
+ "failure_modes": "Applying window-plus-rollback without dependency closure, discarding full-layout state, clipping away the sampling row, or comparing across preemption can invalidate replay; compact queries do not imply historical keys were dropped.",
+ "refusal_point": "Stop if required consumer state is unavailable without a preserved full-length copy, the boundary representation cannot be captured, or comparison preemption moves the anchor. Do not use for clipped-region outputs or required bit-identical preemption replay."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Derive every consumer dependency before computing keep; preserve the required full-layout representation separately from the compact query rows.",
+ "Use the visibility-only diagnostic mode to isolate equivalence from row gathering, and retain one sampling row for a fully clipped chunk.",
+ "Check metadata at chunk/store boundaries, actual prompt-logprob handling and zero preemption during equivalence comparison."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/budget-speculative-verification.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/budget-speculative-verification.json
index 6cca85b49..f50077cc6 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/budget-speculative-verification.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/budget-speculative-verification.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/SKILL.md",
"src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/metadata.json",
@@ -42,7 +42,7 @@
"checkout_ref": "checkout:budget-speculative-verification",
"source_files": {
"src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/SKILL.md": "f157562d1edc905cb4fc07246567e3c5b844efb9fbf5fe39f0d40553a1b1a307",
- "src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/metadata.json": "4d68bf4c1767eaee12ade963bbd551d0e61d1bcd81a984bb02fd3aae46965919",
+ "src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/metadata.json": "1aa80a3109072c1ec211f38dbeb9e9cb9edb56aeac187946b43e65dd7502e46f",
"src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/references/source-notes.md": "6fa955022f16cd598f44fd6521901d1cb868b06cec545cabe2b82769a641b5d6",
"src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/references/validation.md": "673d93f6700052974b340fab25b3438e0e6344960569ac9abec7d7720f662288",
"src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
@@ -88,7 +88,11 @@
"applicability": {
"scope": "Engines with multi-token speculative decoding whose verification forward cost grows with query rows",
"prerequisites": "A calibrated per-position acceptance signal; a measured (rows, step_time) curve; a handle on metadata construction",
- "hardware": "Backend-independent procedure; the reference implementation targets NVIDIA SM100 and Ascend 950 only"
+ "hardware": "Backend-independent procedure; the reference implementation targets NVIDIA SM100 and Ascend 950 only",
+ "input_contract": "Draft width and row representation, the defined individual/conditional versus cumulative acceptance semantics, verify rows/time curve, sampler cost and acceptance round actually available at schedule time.",
+ "produced_artifact": "A per-step verification row budget, per-request query lengths from one through draft_width plus one, cumulative acceptance vectors and predicted-versus-realized acceptance error logs.",
+ "failure_modes": "Confusing verification savings with reduced drafting, misreading confidence semantics or available rounds, allowing later-position confidence to select an earlier token, or generating non-prefix grants can invalidate the budget or sampling claim.",
+ "refusal_point": "Stop if unthrottled drafting or sampler cost dominates, the signal cannot be made cumulative, or no round supplies acceptance estimates; the procedure also does not apply without a per-step cost model."
},
"limitations": [
"Source-derived candidate; no local efficacy validation and no reproduced gain.",
@@ -103,16 +107,26 @@
"references/source-notes.md",
"references/validation.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Measure draft, verify and sampler costs separately and fit the verify curve for graph sizes that will actually replay.",
+ "Preserve the one-row floor and prefix grants, and audit CPU/GPU allocation timing against the per-position decision-independence contract.",
+ "Report verification rows, accepted tokens, total step time and throughput separately; use an independently specified sampling oracle rather than requiring unchanged accepted length."
+ ]
},
"applicability": {
"scope": "Engines with multi-token speculative decoding whose verification forward cost grows with query rows",
"prerequisites": "A calibrated per-position acceptance signal; a measured (rows, step_time) curve; a handle on metadata construction",
- "hardware": "Backend-independent procedure; the reference implementation targets NVIDIA SM100 and Ascend 950 only"
+ "hardware": "Backend-independent procedure; the reference implementation targets NVIDIA SM100 and Ascend 950 only",
+ "input_contract": "Draft width and row representation, the defined individual/conditional versus cumulative acceptance semantics, verify rows/time curve, sampler cost and acceptance round actually available at schedule time.",
+ "produced_artifact": "A per-step verification row budget, per-request query lengths from one through draft_width plus one, cumulative acceptance vectors and predicted-versus-realized acceptance error logs.",
+ "failure_modes": "Confusing verification savings with reduced drafting, misreading confidence semantics or available rounds, allowing later-position confidence to select an earlier token, or generating non-prefix grants can invalidate the budget or sampling claim.",
+ "refusal_point": "Stop if unthrottled drafting or sampler cost dominates, the signal cannot be made cumulative, or no round supplies acceptance estimates; the procedure also does not apply without a per-step cost model."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Measure draft, verify and sampler costs separately and fit the verify curve for graph sizes that will actually replay.",
+ "Preserve the one-row floor and prefix grants, and audit CPU/GPU allocation timing against the per-position decision-independence contract.",
+ "Report verification rows, accepted tokens, total step time and throughput separately; use an independently specified sampling oracle rather than requiring unchanged accepted length."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/calibrate-serving-simulation.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/calibrate-serving-simulation.json
index 1caf6afaf..a51ea8070 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/calibrate-serving-simulation.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/calibrate-serving-simulation.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/methods/calibrate-serving-simulation/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/methods/calibrate-serving-simulation/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/methods/calibrate-serving-simulation/SKILL.md",
"src/inferrsi/data/skill_library/methods/calibrate-serving-simulation/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cgroup-resource-controls.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cgroup-resource-controls.json
index 1429dd4f6..3335f9067 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cgroup-resource-controls.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cgroup-resource-controls.json
@@ -25,8 +25,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/hardware/host/cgroup-resource-controls/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/hardware/host/cgroup-resource-controls/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/hardware/host/cgroup-resource-controls/SKILL.md",
"src/inferrsi/data/skill_library/hardware/host/cgroup-resource-controls/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/checkpoint-recovery-consistency.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/checkpoint-recovery-consistency.json
index ce7c6e5c7..76759c86b 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/checkpoint-recovery-consistency.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/checkpoint-recovery-consistency.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/methods/checkpoint-recovery-consistency/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/methods/checkpoint-recovery-consistency/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/methods/checkpoint-recovery-consistency/SKILL.md",
"src/inferrsi/data/skill_library/methods/checkpoint-recovery-consistency/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/choose-parallel-training-plan.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/choose-parallel-training-plan.json
index e94c12016..6f9faab64 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/choose-parallel-training-plan.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/choose-parallel-training-plan.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/methods/choose-parallel-training-plan/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/methods/choose-parallel-training-plan/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/methods/choose-parallel-training-plan/SKILL.md",
"src/inferrsi/data/skill_library/methods/choose-parallel-training-plan/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cite-check-downstream-pr.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cite-check-downstream-pr.json
index f37b7765e..a17da3bd9 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cite-check-downstream-pr.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cite-check-downstream-pr.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/workflows/cite-check-downstream-pr/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/workflows/cite-check-downstream-pr/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/skill-index.json",
"src/inferrsi/data/skill_library/workflows/cite-check-downstream-pr/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/claim-contract.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/claim-contract.json
index f39ac4b29..376d7a0b0 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/claim-contract.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/claim-contract.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/workflows/claim-contract/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/workflows/claim-contract/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/skill-index.json",
"src/inferrsi/data/skill_library/workflows/claim-contract/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/comfy-ui.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/comfy-ui.json
index e6475bf57..39f0f33d0 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/comfy-ui.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/comfy-ui.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/repositories/comfy-ui/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/repositories/comfy-ui/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/repositories/comfy-ui/LICENSE",
"src/inferrsi/data/skill_library/repositories/comfy-ui/SKILL.md",
@@ -45,7 +45,7 @@
"source_files": {
"src/inferrsi/data/skill_library/repositories/comfy-ui/LICENSE": "3972dc9744f6499f0f9b2dbf76696f2ae7ad8af9b23dde66d6af86c9dfb36986",
"src/inferrsi/data/skill_library/repositories/comfy-ui/SKILL.md": "c993bd6ff8e9b5f82f39f175ec8242341136e4ee0808af414e1d94ad3e196c5e",
- "src/inferrsi/data/skill_library/repositories/comfy-ui/metadata.json": "74742851a002a51c208362cdcd2a59fdab5d1734d8492b1f93fbec3cf21e0841",
+ "src/inferrsi/data/skill_library/repositories/comfy-ui/metadata.json": "81ad1c73b12426e4bf7961889f4b9a1c92739c6be18380402f7866e021cac67b",
"src/inferrsi/data/skill_library/repositories/comfy-ui/references/graph-cache-invalidation.md": "614c9a4e6943d24d7d7be7d632682ce8b1dcc239124a37d9dc1d487113772eac",
"src/inferrsi/data/skill_library/repositories/comfy-ui/references/model-residency.md": "014c68386e7cdfdb7b4f566836e8b2ff61bd5a5a506e7b4723f83fe80c5d4de7",
"src/inferrsi/data/skill_library/repositories/comfy-ui/references/repo-provenance.md": "ca996ef515af42a3ce304747fa3f12a31c00533deaace5ba2ce21aefa0a62e2e",
@@ -104,7 +104,12 @@
}
],
"applicability": {
- "scope": "Source-grounded ComfyUI workflow-output cache and model-weight ownership audit; confirm the actual revision, node registry, cache caller and static/dynamic patcher before transfer."
+ "scope": "Source-grounded ComfyUI workflow-output cache and model-weight ownership audit; confirm the actual revision, node registry, cache caller and static/dynamic patcher before transfer.",
+ "prerequisites": "Access to the pinned provenance and the actual node registry/cache caller; residency questions additionally require the concrete patcher and effective dynamic/static mode. Native memory conclusions need the actual backend.",
+ "input_contract": "Exact revision, API-format prompt/output targets, relevant node classes/constants/links/hidden inputs and external state, selected cache caller/mode and patcher settings, plus available execution/load/memory/timing observations.",
+ "produced_artifact": "A state/owner/key/lifetime/consumer ledger with invalidation or release events, evidence and unresolved boundaries, followed by a discriminating check and its actual result.",
+ "failure_modes": "Treating identical JSON as proof of unchanged external inputs, confusing eviction with key validity, or equating workflow cache hits with device-resident model weights can explain the wrong state contract.",
+ "refusal_point": "Stop when the cause/action is supported or required environment, backend or model resources are unavailable; do not manufacture a hit, hardware result or performance threshold."
},
"limitations": [
"Source-derived candidate; no local executor-cache, device-residency or performance qualification.",
@@ -114,14 +119,25 @@
"provenance_readsets": [
"references/repo-provenance.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Draw the affected dependency cone with declared, lazy and dynamically expanded links, then mark unfingerprinted external state unresolved.",
+ "Separate output/object cache validity and retention from patch identity, partial unload, full detach and allocator-visible release.",
+ "Change one test variable while fixing prompt, output target, model, backend, precision, seed and cache mode; check values and executed nodes before timing."
+ ]
},
"applicability": {
- "scope": "Source-grounded ComfyUI workflow-output cache and model-weight ownership audit; confirm the actual revision, node registry, cache caller and static/dynamic patcher before transfer."
+ "scope": "Source-grounded ComfyUI workflow-output cache and model-weight ownership audit; confirm the actual revision, node registry, cache caller and static/dynamic patcher before transfer.",
+ "prerequisites": "Access to the pinned provenance and the actual node registry/cache caller; residency questions additionally require the concrete patcher and effective dynamic/static mode. Native memory conclusions need the actual backend.",
+ "input_contract": "Exact revision, API-format prompt/output targets, relevant node classes/constants/links/hidden inputs and external state, selected cache caller/mode and patcher settings, plus available execution/load/memory/timing observations.",
+ "produced_artifact": "A state/owner/key/lifetime/consumer ledger with invalidation or release events, evidence and unresolved boundaries, followed by a discriminating check and its actual result.",
+ "failure_modes": "Treating identical JSON as proof of unchanged external inputs, confusing eviction with key validity, or equating workflow cache hits with device-resident model weights can explain the wrong state contract.",
+ "refusal_point": "Stop when the cause/action is supported or required environment, backend or model resources are unavailable; do not manufacture a hit, hardware result or performance threshold."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Draw the affected dependency cone with declared, lazy and dynamically expanded links, then mark unfingerprinted external state unresolved.",
+ "Separate output/object cache validity and retention from patch identity, partial unload, full detach and allocator-visible release.",
+ "Change one test variable while fixing prompt, output target, model, backend, precision, seed and cache mode; check values and executed nodes before timing."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/competitor-scan.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/competitor-scan.json
index 8cd803b37..fac9eb2db 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/competitor-scan.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/competitor-scan.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/workflows/competitor-scan/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/workflows/competitor-scan/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/skill-index.json",
"src/inferrsi/data/skill_library/workflows/competitor-scan/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dcp-async-save-performance.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dcp-async-save-performance.json
index aa309c547..fcffb14bc 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dcp-async-save-performance.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dcp-async-save-performance.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/SKILL.md",
"src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/metadata.json",
@@ -41,7 +41,7 @@
"checkout_ref": "checkout:dcp-async-save-performance",
"source_files": {
"src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/SKILL.md": "645139ee0c88ec865cad952e24cda35f4648d4b777b732c9b7d274b6c7830845",
- "src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/metadata.json": "db7ad368394ccbfdeff6a16e3423c300d51305b1f95ff5c3936df91738a8cc35",
+ "src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/metadata.json": "49764fa454403fb38891d82ee63de1e07a90d3ff6a3cd28550aee7ea34cbdcb0",
"src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/references/sources.md": "22afe4f4d77f441096cc657d55d763473bfcbea8fb531c8e493fe49ab84955f7",
"src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
},
@@ -71,12 +71,33 @@
"provenance_readsets": [
"references/sources.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "applicability": {
+ "scope": "Source-supported diagnosis of training-visible PyTorch DCP save pauses, snapshot/staging/upload lifetimes and bounded CPU staging-buffer reuse; distributed recovery validity and backend durability need separate evidence.",
+ "prerequisites": "Known correspondence between the installed PyTorch API and source/version notes, identified checkpointed-state mutation boundaries and an initial one-outstanding-save policy before overlap is considered.",
+ "input_contract": "Installed PyTorch revision, stager, storage writer and thread/process executor, checkpoint contents, ranks per host and state-mutation boundaries; performance assessment needs caller/staging/upload waits and host/resource observations.",
+ "produced_artifact": "A completion/lifetime diagnosis and smallest justified staging/cache/backpressure lever, with separate caller-pause, staging, upload and step-time measurements proposed for authorized validation.",
+ "failure_modes": "Mutating source state before staging completes, overwriting cached snapshots before prior upload completes, ignoring caller-side state_dict work or treating async return as durability can corrupt the snapshot or misstate the pause.",
+ "refusal_point": "Reject staging overlap that races checkpointed-state mutation; do not infer backend durability from asynchronous return or writer completion."
+ },
+ "recommendations": [
+ "Place the staging wait before the earliest mutation of any checkpointed state, including forward-mutated buffers rather than only optimizer.step.",
+ "Wait for the previous upload before reusing a stager's CPU storage; observe the final upload before closing caller-owned staging resources.",
+ "Measure checkpoint and intervening steps with the same state/cadence, retaining first-use and cached-buffer costs plus per-host memory and I/O pressure."
+ ]
+ },
+ "applicability": {
+ "scope": "Source-supported diagnosis of training-visible PyTorch DCP save pauses, snapshot/staging/upload lifetimes and bounded CPU staging-buffer reuse; distributed recovery validity and backend durability need separate evidence.",
+ "prerequisites": "Known correspondence between the installed PyTorch API and source/version notes, identified checkpointed-state mutation boundaries and an initial one-outstanding-save policy before overlap is considered.",
+ "input_contract": "Installed PyTorch revision, stager, storage writer and thread/process executor, checkpoint contents, ranks per host and state-mutation boundaries; performance assessment needs caller/staging/upload waits and host/resource observations.",
+ "produced_artifact": "A completion/lifetime diagnosis and smallest justified staging/cache/backpressure lever, with separate caller-pause, staging, upload and step-time measurements proposed for authorized validation.",
+ "failure_modes": "Mutating source state before staging completes, overwriting cached snapshots before prior upload completes, ignoring caller-side state_dict work or treating async return as durability can corrupt the snapshot or misstate the pause.",
+ "refusal_point": "Reject staging overlap that races checkpointed-state mutation; do not infer backend durability from asynchronous return or writer completion."
},
- "applicability": {},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Place the staging wait before the earliest mutation of any checkpointed state, including forward-mutated buffers rather than only optimizer.step.",
+ "Wait for the previous upload before reusing a stager's CPU storage; observe the final upload before closing caller-owned staging resources.",
+ "Measure checkpoint and intervening steps with the same state/cadence, retaining first-use and cached-buffer costs plus per-host memory and I/O pressure."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dense-performance-evidence.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dense-performance-evidence.json
index 60d8c1c02..353b0db04 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dense-performance-evidence.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dense-performance-evidence.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/methods/dense-performance-evidence/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/methods/dense-performance-evidence/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/methods/dense-performance-evidence/SKILL.md",
"src/inferrsi/data/skill_library/methods/dense-performance-evidence/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diagnose-vllm-admission-pressure.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diagnose-vllm-admission-pressure.json
index 5cf0a0aa3..ffd38b4d1 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diagnose-vllm-admission-pressure.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diagnose-vllm-admission-pressure.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/SKILL.md",
"src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/metadata.json",
@@ -52,7 +52,7 @@
"checkout_ref": "checkout:diagnose-vllm-admission-pressure",
"source_files": {
"src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/SKILL.md": "d99fc28d48127d3bc627f2a0cab6f3f16158a77da49428838a08865adaea9905",
- "src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/metadata.json": "78459416734f61a8faa819d9a1a3ad7b9a985e8bf8e149813c604d15be6bb92f",
+ "src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/metadata.json": "a4a4c4b7645852bce4abafde8e4128baacbdf1bc2a69362f7c3120db597fbe40",
"src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/references/papers-20261015/INTEGRATION_NOTES.md": "82b24be82451ee121892f61575d0cf4ecb9920813a25ccb26cce2b45dda6d8b3",
"src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/references/papers-20261015/cases/migration-is-not-free-memory.md": "2ac168f2725cba7fc92dfcc67c01bf2258dff6a2dbe1ddb063bc60c175d82d0a",
"src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/references/papers-20261015/cases/parallelism-change-with-state-cost.md": "5accc81086e4500d66f894ef6af9951e1cbb76941f6c631aab5c5561a786d8d8",
@@ -106,12 +106,33 @@
"references/papers-20261015/references/orca-readset.md",
"references/source-notes.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "applicability": {
+ "scope": "Pinned vLLM V1 branch-based diagnosis of queued requests, low scheduled-token counts and KV-pressure preemption, separating admission/token/input/allocation/connector-wait constraints.",
+ "prerequisites": "The deployed scheduler must be mapped to the pinned branches, with effective configuration and a bounded request/step trace; another revision requires re-establishing that correspondence.",
+ "input_contract": "Deployed revision, scheduler class/configuration, connector role, KV layout, policy and speculative/encoder/concurrent-batch features; per-step queues, tokens, both budgets, allocation results, victims, transfer/free state and timing.",
+ "produced_artifact": "A per-step accounting ledger, observed limiting branch and competing explanation, missing instrumentation and one conditional configuration/progress hypothesis with predicted effects and trade-offs.",
+ "failure_modes": "Inferring memory exhaustion from a queue or high KV usage, conflating token/input budgets, ignoring pending frees or reading async token accounting as load completion can attribute admission stalls to the wrong branch.",
+ "refusal_point": "When required state is missing, leave the diagnosis unresolved and request the smallest trace; escalate allocation or connector behavior outside the readset and stop before live mutation or performance claims."
+ },
+ "recommendations": [
+ "Trace the first actual blocking or skip branch and record running versus admission caps and token versus input budgets separately.",
+ "Inspect deferred frees and request-free eligibility before blaming preemption victim selection, then reconcile restored budgets for an already-scheduled victim.",
+ "Reconstruct waiting admission independently, including queue precedence and async prefix states; change no setting without a branch-specific hypothesis."
+ ]
+ },
+ "applicability": {
+ "scope": "Pinned vLLM V1 branch-based diagnosis of queued requests, low scheduled-token counts and KV-pressure preemption, separating admission/token/input/allocation/connector-wait constraints.",
+ "prerequisites": "The deployed scheduler must be mapped to the pinned branches, with effective configuration and a bounded request/step trace; another revision requires re-establishing that correspondence.",
+ "input_contract": "Deployed revision, scheduler class/configuration, connector role, KV layout, policy and speculative/encoder/concurrent-batch features; per-step queues, tokens, both budgets, allocation results, victims, transfer/free state and timing.",
+ "produced_artifact": "A per-step accounting ledger, observed limiting branch and competing explanation, missing instrumentation and one conditional configuration/progress hypothesis with predicted effects and trade-offs.",
+ "failure_modes": "Inferring memory exhaustion from a queue or high KV usage, conflating token/input budgets, ignoring pending frees or reading async token accounting as load completion can attribute admission stalls to the wrong branch.",
+ "refusal_point": "When required state is missing, leave the diagnosis unresolved and request the smallest trace; escalate allocation or connector behavior outside the readset and stop before live mutation or performance claims."
},
- "applicability": {},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Trace the first actual blocking or skip branch and record running versus admission caps and token versus input budgets separately.",
+ "Inspect deferred frees and request-free eligibility before blaming preemption victim selection, then reconcile restored budgets for an already-scheduled victim.",
+ "Reconstruct waiting admission independently, including queue precedence and async prefix states; change no setting without a branch-specific hypothesis."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diffusers-invocation-state.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diffusers-invocation-state.json
index f706019fc..d6b227f29 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diffusers-invocation-state.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diffusers-invocation-state.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/SKILL.md",
"src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/metadata.json",
@@ -42,7 +42,7 @@
"checkout_ref": "checkout:diffusers-invocation-state",
"source_files": {
"src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/SKILL.md": "7581b56336d07213a87d01477793ae1a8acf0b7cf876fce36c9a761ee32c1ff4",
- "src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/metadata.json": "68ac7f64b7ab947fc2db664bc3990f102c1cc5539b42350aa5818d0bbf17d86c",
+ "src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/metadata.json": "66f8d20c1644aba964adfff3b439bfd3ecd3875885191e6ce10721bc50877b1a",
"src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/references/ownership.md": "40688cc44183027efbf2fca8d3548c299cdf2e33fef54f76ab3549353eb9bd1b",
"src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/references/source-map.md": "10562a75683bd9bd07ec2e37a1cbd0f93f2534d6f85b55a566adb2d97485215f",
"src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/references/validation.md": "5a0b854b87d091d7d505b0bdd5c03330052507a851ab6edf35b93a6675f2d4f7",
@@ -84,7 +84,12 @@
}
],
"applicability": {
- "scope": "FluxPipeline and FlowMatchEulerDiscreteScheduler invocation ownership at this source pin; inspect the actual caller contract."
+ "scope": "FluxPipeline and FlowMatchEulerDiscreteScheduler invocation ownership at this source pin; inspect the actual caller contract.",
+ "prerequisites": "Access to the pinned ownership path and an identified caller admission/serialization contract; other pipeline or scheduler classes need their own contracts.",
+ "input_contract": "Diffusers/PyTorch revisions, pipeline/scheduler classes and object identities, construction/from_pipe path, actual components, scheduler config and resolved timesteps/sigmas, shape/dtype, callback/interrupt timing, output/cleanup path and overlapping intervals.",
+ "produced_artifact": "An object-identity/ownership map and call-to-schedule-reset/step/interrupt/output/cleanup trace, the caller admission contract, one next check and actual evidence status.",
+ "failure_modes": "Assuming from_pipe deep-copies scheduler state, passing an enumeration index instead of a scheduler timestep, or treating cooperative iteration skipping as kernel cancellation/exception cleanup can misidentify invocation ownership.",
+ "refusal_point": "Stop when the ownership question is resolved or a required fixture/backend is absent; do not infer service-wide race or serialization behavior from this readset."
},
"limitations": [
"Source-derived candidate; no installed API, native scheduler, model, concurrency or service validation.",
@@ -94,14 +99,25 @@
"provenance_readsets": [
"references/source-map.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Check wrapper and scheduler identities explicitly and follow the effective schedule reset and step-index progression.",
+ "Trace interruption through its loop-boundary consumer and normal decode/cleanup tail while preserving the caller's existing shared-state serialization.",
+ "Use a real scheduler CPU check only for its scheduler cycle; keep pipeline reuse/interruption and serving claims tied to matching fixtures and execution evidence."
+ ]
},
"applicability": {
- "scope": "FluxPipeline and FlowMatchEulerDiscreteScheduler invocation ownership at this source pin; inspect the actual caller contract."
+ "scope": "FluxPipeline and FlowMatchEulerDiscreteScheduler invocation ownership at this source pin; inspect the actual caller contract.",
+ "prerequisites": "Access to the pinned ownership path and an identified caller admission/serialization contract; other pipeline or scheduler classes need their own contracts.",
+ "input_contract": "Diffusers/PyTorch revisions, pipeline/scheduler classes and object identities, construction/from_pipe path, actual components, scheduler config and resolved timesteps/sigmas, shape/dtype, callback/interrupt timing, output/cleanup path and overlapping intervals.",
+ "produced_artifact": "An object-identity/ownership map and call-to-schedule-reset/step/interrupt/output/cleanup trace, the caller admission contract, one next check and actual evidence status.",
+ "failure_modes": "Assuming from_pipe deep-copies scheduler state, passing an enumeration index instead of a scheduler timestep, or treating cooperative iteration skipping as kernel cancellation/exception cleanup can misidentify invocation ownership.",
+ "refusal_point": "Stop when the ownership question is resolved or a required fixture/backend is absent; do not infer service-wide race or serialization behavior from this readset."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Check wrapper and scheduler identities explicitly and follow the effective schedule reset and step-index progression.",
+ "Trace interruption through its loop-boundary consumer and normal decode/cleanup tail while preserving the caller's existing shared-state serialization.",
+ "Use a real scheduler CPU check only for its scheduler cycle; keep pipeline reuse/interruption and serving claims tied to matching fixtures and execution evidence."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/draft-acceptance-issue.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/draft-acceptance-issue.json
index 35c1146f7..8c2305102 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/draft-acceptance-issue.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/draft-acceptance-issue.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/workflows/draft-acceptance-issue/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/workflows/draft-acceptance-issue/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/skill-index.json",
"src/inferrsi/data/skill_library/workflows/draft-acceptance-issue/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dwarfstar.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dwarfstar.json
index b60d6e126..ed140c6e6 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dwarfstar.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dwarfstar.json
@@ -25,8 +25,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/repositories/dwarfstar/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/repositories/dwarfstar/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/repositories/dwarfstar/LICENSE",
"src/inferrsi/data/skill_library/repositories/dwarfstar/SKILL.md",
@@ -57,7 +57,7 @@
"source_files": {
"src/inferrsi/data/skill_library/repositories/dwarfstar/LICENSE": "4a8a1358f5b048627d71e4988de0bafd386f4988991929629e6cb9e29b7b9b2f",
"src/inferrsi/data/skill_library/repositories/dwarfstar/SKILL.md": "a04a252ecdcd4710a3273595f4bd1f3eb8e24d5bb1bfe005f058a78388254dc9",
- "src/inferrsi/data/skill_library/repositories/dwarfstar/metadata.json": "e1f0c2dc3d7ea9278039c2e8c3e8ca4f8e3aba33504b7784d628e903f0b5fc8c",
+ "src/inferrsi/data/skill_library/repositories/dwarfstar/metadata.json": "c883f6bc05e585f072cf03f734b7a296a65842c7921d3afcfe89e1977405fbd2",
"src/inferrsi/data/skill_library/repositories/dwarfstar/references/repo-provenance.md": "7d72f9c85ee60fe003f5f5a585b52e9989ebcedca625b584fa39a976d4433998",
"src/inferrsi/data/skill_library/repositories/dwarfstar/references/repo-routing-metadata.json": "8e195b420b67151932af5f435c284f49677cc7863e650c83cfa9e762fd00db0b",
"src/inferrsi/data/skill_library/repositories/dwarfstar/references/source-identity.json": "a38fe07d003accd83b7626d7e189f7b49b4b73e7a223e4d441b0e33add5d9c91",
@@ -154,7 +154,12 @@
"method": "User-approved native C adaptation and partial drafting; manual stable-ID subskill orchestration and integrated create/verify stages. Required-backend blocks disable automatic admission."
},
"applicability": {
- "scope": "Pinned DwarfStar source investigation and experiment planning; not arbitrary GGUF support or cross-backend qualification."
+ "scope": "Pinned DwarfStar source investigation and experiment planning; not arbitrary GGUF support or cross-backend qualification.",
+ "prerequisites": "Access to the pinned native-source contract and relevant memory/state/qualification route; native verification needs a compatible build, matching hardware/weights and task-specific execution authority.",
+ "input_contract": "Engine revision/patches, actual model artifact/representation, selected backend/build and topology, memory/storage budget, context/chunk/session count, image inputs, sampling/speculation policy, cache state and quality/service requirements.",
+ "produced_artifact": "A bounded executing-tuple report with observed signal, source-backed mechanism, alternate explanation, proposed falsifier, actual check outcome and unresolved backend/model/service gap.",
+ "failure_modes": "Inferring model support from flags, aliases, file mapping or quantization names can select the wrong graph; treating CPU bookkeeping or parser checks as native backend/service qualification can overstate the result.",
+ "refusal_point": "Stop at unavailable hardware/weights, unsafe exposure or the authorized budget; retain memory-guard refusals rather than disabling the guard to claim fit."
},
"limitations": [
"Metal/CUDA/ROCm inference, actual model output, SSD runtime I/O, GPU recurrent rollback, physical distributed behavior, service quality/performance and skill efficacy remain unverified.",
@@ -166,14 +171,25 @@
"references/repo-provenance.md",
"references/source-identity.json"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Choose the memory, serving-state or qualification route by the observed symptom and identify the owner, validity key and release/rollback boundary.",
+ "Freeze comparison controls, test the smallest invariant first and retain fallback messages, cancellations, refusals and unavailable fixtures.",
+ "Base model support on the actual loaded artifact and graph; use native backend evidence for model, tensor-parallel or service conclusions."
+ ]
},
"applicability": {
- "scope": "Pinned DwarfStar source investigation and experiment planning; not arbitrary GGUF support or cross-backend qualification."
+ "scope": "Pinned DwarfStar source investigation and experiment planning; not arbitrary GGUF support or cross-backend qualification.",
+ "prerequisites": "Access to the pinned native-source contract and relevant memory/state/qualification route; native verification needs a compatible build, matching hardware/weights and task-specific execution authority.",
+ "input_contract": "Engine revision/patches, actual model artifact/representation, selected backend/build and topology, memory/storage budget, context/chunk/session count, image inputs, sampling/speculation policy, cache state and quality/service requirements.",
+ "produced_artifact": "A bounded executing-tuple report with observed signal, source-backed mechanism, alternate explanation, proposed falsifier, actual check outcome and unresolved backend/model/service gap.",
+ "failure_modes": "Inferring model support from flags, aliases, file mapping or quantization names can select the wrong graph; treating CPU bookkeeping or parser checks as native backend/service qualification can overstate the result.",
+ "refusal_point": "Stop at unavailable hardware/weights, unsafe exposure or the authorized budget; retain memory-guard refusals rather than disabling the guard to claim fit."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Choose the memory, serving-state or qualification route by the observed symptom and identify the owner, validity key and release/rollback boundary.",
+ "Freeze comparison controls, test the smallest invariant first and retain fallback messages, cancellations, refusals and unavailable fixtures.",
+ "Base model support on the actual loaded artifact and graph; use native backend evidence for model, tensor-parallel or service conclusions."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/evaluate-elastic-serving-placement.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/evaluate-elastic-serving-placement.json
index 701eedf8b..350486a6d 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/evaluate-elastic-serving-placement.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/evaluate-elastic-serving-placement.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/methods/evaluate-elastic-serving-placement/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/methods/evaluate-elastic-serving-placement/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/methods/evaluate-elastic-serving-placement/SKILL.md",
"src/inferrsi/data/skill_library/methods/evaluate-elastic-serving-placement/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/generation-cache-lifecycle.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/generation-cache-lifecycle.json
index b1f38ce58..b8f4eeb34 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/generation-cache-lifecycle.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/generation-cache-lifecycle.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/SKILL.md",
"src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/metadata.json",
@@ -43,7 +43,7 @@
"checkout_ref": "checkout:generation-cache-lifecycle",
"source_files": {
"src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/SKILL.md": "bf1177cb9985c12a9bf11edb64771018e1e314c4713920de8fd4a989ff70115f",
- "src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/metadata.json": "d86cecb615e93ba0f6be09d1ead102cd5a17068c07e1be2549ac5cd8f7519279",
+ "src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/metadata.json": "c2e30a2a3365dd27cac7916dc26aa356f1277d07ebee891c173ee37b77111fa1",
"src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/references/cache-stop.md": "bba6b5c73261065309cbcf11b1e591eef48916f0c79ceaaf74d98a9fa96b0992",
"src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/references/generation-batch.md": "f16d3c57d45e63bcaffd3871aa41c76972ee11f7dbf1e12b45d9b170303cad9b",
"src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/references/source-map.md": "91d26181f246307f2113f94b12ac9587d689b654dcd216e9f14b11cf5ebb650c",
@@ -89,7 +89,12 @@
}
],
"applicability": {
- "scope": "MLX-LM generation, batch-row and prefix-cache diagnosis; model/backend/runtime compatibility requires separate qualification."
+ "scope": "MLX-LM generation, batch-row and prefix-cache diagnosis; model/backend/runtime compatibility requires separate qualification.",
+ "prerequisites": "Access to the independently pinned MLX-LM generation/cache contracts and the selected entry path; model, backend and runtime compatibility is separately established for any native claim.",
+ "input_contract": "MLX-LM/MLX revisions and builds, backend/device, model/tokenizer/adapter/draft identities, entry point, tokenized prompts, cache classes, batching/prefill limits, per-request sampling/processors/stops, seed and KV quantization; failure traces include UIDs, stages, lengths, terminal reason and next-request result.",
+ "produced_artifact": "An entry-path decision, UID/stage/cache-row table, prefix identity and cache-class assumptions, terminal/cleanup trace and actual evidence status with a targeted verification plan.",
+ "failure_modes": "Ignoring server batching gates, filtering only some per-request row state, reusing a text-only prefix key or treating a stop flag as immediate MLX work cancellation can corrupt lifecycle reasoning.",
+ "refusal_point": "Stop at the first unsupported contract or missing required fixture/backend; retain a targeted plan without inferring paged allocation, radix sharing or instantaneous prefill interruption."
},
"limitations": [
"Source-derived candidate; no local native-backend, model, service or efficacy validation.",
@@ -98,14 +103,25 @@
"provenance_readsets": [
"references/source-map.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Resolve the library versus server entry and its batching gates before transferring BatchGenerator behavior.",
+ "Keep UID, caches, tokens, samplers, processors, lengths, stop matchers and pending outputs aligned under the same removal selection, including prompt-stage requests.",
+ "Check exact token/model/adapter/draft identity and cache-type copy/trim semantics, then trace cooperative removal through survivor and next-request behavior."
+ ]
},
"applicability": {
- "scope": "MLX-LM generation, batch-row and prefix-cache diagnosis; model/backend/runtime compatibility requires separate qualification."
+ "scope": "MLX-LM generation, batch-row and prefix-cache diagnosis; model/backend/runtime compatibility requires separate qualification.",
+ "prerequisites": "Access to the independently pinned MLX-LM generation/cache contracts and the selected entry path; model, backend and runtime compatibility is separately established for any native claim.",
+ "input_contract": "MLX-LM/MLX revisions and builds, backend/device, model/tokenizer/adapter/draft identities, entry point, tokenized prompts, cache classes, batching/prefill limits, per-request sampling/processors/stops, seed and KV quantization; failure traces include UIDs, stages, lengths, terminal reason and next-request result.",
+ "produced_artifact": "An entry-path decision, UID/stage/cache-row table, prefix identity and cache-class assumptions, terminal/cleanup trace and actual evidence status with a targeted verification plan.",
+ "failure_modes": "Ignoring server batching gates, filtering only some per-request row state, reusing a text-only prefix key or treating a stop flag as immediate MLX work cancellation can corrupt lifecycle reasoning.",
+ "refusal_point": "Stop at the first unsupported contract or missing required fixture/backend; retain a targeted plan without inferring paged allocation, radix sharing or instantaneous prefill interruption."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Resolve the library versus server entry and its batching gates before transferring BatchGenerator behavior.",
+ "Keep UID, caches, tokens, samplers, processors, lengths, stop matchers and pending outputs aligned under the same removal selection, including prompt-stage requests.",
+ "Check exact token/model/adapter/draft identity and cache-type copy/trim semantics, then trace cooperative removal through survivor and next-request behavior."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-memory-management.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-memory-management.json
index 35039c25e..4be6a69c7 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-memory-management.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-memory-management.json
@@ -30,8 +30,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/hardware/host/host-memory-management/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/hardware/host/host-memory-management/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/hardware/host/host-memory-management/SKILL.md",
"src/inferrsi/data/skill_library/hardware/host/host-memory-management/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-overhead-optimization.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-overhead-optimization.json
index 5e56342b1..e4cd8ff0f 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-overhead-optimization.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-overhead-optimization.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/hardware/host/host-overhead-optimization/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/hardware/host/host-overhead-optimization/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/hardware/host/host-overhead-optimization/SKILL.md",
"src/inferrsi/data/skill_library/hardware/host/host-overhead-optimization/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/husky.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/husky.json
index f88743bb3..53db4dfaa 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/husky.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/husky.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/repositories/husky/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/repositories/husky/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/repositories/husky/SKILL.md",
"src/inferrsi/data/skill_library/repositories/husky/metadata.json",
@@ -44,7 +44,7 @@
"checkout_ref": "checkout:husky",
"source_files": {
"src/inferrsi/data/skill_library/repositories/husky/SKILL.md": "ac86afdabf9adffb684cb635971934438724bdfde8628ec549727ea23c40cf3b",
- "src/inferrsi/data/skill_library/repositories/husky/metadata.json": "9b720bc6fcb65b6dbddba36d6187b4bee86346a1bc633f55a75bd43dbe4fa04e",
+ "src/inferrsi/data/skill_library/repositories/husky/metadata.json": "b1cb0a5be02d1170125bfc3e16c044b9c1463a611a14d36fac40638524fc2e52",
"src/inferrsi/data/skill_library/repositories/husky/references/author-report.md": "4c1351ca42b2535a7c03c4da87706040650ef10d86eadcdb496e881a147fc0f7",
"src/inferrsi/data/skill_library/repositories/husky/references/repo-provenance.md": "37de912e2e2a36b44233560e6a8368b994428bff587ed4f04b1cb5fbaba0a9e7",
"src/inferrsi/data/skill_library/repositories/husky/references/source-availability.json": "d3523e53c89d4aee5bfaa17db84501aa4c4eddd59bcee486df5d48d3aa48173d",
@@ -105,7 +105,12 @@
"method": "Single coherent research workflow, bounded source summary, original verification questions and offline schema checks. Repository/environment/native stages remain unavailable."
},
"applicability": {
- "scope": "Review Husky document claims and specify missing evidence; common catalog scope is not inference backend compatibility."
+ "scope": "Review Husky document claims and specify missing evidence; common catalog scope is not inference backend compatibility.",
+ "prerequisites": "Access to the bounded official-document inventory and verification protocol; implementation and native reproduction remain unavailable until separately supplied and authenticated.",
+ "input_contract": "The exact claim, source/view date, workload and comparison intent, supporting evidence classes and available implementation/model/output/measurement artifacts; unknown identity, licensing and availability remain explicit.",
+ "produced_artifact": "A reviewable claim/evidence request recording source availability, mechanism hypothesis, counter-signal, missing identities/artifacts, proposed falsifying check and actual result or bounded INCONCLUSIVE conclusion.",
+ "failure_modes": "Treating a public-page reading as code inspection, proposed extensions as shipped behavior, summed independent ratios as one gain, or a structurally valid self-reported card as authentic evidence can overstate qualification.",
+ "refusal_point": "Stop at missing source/receipts, unsafe or private data, unsupported resources or the authorized budget; unavailable artifacts yield an unverified/INCONCLUSIVE research conclusion."
},
"limitations": [
"No canonical public repository, implementation revision, retained receipts or source-code license was identified; unknowns remain unknown.",
@@ -119,14 +124,25 @@
"references/repo-provenance.md",
"references/source-availability.json"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Separate author reports, inspected implementation, executed experiments and new interpretations, including negative findings and future work.",
+ "Request immutable identities and artifacts, then state one mechanism, competing explanation and smallest ablation with frozen semantic requirements.",
+ "Use the offline card auditor only to identify structural gaps; it authenticates neither artifacts nor execution and leaves qualification unadjudicated."
+ ]
},
"applicability": {
- "scope": "Review Husky document claims and specify missing evidence; common catalog scope is not inference backend compatibility."
+ "scope": "Review Husky document claims and specify missing evidence; common catalog scope is not inference backend compatibility.",
+ "prerequisites": "Access to the bounded official-document inventory and verification protocol; implementation and native reproduction remain unavailable until separately supplied and authenticated.",
+ "input_contract": "The exact claim, source/view date, workload and comparison intent, supporting evidence classes and available implementation/model/output/measurement artifacts; unknown identity, licensing and availability remain explicit.",
+ "produced_artifact": "A reviewable claim/evidence request recording source availability, mechanism hypothesis, counter-signal, missing identities/artifacts, proposed falsifying check and actual result or bounded INCONCLUSIVE conclusion.",
+ "failure_modes": "Treating a public-page reading as code inspection, proposed extensions as shipped behavior, summed independent ratios as one gain, or a structurally valid self-reported card as authentic evidence can overstate qualification.",
+ "refusal_point": "Stop at missing source/receipts, unsafe or private data, unsupported resources or the authorized budget; unavailable artifacts yield an unverified/INCONCLUSIVE research conclusion."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Separate author reports, inspected implementation, executed experiments and new interpretations, including negative findings and future work.",
+ "Request immutable identities and artifacts, then state one mechanism, competing explanation and smallest ablation with frozen semantic requirements.",
+ "Use the offline card auditor only to identify structural gaps; it authenticates neither artifacts nor execution and leaves qualification unadjudicated."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-benchmark-design.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-benchmark-design.json
index 89a2500c9..d71a20ca9 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-benchmark-design.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-benchmark-design.json
@@ -30,8 +30,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/methods/inference-benchmark-design/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/methods/inference-benchmark-design/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/methods/inference-benchmark-design/SKILL.md",
"src/inferrsi/data/skill_library/methods/inference-benchmark-design/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-optimization-casebook.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-optimization-casebook.json
index 2b589b439..03a0ff9b7 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-optimization-casebook.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-optimization-casebook.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/methods/inference-optimization-casebook/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/methods/inference-optimization-casebook/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/methods/inference-optimization-casebook/SKILL.md",
"src/inferrsi/data/skill_library/methods/inference-optimization-casebook/cases/admission-drain-occupancy.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kernel-writing-guide.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kernel-writing-guide.json
index 7f7a01ee7..750d4b091 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kernel-writing-guide.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kernel-writing-guide.json
@@ -39,8 +39,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/methods/kernel-writing-guide/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/methods/kernel-writing-guide/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/methods/kernel-writing-guide/SKILL.md",
"src/inferrsi/data/skill_library/methods/kernel-writing-guide/cases/causal-conv1d-dispatch-state.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kv-tier-management.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kv-tier-management.json
index 7864009ba..de9737b7c 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kv-tier-management.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kv-tier-management.json
@@ -24,8 +24,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/repositories/lmcache/kv-tier-management/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/repositories/lmcache/kv-tier-management/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/repositories/lmcache/kv-tier-management/SKILL.md",
"src/inferrsi/data/skill_library/repositories/lmcache/kv-tier-management/metadata.json",
@@ -41,7 +41,7 @@
"checkout_ref": "checkout:kv-tier-management",
"source_files": {
"src/inferrsi/data/skill_library/repositories/lmcache/kv-tier-management/SKILL.md": "e4b78f368d70c8abb74a935acb910414015f1aeec1b50f7a295e836822ec4b27",
- "src/inferrsi/data/skill_library/repositories/lmcache/kv-tier-management/metadata.json": "c221189b8d85fac3bd41ef1c7cd560f1aebc8cfa0381c182d1d571a7169be829",
+ "src/inferrsi/data/skill_library/repositories/lmcache/kv-tier-management/metadata.json": "85f7242b83334af11f57799875738b498bfe4680a51f36727792dfbe91806704",
"src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
},
"skill_metadata": {
@@ -69,12 +69,31 @@
"Original typed source inventory is not normalized; retain source identities and limits in the procedure/readsets."
],
"provenance_readsets": [],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "applicability": {
+ "scope": "LMCache multiprocess runtime KV tier policy across engine GPU cache, host L1 and configured L2 adapters, covering store/retention, prefetch admission, eviction and RAM pressure; startup loading and checkpoint recovery have separate owners.",
+ "prerequisites": "The deployment uses the inspected MP-mode semantics rather than single-process knobs; selected GDS/DAX/raw-block paths need their own platform and adapter prerequisites. Lazy offload additionally requires vLLM prefix caching.",
+ "input_contract": "MP versus single-process mode, store/retention policy and rationale, L1 manager capacities and eviction settings, L2 adapter capacities/eviction, chunk-size distribution and prefetch in-flight setting, together with token-level hit and pressure counters.",
+ "produced_artifact": "A mode- and adapter-specific tier-policy choice with retention rationale, an ordered RAM-release plan and a verification record separating L1/L2 token hits, prefetch pressure, store lag and per-manager usage.",
+ "failure_modes": "Mixing MP and in-process controls, treating key-count eviction as a byte fraction for unequal chunks, omitting fs capacity management, or increasing count-based prefetch admission without tracking L1 pressure can make the policy or memory claim unusable."
+ },
+ "recommendations": [
+ "Choose store retention and prefetched-key retention separately; use skip_l1 with noop only for buffer-only operation and account for failed stores remaining in L1.",
+ "Tune the count-based prefetch in-flight limit against L1 allocation failures and in-flight bytes; eager or warm prefetch can raise pressure before useful reuse.",
+ "Record per-adapter capacity/eviction and actual chunk-size uniformity; report token-level L1/L2 hits and sum per-l1_tag bytes rather than ratios."
+ ]
+ },
+ "applicability": {
+ "scope": "LMCache multiprocess runtime KV tier policy across engine GPU cache, host L1 and configured L2 adapters, covering store/retention, prefetch admission, eviction and RAM pressure; startup loading and checkpoint recovery have separate owners.",
+ "prerequisites": "The deployment uses the inspected MP-mode semantics rather than single-process knobs; selected GDS/DAX/raw-block paths need their own platform and adapter prerequisites. Lazy offload additionally requires vLLM prefix caching.",
+ "input_contract": "MP versus single-process mode, store/retention policy and rationale, L1 manager capacities and eviction settings, L2 adapter capacities/eviction, chunk-size distribution and prefetch in-flight setting, together with token-level hit and pressure counters.",
+ "produced_artifact": "A mode- and adapter-specific tier-policy choice with retention rationale, an ordered RAM-release plan and a verification record separating L1/L2 token hits, prefetch pressure, store lag and per-manager usage.",
+ "failure_modes": "Mixing MP and in-process controls, treating key-count eviction as a byte fraction for unequal chunks, omitting fs capacity management, or increasing count-based prefetch admission without tracking L1 pressure can make the policy or memory claim unusable."
},
- "applicability": {},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Choose store retention and prefetched-key retention separately; use skip_l1 with noop only for buffer-only operation and account for failed stores remaining in L1.",
+ "Tune the count-based prefetch in-flight limit against L1 allocation failures and in-flight bytes; eager or warm prefetch can raise pressure before useful reuse.",
+ "Record per-adapter capacity/eviction and actual chunk-size uniformity; report token-level L1/L2 hits and sum per-l1_tag bytes rather than ratios."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lazy-evaluation-stream-lifetimes.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lazy-evaluation-stream-lifetimes.json
index ceaa9e2f9..7def78509 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lazy-evaluation-stream-lifetimes.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lazy-evaluation-stream-lifetimes.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/SKILL.md",
"src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/metadata.json",
@@ -42,7 +42,7 @@
"checkout_ref": "checkout:lazy-evaluation-stream-lifetimes",
"source_files": {
"src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/SKILL.md": "a95569bcc0da6e715fd74783a6a1e0489e92b680c582f15dafbf98df589d505a",
- "src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/metadata.json": "9c2293c8956508de987c886c7224d673d22912e7e915b026b05abdd287e97401",
+ "src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/metadata.json": "cfa57ac908025135113d04c3d64b9967f8d9a3af4a00707673d94b8227c85808",
"src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/references/execution-lifetime.md": "c4bc41e08aca68f661325f61e6f0db1760a079dc362f295c847bdc374fee3bf6",
"src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/references/source-map.md": "798b9cb89ed6b885eb47464ca9f137d2802a54571b9327e631d7742d3e5abd5d",
"src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/references/validation.md": "ebfc00bf7d6ab0a1dc415953ae84d56ccdc713d5614aff85e433c4a843793bff",
@@ -94,7 +94,12 @@
}
],
"applicability": {
- "scope": "MLX evaluation/stream/buffer diagnosis at the pinned implementation; qualify the actual backend separately."
+ "scope": "MLX evaluation/stream/buffer diagnosis at the pinned implementation; qualify the actual backend separately.",
+ "prerequisites": "Access to the pinned MLX execution/lifetime contract and a bounded reproduction with an identified evaluation/read boundary; device-dependent conclusions need the actual selected backend.",
+ "input_contract": "MLX revision/build, backend/device, Python/C++ entry, array shapes/dtypes, primitive streams, evaluation/read boundary and transformation context; timing needs warmup/work boundaries, while memory needs live references, in-flight buffers, allocator cache and synchronized observations.",
+ "produced_artifact": "A producer/evaluation/dependency-wait/backend/availability-cleanup trace, ownership table, observed symptom and one discriminating check with build-bound actual evidence status.",
+ "failure_modes": "Timing graph construction or submission as execution, assuming evaluated means completed, synchronizing only the default stream, or treating graph detachment/Python deletion as reusable storage can misdiagnose time or memory.",
+ "refusal_point": "Stop when the trace answers the question; if the required backend is absent, retain the source-supported diagnosis and exact runtime gap rather than substituting CPU import success."
},
"limitations": [
"Source-derived candidate; no local native-backend, model, service or efficacy validation.",
@@ -103,14 +108,25 @@
"provenance_readsets": [
"references/source-map.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Separate graph construction, async_eval submission, eval and host reads, and record which stream the measurement actually synchronizes.",
+ "Preserve the dependency-wait path and selected backend buffer-retention/donation contract before changing streams or assuming storage can be reused.",
+ "Distinguish live, in-flight and allocator-cached memory; inspect the real error/cleanup path before proposing a leak fix or hiding the failure."
+ ]
},
"applicability": {
- "scope": "MLX evaluation/stream/buffer diagnosis at the pinned implementation; qualify the actual backend separately."
+ "scope": "MLX evaluation/stream/buffer diagnosis at the pinned implementation; qualify the actual backend separately.",
+ "prerequisites": "Access to the pinned MLX execution/lifetime contract and a bounded reproduction with an identified evaluation/read boundary; device-dependent conclusions need the actual selected backend.",
+ "input_contract": "MLX revision/build, backend/device, Python/C++ entry, array shapes/dtypes, primitive streams, evaluation/read boundary and transformation context; timing needs warmup/work boundaries, while memory needs live references, in-flight buffers, allocator cache and synchronized observations.",
+ "produced_artifact": "A producer/evaluation/dependency-wait/backend/availability-cleanup trace, ownership table, observed symptom and one discriminating check with build-bound actual evidence status.",
+ "failure_modes": "Timing graph construction or submission as execution, assuming evaluated means completed, synchronizing only the default stream, or treating graph detachment/Python deletion as reusable storage can misdiagnose time or memory.",
+ "refusal_point": "Stop when the trace answers the question; if the required backend is absent, retain the source-supported diagnosis and exact runtime gap rather than substituting CPU import success."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Separate graph construction, async_eval submission, eval and host reads, and record which stream the measurement actually synchronizes.",
+ "Preserve the dependency-wait path and selected backend buffer-retention/donation contract before changing streams or assuming storage can be reused.",
+ "Distinguish live, in-flight and allocator-cached memory; inspect the real error/cleanup path before proposing a leak fix or hiding the failure."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lily.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lily.json
index ee17263cf..b0f8eaf32 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lily.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lily.json
@@ -25,8 +25,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/repositories/lily/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/repositories/lily/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/repositories/lily/LICENSE",
"src/inferrsi/data/skill_library/repositories/lily/NOTICE.md",
@@ -61,7 +61,7 @@
"src/inferrsi/data/skill_library/repositories/lily/LICENSE": "b9e2b50a48a09fb39dafa71ab1a11a6e87389addda17faa99d55f5f74a0001ac",
"src/inferrsi/data/skill_library/repositories/lily/NOTICE.md": "7a01e666044fb8aca1fd88c636bca43bcc95b8b1fecbe101dce4c7c80fbb5759",
"src/inferrsi/data/skill_library/repositories/lily/SKILL.md": "db815674d5d36964902041d310d3df844d00ab2e6bef0ce4b72ed2696ab2b9aa",
- "src/inferrsi/data/skill_library/repositories/lily/metadata.json": "6d9d54e84f957881cc26894eb82ae666f106cfc6c505924f9a6fb6b35eab457f",
+ "src/inferrsi/data/skill_library/repositories/lily/metadata.json": "52f01d58613e1a93e9cfd958ceca9ab5061baa07c5f94572393520434892b420",
"src/inferrsi/data/skill_library/repositories/lily/references/repo-provenance.md": "96182f6cb3666f89adf1aed6e216c0a84ce3a04a0685a4257e286100ba8ee367",
"src/inferrsi/data/skill_library/repositories/lily/references/repo-routing-metadata.json": "f2f3becb51056e03081c7c4140e19572a6318c3d63015884df56eb49d38a00fc",
"src/inferrsi/data/skill_library/repositories/lily/references/source-identity.json": "4595559c11611c855be150919fca3eeb89e7f96c7a8dd133b9f79a5cd8e05e62",
@@ -177,7 +177,12 @@
"method": "User-approved native Rust/Metal and partial drafting adaptation; stable-ID lanes, integrated source graph and assertion/helper verification. Missing compiler/backend/checkpoint remains explicit."
},
"applicability": {
- "scope": "Exact Lily source/checkpoint/layout/API/cache/qualification investigation; common catalog classification is not CPU inference or cross-device support."
+ "scope": "Exact Lily source/checkpoint/layout/API/cache/qualification investigation; common catalog classification is not CPU inference or cross-device support.",
+ "prerequisites": "Access to the exact Lily source/model-layout/API contracts; native checks require the matching Rust/Metal/checkpoint environment and frozen comparison requirements.",
+ "input_contract": "Source revision, Cargo lock/toolchain, macOS/chip/GPU family, checkpoint and tensor payload identities, affine configuration/layout, tokenizer/template/stops, sequence/output budgets, greedy policy, input history, prefix-cache state and quality/measurement requirements.",
+ "produced_artifact": "A source/model/device-bound representation, serving-cache or qualification diagnosis with the observation, invariant, alternate cause, smallest falsifier, actual result and remaining native gap.",
+ "failure_modes": "Equal bit width, file size, API names or prompt_cache_key do not establish tensor/token/state equivalence; ignoring the extra lookahead slot can understate request capacity, and green CPU references do not replace matched-model Metal goldens.",
+ "refusal_point": "Stop at unsupported formats/APIs, missing compiler/backend/checkpoint, unfrozen gates or the authorized budget; do not weaken ignored/golden tests to obtain a green result."
},
"limitations": [
"Cargo/Rust1.92, supported M5/macOS26/Metal and model resources are unavailable; native evidence remains BLOCKED_REQUIRED_BACKEND.",
@@ -190,14 +195,25 @@
"references/repo-provenance.md",
"references/source-identity.json"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Resolve exact affine representation and strict API semantics, then trace token frontier, cache key and restore/release ownership.",
+ "Retain the source's extra request lookahead slot in capacity arithmetic rather than counting only prompt and visible output.",
+ "Separate reported benchmark version from the checked-out runner and distinguish offline screening, native compilation, Metal operators and matched-model goldens."
+ ]
},
"applicability": {
- "scope": "Exact Lily source/checkpoint/layout/API/cache/qualification investigation; common catalog classification is not CPU inference or cross-device support."
+ "scope": "Exact Lily source/checkpoint/layout/API/cache/qualification investigation; common catalog classification is not CPU inference or cross-device support.",
+ "prerequisites": "Access to the exact Lily source/model-layout/API contracts; native checks require the matching Rust/Metal/checkpoint environment and frozen comparison requirements.",
+ "input_contract": "Source revision, Cargo lock/toolchain, macOS/chip/GPU family, checkpoint and tensor payload identities, affine configuration/layout, tokenizer/template/stops, sequence/output budgets, greedy policy, input history, prefix-cache state and quality/measurement requirements.",
+ "produced_artifact": "A source/model/device-bound representation, serving-cache or qualification diagnosis with the observation, invariant, alternate cause, smallest falsifier, actual result and remaining native gap.",
+ "failure_modes": "Equal bit width, file size, API names or prompt_cache_key do not establish tensor/token/state equivalence; ignoring the extra lookahead slot can understate request capacity, and green CPU references do not replace matched-model Metal goldens.",
+ "refusal_point": "Stop at unsupported formats/APIs, missing compiler/backend/checkpoint, unfrozen gates or the authorized budget; do not weaken ignored/golden tests to obtain a green result."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Resolve exact affine representation and strict API semantics, then trace token frontier, cache key and restore/release ownership.",
+ "Retain the source's extra request lookahead slot in capacity arithmetic rather than counting only prompt and visible output.",
+ "Separate reported benchmark version from the checked-out runner and distinguish offline screening, native compilation, Metal operators and matched-model goldens."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lithos-metal.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lithos-metal.json
index 855174e9e..86745e8f3 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lithos-metal.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lithos-metal.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/repositories/lithos-metal/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/repositories/lithos-metal/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/repositories/lithos-metal/LICENSE",
"src/inferrsi/data/skill_library/repositories/lithos-metal/SKILL.md",
@@ -60,7 +60,7 @@
"source_files": {
"src/inferrsi/data/skill_library/repositories/lithos-metal/LICENSE": "cfc7749b96f63bd31c3c42b5c471bf756814053e847c10f3eb003417bc523d30",
"src/inferrsi/data/skill_library/repositories/lithos-metal/SKILL.md": "4c71ba2f26c275829466bba2412738aa27799842e642c75cdfd36d9ed8713f7f",
- "src/inferrsi/data/skill_library/repositories/lithos-metal/metadata.json": "ea2d4e73e33a49433a2834d19489653f828e56f69c4975cc32cf508176eb24ae",
+ "src/inferrsi/data/skill_library/repositories/lithos-metal/metadata.json": "c5fa717d6f79d54e6aa2b0c5b672e80ac7412fbe5901ae24fda78a24c584ba67",
"src/inferrsi/data/skill_library/repositories/lithos-metal/references/mlx-license.md": "ccfab7ccb2ea306f71531c8ca77bb55507606cd90768b1e32b8b52ab5b48cf01",
"src/inferrsi/data/skill_library/repositories/lithos-metal/references/notices.md": "b2c9d2481ef2c9d16b3240c951ae65db417e5d309cd06b7917db9d84fe175886",
"src/inferrsi/data/skill_library/repositories/lithos-metal/references/package-api.md": "4f3ecfabafb2a15eddf21ccf964b8c37635bbb7862418a2c404e7049c0fa2898",
@@ -132,7 +132,11 @@
}
],
"applicability": {
- "scope": "Pinned Lithos Metal 0.1.2 source/API contracts; Apple silicon Metal runtime; model and recipe qualification required separately."
+ "scope": "Pinned Lithos Metal 0.1.2 source/API contracts; Apple silicon Metal runtime; model and recipe qualification required separately.",
+ "prerequisites": "Access to the pinned package/API and selected compiler, generation or serving contracts; revision/API drift needs a fresh bounded inspection, and efficacy claims need campaign-owned gates and matching native/model evidence.",
+ "input_contract": "Actual source/package version, backend/configuration, chip/core count and OS, target/draft checkpoint identities, context and prompt-token IDs, sampling/numerical options and the affected recipe/runtime/session/cache boundary.",
+ "produced_artifact": "A routed region/dependency, committed-state or cache-identity/lifecycle ledger, followed by the smallest supported finding, missing checks and next bounded action.",
+ "failure_modes": "Treating session adoption/cache hits as generation correctness, fewer dispatches as faster rounds, another chip recipe as the default, CPU cache checks as GPU rollback or health liveness as loaded/warm model state can yield unsupported conclusions."
},
"limitations": [
"Source-derived candidate; empirical efficacy validation not_run.",
@@ -147,14 +151,24 @@
"disable-model-invocation": true,
"metadata": {
"disco-role": "operating"
- }
+ },
+ "recommendations": [
+ "Follow session replacement across recipe geometry, committed generation state and serving adoption/cache identity before interpreting a cache hit.",
+ "Keep an unfused/native or target-only control with matching decoded weights, numerical mode, row bounds and sampling policy; verify dependencies and persistent state before timing.",
+ "Distinguish package/CPU checks, synthetic Metal execution, model continuation and HTTP workload evidence; chip recipes and target-compute projections do not establish another device or E2E rate."
+ ]
},
"applicability": {
- "scope": "Pinned Lithos Metal 0.1.2 source/API contracts; Apple silicon Metal runtime; model and recipe qualification required separately."
+ "scope": "Pinned Lithos Metal 0.1.2 source/API contracts; Apple silicon Metal runtime; model and recipe qualification required separately.",
+ "prerequisites": "Access to the pinned package/API and selected compiler, generation or serving contracts; revision/API drift needs a fresh bounded inspection, and efficacy claims need campaign-owned gates and matching native/model evidence.",
+ "input_contract": "Actual source/package version, backend/configuration, chip/core count and OS, target/draft checkpoint identities, context and prompt-token IDs, sampling/numerical options and the affected recipe/runtime/session/cache boundary.",
+ "produced_artifact": "A routed region/dependency, committed-state or cache-identity/lifecycle ledger, followed by the smallest supported finding, missing checks and next bounded action.",
+ "failure_modes": "Treating session adoption/cache hits as generation correctness, fewer dispatches as faster rounds, another chip recipe as the default, CPU cache checks as GPU rollback or health liveness as loaded/warm model state can yield unsupported conclusions."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Follow session replacement across recipe geometry, committed generation state and serving adoption/cache identity before interpreting a cache hit.",
+ "Keep an unfused/native or target-only control with matching decoded weights, numerical mode, row bounds and sampling policy; verify dependencies and persistent state before timing.",
+ "Distinguish package/CPU checks, synthetic Metal execution, model continuation and HTTP workload evidence; chip recipes and target-compute projections do not establish another device or E2E rate."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/local-dream.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/local-dream.json
index 61a92eba6..1db732a43 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/local-dream.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/local-dream.json
@@ -29,8 +29,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/repositories/local-dream/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/repositories/local-dream/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/repositories/local-dream/SKILL.md",
"src/inferrsi/data/skill_library/repositories/local-dream/metadata.json",
@@ -53,7 +53,7 @@
"checkout_ref": "checkout:local-dream",
"source_files": {
"src/inferrsi/data/skill_library/repositories/local-dream/SKILL.md": "e51fd3c71d693070abeba57ca67dcca82c7347cb94eb98f43e7c455bcacbaa46",
- "src/inferrsi/data/skill_library/repositories/local-dream/metadata.json": "ab33602252d0f5bb0797c3e1f12466b55a82aeaff10cd24e42a99a5224c8e593",
+ "src/inferrsi/data/skill_library/repositories/local-dream/metadata.json": "63b71cc6a237d8c4e00db875d15594df09d6e3cf87a1cb25183a311b4d480804",
"src/inferrsi/data/skill_library/repositories/local-dream/references/android-lifecycle.md": "f85e5fce39f5de5f88548fd8c21978240c810f899e03ff05ed72eb3508ee5e36",
"src/inferrsi/data/skill_library/repositories/local-dream/references/native-contracts.md": "40206892f5e421f9d8edf303870f5c4aefa3e87123537178d45f438f3d04b7ce",
"src/inferrsi/data/skill_library/repositories/local-dream/references/packaging-and-abi.md": "9a19bca59e19c3bfbfc1a15f5fc756c13d954ace367812aeb000551f4484ad06",
@@ -154,7 +154,11 @@
}
],
"applicability": {
- "scope": "Selected Android app/process and native diffusion boundary contracts; common is a discovery category, not CPU/device compatibility or native qualification."
+ "scope": "Selected Android app/process and native diffusion boundary contracts; common is a discovery category, not CPU/device compatibility or native qualification.",
+ "prerequisites": "Access to the pinned Android lifecycle/native/packaging contracts and identified installed artifacts; stronger conclusions require an authorized, frozen workload on the actual selected Android/ARM64 backend.",
+ "input_contract": "Source revision, installed APK/version/flavor/ABI, native executable and loaded DiT library, QNN/Hexagon/MNN builds, model/export/weights/tokenizer identities, requested and serving configuration, and lifecycle/readiness/generation/cache observations.",
+ "produced_artifact": "A bounded failed-boundary report with counter-signals, identities observed versus declared, the smallest missing discriminator and retained INCONCLUSIVE native scope.",
+ "failure_modes": "Reading UI flags or equal declared configurations as actual dispatch/readiness, treating cancellation/SSE progress as native completion, or interpreting model-directory caches as authenticated model-revision namespaces can misidentify stale state."
},
"limitations": [
"Source-derived candidate; required Android ARM64/MNN/QNN/Hexagon/model execution BLOCKED_REQUIRED_BACKEND / not_run; qualification INCONCLUSIVE.",
@@ -169,14 +173,24 @@
"references/repo-provenance.md",
"references/source-identity.json"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Compare requested and serving model/backend/dimensions/bind address through stop/reopen and health observations; never enable host mode as a diagnostic default.",
+ "Trace the actual MNN/QNN/Anima/DiT path and separate native library/ABI identity from UI runOnCpu flags and declared SDK labels.",
+ "Keep progress, cancellation, native completion, release and saved final output separate; check prompt-cache mode/version/dimensions and textual-inversion caller guards before proposing cache changes."
+ ]
},
"applicability": {
- "scope": "Selected Android app/process and native diffusion boundary contracts; common is a discovery category, not CPU/device compatibility or native qualification."
+ "scope": "Selected Android app/process and native diffusion boundary contracts; common is a discovery category, not CPU/device compatibility or native qualification.",
+ "prerequisites": "Access to the pinned Android lifecycle/native/packaging contracts and identified installed artifacts; stronger conclusions require an authorized, frozen workload on the actual selected Android/ARM64 backend.",
+ "input_contract": "Source revision, installed APK/version/flavor/ABI, native executable and loaded DiT library, QNN/Hexagon/MNN builds, model/export/weights/tokenizer identities, requested and serving configuration, and lifecycle/readiness/generation/cache observations.",
+ "produced_artifact": "A bounded failed-boundary report with counter-signals, identities observed versus declared, the smallest missing discriminator and retained INCONCLUSIVE native scope.",
+ "failure_modes": "Reading UI flags or equal declared configurations as actual dispatch/readiness, treating cancellation/SSE progress as native completion, or interpreting model-directory caches as authenticated model-revision namespaces can misidentify stale state."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Compare requested and serving model/backend/dimensions/bind address through stop/reopen and health observations; never enable host mode as a diagnostic default.",
+ "Trace the actual MNN/QNN/Anima/DiT path and separate native library/ABI identity from UI runOnCpu flags and declared SDK labels.",
+ "Keep progress, cancellation, native completion, release and saved final output separate; check prompt-cache mode/version/dimensions and textual-inversion caller guards before proposing cache changes."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mistral-rs.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mistral-rs.json
index 018149f60..2dc556cb4 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mistral-rs.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mistral-rs.json
@@ -26,8 +26,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/repositories/mistral-rs/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/repositories/mistral-rs/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/repositories/mistral-rs/LICENSE",
"src/inferrsi/data/skill_library/repositories/mistral-rs/SKILL.md",
@@ -67,7 +67,7 @@
"source_files": {
"src/inferrsi/data/skill_library/repositories/mistral-rs/LICENSE": "b40bb09c7b3b6f697a8920cf9d2cb3ff8cfcf77a0fcb6d264a92842b43c4dcae",
"src/inferrsi/data/skill_library/repositories/mistral-rs/SKILL.md": "8917b30fb86437dea7508275ccce1992194d6e0c4858d537a410bf1f900953d0",
- "src/inferrsi/data/skill_library/repositories/mistral-rs/metadata.json": "5e88e2ca64b848b8177336351e03d1fbcb5c495fd54217daa83710ef6faccff0",
+ "src/inferrsi/data/skill_library/repositories/mistral-rs/metadata.json": "3ca0ca5f3a3c9e4c3e2c7d4587940b83123cf05f4e3213b305ccdd1ae89cd679",
"src/inferrsi/data/skill_library/repositories/mistral-rs/references/apache-component-license.md": "c71d239df91726fc519c6eb72d318ec65820627232b2f796219e87dcf35d0ab4",
"src/inferrsi/data/skill_library/repositories/mistral-rs/references/paged-component-notice.md": "a1718d8c5a25f15e61bea98a158880e96faa5b7174e82c4e1470fbacb23a3559",
"src/inferrsi/data/skill_library/repositories/mistral-rs/references/repo-provenance.md": "fab87dacbe0a7298e42137cab7409f443ece0dfefba60961040031e36d96adc8",
@@ -175,7 +175,11 @@
"method": "Stable-ID bounded decode/prefill/qualification author lanes, whole-graph reconciliation, hard cases and offline helper/static/scaffold/wheel checks; compiler and actual Apple evidence not_run."
},
"applicability": {
- "scope": "Selected native Metal paged decode/prefill layout/metadata/workspace and preparation/evidence investigation. Common catalog categorization is not CPU inference or hardware certification."
+ "scope": "Selected native Metal paged decode/prefill layout/metadata/workspace and preparation/evidence investigation. Common catalog categorization is not CPU inference or hardware certification.",
+ "prerequisites": "Access to the pinned Rust/Candle/MSL decode, prefill and qualification contracts; native comparisons need the matched Rust/Apple/Candle/shader stack and separately authorized frozen references.",
+ "input_contract": "mistral.rs/Candle/native artifact revisions, Rust toolchain, Apple SDK/OS/device, actual model/cache dtype and representation, shape/strides/offsets, block table, context and execution limits.",
+ "produced_artifact": "A source/build-bound route and invariant report with actual versus declared facts, safe support result, counter-signal and smallest next native check; unavailable Metal/kernel/model evidence remains blocked.",
+ "failure_modes": "Equating element counts with physical alias/layout identity, transferring decode geometry to prefill, or treating prewarm/pipeline preparation as actual dispatch can misclassify eligibility and native correctness."
},
"limitations": [
"BLOCKED_REQUIRED_BACKEND: actual Rust/Candle/Apple Metal compilation, native tensor/cache/dispatch/parity/model/quality/service/performance evidence not_run.",
@@ -188,14 +192,24 @@
"references/repo-provenance.md",
"references/source-identity.json"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Select decode, prefill or qualification by the failed contract, then follow cache/metadata bytes from producer through view to native consumer.",
+ "Keep static predicates, pipeline preparation, kernel dispatch, numerical correctness and model/service outcomes as separate evidence steps; retain skipped or gated tests.",
+ "Use the matched Candle revision and native shader/build inputs; broad workspace tests can require model fixtures, tokens and downloads."
+ ]
},
"applicability": {
- "scope": "Selected native Metal paged decode/prefill layout/metadata/workspace and preparation/evidence investigation. Common catalog categorization is not CPU inference or hardware certification."
+ "scope": "Selected native Metal paged decode/prefill layout/metadata/workspace and preparation/evidence investigation. Common catalog categorization is not CPU inference or hardware certification.",
+ "prerequisites": "Access to the pinned Rust/Candle/MSL decode, prefill and qualification contracts; native comparisons need the matched Rust/Apple/Candle/shader stack and separately authorized frozen references.",
+ "input_contract": "mistral.rs/Candle/native artifact revisions, Rust toolchain, Apple SDK/OS/device, actual model/cache dtype and representation, shape/strides/offsets, block table, context and execution limits.",
+ "produced_artifact": "A source/build-bound route and invariant report with actual versus declared facts, safe support result, counter-signal and smallest next native check; unavailable Metal/kernel/model evidence remains blocked.",
+ "failure_modes": "Equating element counts with physical alias/layout identity, transferring decode geometry to prefill, or treating prewarm/pipeline preparation as actual dispatch can misclassify eligibility and native correctness."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Select decode, prefill or qualification by the failed contract, then follow cache/metadata bytes from producer through view to native consumer.",
+ "Keep static predicates, pipeline preparation, kernel dispatch, numerical correctness and model/service outcomes as separate evidence steps; retain skipped or gated tests.",
+ "Use the matched Candle revision and native shader/build inputs; broad workspace tests can require model fixtures, tokens and downloads."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlc-llm.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlc-llm.json
index 0d61044c8..96d525e6a 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlc-llm.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlc-llm.json
@@ -27,8 +27,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/repositories/mlc-llm/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/repositories/mlc-llm/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/repositories/mlc-llm/LICENSE",
"src/inferrsi/data/skill_library/repositories/mlc-llm/SKILL.md",
@@ -71,7 +71,7 @@
"source_files": {
"src/inferrsi/data/skill_library/repositories/mlc-llm/LICENSE": "c71d239df91726fc519c6eb72d318ec65820627232b2f796219e87dcf35d0ab4",
"src/inferrsi/data/skill_library/repositories/mlc-llm/SKILL.md": "e0fd1accb4216191c56e7e4a8074633f36d9063727c2546e0d2eed14220d8cf4",
- "src/inferrsi/data/skill_library/repositories/mlc-llm/metadata.json": "c9927e284e44aa09733424c024409ec8efc9593520418bb7d2e53ef301363681",
+ "src/inferrsi/data/skill_library/repositories/mlc-llm/metadata.json": "f20ca326ca0966e8d56b5fa698f971001a4af77370a145631e98476601610fa9",
"src/inferrsi/data/skill_library/repositories/mlc-llm/references/mlc-notice.md": "3a7dfb37bb96ce6f5a1cc0d6d4c8da3b8a35e6ec4692647ba0a37f6155c7e694",
"src/inferrsi/data/skill_library/repositories/mlc-llm/references/repo-provenance.md": "423f21e1de6d7ab2e11c0bdf540357efde025ae4535bdf76ed9770b1a59ba35a",
"src/inferrsi/data/skill_library/repositories/mlc-llm/references/repo-routing-metadata.json": "5309c27e7b60c4fe7aaef28bf6911e7e3c936b58b1a907aa8486bf0e40b4c0d0",
@@ -186,7 +186,11 @@
}
],
"applicability": {
- "scope": "Selected artifact/Apple target/pinned TVM Metal consumer and KV configuration source contracts. Common catalog categorization is not CPU inference or Apple qualification."
+ "scope": "Selected artifact/Apple target/pinned TVM Metal consumer and KV configuration source contracts. Common catalog categorization is not CPU inference or Apple qualification.",
+ "prerequisites": "Access to the pinned MLC and exact TVM fork/gitlink contracts; native checks require separately prepared matching MLC/TVM/FFI and the applicable Apple SDK/device stack.",
+ "input_contract": "MLC revision, TVM gitlink/build, model/tokenizer/quantized-parameter identity, target/host/SDK, compiled artifact/format, runtime device and the actual compiled/runtime configuration sources.",
+ "produced_artifact": "An identity and actual-versus-declared boundary report for artifact, Metal runtime or KV-limit ownership, with a negative signal and smallest missing native discriminator.",
+ "failure_modes": "Treating a schema hash as payload authentication, an opt-in WebLLM artifact contract as a native-engine ABI, serializable Metal fallback as runnable, or automatic capacity estimates as free memory can create unsupported qualification claims."
},
"limitations": [
"Source-derived candidate; original compiled CPU/native tests, compiler/live API/CLI/model/Metal/fit/quality/service/performance execution not_run; required native BLOCKED_REQUIRED_BACKEND.",
@@ -200,14 +204,24 @@
"references/repo-provenance.md",
"references/source-identity.json"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Identify the interface's actual consumer before using a valid artifact manifest as evidence of native model/task support.",
+ "Separate codegen/export from runtime factory/load, pipeline creation, dispatch and stream completion; a serializable kind=metal fallback can still be non-runnable.",
+ "Trace runtime preference, compile-time bounds, nested workspace inputs and explicit overrides separately before interpreting a KV or working-set estimate."
+ ]
},
"applicability": {
- "scope": "Selected artifact/Apple target/pinned TVM Metal consumer and KV configuration source contracts. Common catalog categorization is not CPU inference or Apple qualification."
+ "scope": "Selected artifact/Apple target/pinned TVM Metal consumer and KV configuration source contracts. Common catalog categorization is not CPU inference or Apple qualification.",
+ "prerequisites": "Access to the pinned MLC and exact TVM fork/gitlink contracts; native checks require separately prepared matching MLC/TVM/FFI and the applicable Apple SDK/device stack.",
+ "input_contract": "MLC revision, TVM gitlink/build, model/tokenizer/quantized-parameter identity, target/host/SDK, compiled artifact/format, runtime device and the actual compiled/runtime configuration sources.",
+ "produced_artifact": "An identity and actual-versus-declared boundary report for artifact, Metal runtime or KV-limit ownership, with a negative signal and smallest missing native discriminator.",
+ "failure_modes": "Treating a schema hash as payload authentication, an opt-in WebLLM artifact contract as a native-engine ABI, serializable Metal fallback as runnable, or automatic capacity estimates as free memory can create unsupported qualification claims."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Identify the interface's actual consumer before using a valid artifact manifest as evidence of native model/task support.",
+ "Separate codegen/export from runtime factory/load, pipeline creation, dispatch and stream completion; a serializable kind=metal fallback can still be non-runnable.",
+ "Trace runtime preference, compile-time bounds, nested workspace inputs and explicit overrides separately before interpreting a KV or working-set estimate."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlx-swift-lm.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlx-swift-lm.json
index 232ce428d..f6f5e3c39 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlx-swift-lm.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlx-swift-lm.json
@@ -27,8 +27,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/repositories/mlx-swift-lm/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/repositories/mlx-swift-lm/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/repositories/mlx-swift-lm/LICENSE",
"src/inferrsi/data/skill_library/repositories/mlx-swift-lm/SKILL.md",
@@ -62,7 +62,7 @@
"source_files": {
"src/inferrsi/data/skill_library/repositories/mlx-swift-lm/LICENSE": "7b1d86acac816ce8192209d7e01141c129b4d48d474c15cc5fb4af62f497c323",
"src/inferrsi/data/skill_library/repositories/mlx-swift-lm/SKILL.md": "501f73e7b327b04b6b80501b42b20dab1ac15c875817e6f4c4b66fe3e69b4fc4",
- "src/inferrsi/data/skill_library/repositories/mlx-swift-lm/metadata.json": "014131f345a28d3367313ed5a4cce858fe4c93d4d47a5f03eb89f609592e032f",
+ "src/inferrsi/data/skill_library/repositories/mlx-swift-lm/metadata.json": "6d5bd7d5f3566b75c8c0b53be95d524db039a517566826c211c978987d78b253",
"src/inferrsi/data/skill_library/repositories/mlx-swift-lm/references/repo-provenance.md": "f1b15edf89c597db0c3fc0bff7db6b2451924ce9273c39623bc0cbdccb568d31",
"src/inferrsi/data/skill_library/repositories/mlx-swift-lm/references/repo-routing-metadata.json": "8fc804347771cd768a7ed027f2041d96c9832fe5f8c5869e6593a3021b1bef8b",
"src/inferrsi/data/skill_library/repositories/mlx-swift-lm/references/scope-and-notices.md": "18e56805a8f275d35af3d8e42f3aa1525871b43400921a5d2dca4897c2047214",
@@ -175,7 +175,11 @@
"method": "Stable-ID lifecycle and recurrent-kernel author lanes, integrated root/local references, hard cases and offline support/static/scaffold/package checks; native Swift/Xcode/MLX compiler/tests not_run."
},
"applicability": {
- "scope": "Selected Swift ownership/cancellation/settling and actual GatedDelta Metal/ops contracts. Common categorization is not CPU inference or Apple qualification."
+ "scope": "Selected Swift ownership/cancellation/settling and actual GatedDelta Metal/ops contracts. Common categorization is not CPU inference or Apple qualification.",
+ "prerequisites": "Access to the Swift lifecycle and selected GatedDelta contracts, with resolved MLX/native identities for a comparison; native checks require a prepared matching Swift/Xcode/Apple stack and separate authorization.",
+ "input_contract": "Exact source revision, resolved Swift/MLX dependencies, Xcode/SDK/OS/device, API family and native artifact identity; the selected question supplies lifecycle ownership/completion observations or recurrent producer shape/dtype/layout and actual consumer/fallback.",
+ "produced_artifact": "An identity and actual-versus-declared report for generation ownership or recurrent kernel invariants, with safe support result, counter-signal and smallest separately authorized native check.",
+ "failure_modes": "Conflating lexical isolation with exclusive access across await or completed task join, treating useKernel as observed dispatch, calling an ops fallback CPU execution, or assuming a manifest range is a resolved dependency can invalidate the conclusion."
},
"limitations": [
"BLOCKED_REQUIRED_BACKEND: native Swift/MLX compiler, host-control tests and required Apple Metal cancellation/backgrounding/kernel/cache/model/fit/quality/service/performance evidence not_run.",
@@ -189,14 +193,24 @@
"references/repo-provenance.md",
"references/source-identity.json"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Resolve mutable-context/cache ownership and task settlement separately from lexical isolation and queued GPU work before cache reuse or producer shutdown.",
+ "Follow recurrent shape/dtype/layout to the actual Metal or ops consumer; predicate eligibility and useKernel alone do not prove dispatch.",
+ "Record the actual dependency resolution and use the documented Xcode macOS test route; the source states swift test does not work for this package."
+ ]
},
"applicability": {
- "scope": "Selected Swift ownership/cancellation/settling and actual GatedDelta Metal/ops contracts. Common categorization is not CPU inference or Apple qualification."
+ "scope": "Selected Swift ownership/cancellation/settling and actual GatedDelta Metal/ops contracts. Common categorization is not CPU inference or Apple qualification.",
+ "prerequisites": "Access to the Swift lifecycle and selected GatedDelta contracts, with resolved MLX/native identities for a comparison; native checks require a prepared matching Swift/Xcode/Apple stack and separate authorization.",
+ "input_contract": "Exact source revision, resolved Swift/MLX dependencies, Xcode/SDK/OS/device, API family and native artifact identity; the selected question supplies lifecycle ownership/completion observations or recurrent producer shape/dtype/layout and actual consumer/fallback.",
+ "produced_artifact": "An identity and actual-versus-declared report for generation ownership or recurrent kernel invariants, with safe support result, counter-signal and smallest separately authorized native check.",
+ "failure_modes": "Conflating lexical isolation with exclusive access across await or completed task join, treating useKernel as observed dispatch, calling an ops fallback CPU execution, or assuming a manifest range is a resolved dependency can invalidate the conclusion."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Resolve mutable-context/cache ownership and task settlement separately from lexical isolation and queued GPU work before cache reuse or producer shutdown.",
+ "Follow recurrent shape/dtype/layout to the actual Metal or ops consumer; predicate eligibility and useKernel alone do not prove dispatch.",
+ "Record the actual dependency resolution and use the documented Xcode macOS test route; the source states swift test does not work for this package."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/npu-skill-factory.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/npu-skill-factory.json
index f42d1bac1..96b5d3a7f 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/npu-skill-factory.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/npu-skill-factory.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/workflows/npu-skill-factory/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/workflows/npu-skill-factory/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/skill-index.json",
"src/inferrsi/data/skill_library/workflows/npu-skill-factory/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/optimization-search-policy.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/optimization-search-policy.json
index e6d1bfee7..e21b5eb9a 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/optimization-search-policy.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/optimization-search-policy.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/methods/optimization-search-policy/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/methods/optimization-search-policy/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/methods/optimization-search-policy/SKILL.md",
"src/inferrsi/data/skill_library/methods/optimization-search-policy/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/os-process-placement.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/os-process-placement.json
index 1ad9db926..5e8eeec65 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/os-process-placement.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/os-process-placement.json
@@ -24,8 +24,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/hardware/host/os-process-placement/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/hardware/host/os-process-placement/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/hardware/host/os-process-placement/SKILL.md",
"src/inferrsi/data/skill_library/hardware/host/os-process-placement/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/performance-ceiling-analysis.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/performance-ceiling-analysis.json
index 860ef2407..b1c24708d 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/performance-ceiling-analysis.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/performance-ceiling-analysis.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/SKILL.md",
"src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/plan-activation-residency.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/plan-activation-residency.json
index e69d8e94d..264d3b83d 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/plan-activation-residency.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/plan-activation-residency.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/methods/plan-activation-residency/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/methods/plan-activation-residency/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/methods/plan-activation-residency/SKILL.md",
"src/inferrsi/data/skill_library/methods/plan-activation-residency/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/project-maya.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/project-maya.json
index 9f2f5253d..d31f30a12 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/project-maya.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/project-maya.json
@@ -25,8 +25,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/repositories/project-maya/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/repositories/project-maya/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/repositories/project-maya/LICENSE",
"src/inferrsi/data/skill_library/repositories/project-maya/SKILL.md",
@@ -63,7 +63,7 @@
"source_files": {
"src/inferrsi/data/skill_library/repositories/project-maya/LICENSE": "574cffb7c57769a9271043ac0c2b07c5e7e7621cfb81081b901d8981e791d121",
"src/inferrsi/data/skill_library/repositories/project-maya/SKILL.md": "c619a3472be02147a847a24ca97c56afd8b56b6ffd22ce66fcae453155e34bab",
- "src/inferrsi/data/skill_library/repositories/project-maya/metadata.json": "c9f7879ca512c07e9bdf6d1d1a52de055a0fa210b8838c205e27a34436324834",
+ "src/inferrsi/data/skill_library/repositories/project-maya/metadata.json": "c48af69c430f23a8b39a2dbc09495bef54b48147c64068b39cfb3551f4e94c63",
"src/inferrsi/data/skill_library/repositories/project-maya/references/license-boundaries.md": "090839df90d6e96f55794c2ad821d7154dc6a6aa4211cdbb08a1e5f6f2e52a92",
"src/inferrsi/data/skill_library/repositories/project-maya/references/repo-provenance.md": "7bb6b6c14d1a9b77e906804505c723569f952e5d47559e636d2e6621994216bf",
"src/inferrsi/data/skill_library/repositories/project-maya/references/repo-routing-metadata.json": "0acbb111e92316e74d537200cd033f3d7d0f69972bf0953c8cea9dfb2ecb4a34",
@@ -225,7 +225,12 @@
"method": "Stable-ID isolated author lanes, integrated runtime graph, assertion/helper/static and selected direct-compiler CPU verification; required accelerator/model blocks retained."
},
"applicability": {
- "scope": "Model-specific dispatch/residency, API/native state and qualification investigation. Common catalog placement is not CPU inference or cross-device certification."
+ "scope": "Model-specific dispatch/residency, API/native state and qualification investigation. Common catalog placement is not CPU inference or cross-device certification.",
+ "prerequisites": "Access to the selected model-family runtime/state/qualification contract; native changes need frozen correctness/comparison requirements and the matching authorized CUDA/HIP/model resources.",
+ "input_contract": "Full source revision, actual executable/model family, model and asset identities, representation, compiler/runtime/device, rendered workload, sampling policy, single/parallel route, exact tokens/images and relevant completion observations.",
+ "produced_artifact": "A route/identity/evidence-class report with invariant, proposed check, actual result, counter-signal, missing resource and next authorized action; missing native requirements remain INCONCLUSIVE.",
+ "failure_modes": "Transferring coverage through a common facade or quantization name, confusing API history with native KV/private state, double-counting unified RAM/VRAM, or repairing a slot header to appear compatible can invalidate diagnosis.",
+ "refusal_point": "When required CUDA/HIP/model evidence is unavailable, retain BLOCKED_REQUIRED_BACKEND and an unverified/INCONCLUSIVE conclusion; exit-zero skips or CPU simulations cannot clear it."
},
"limitations": [
"BLOCKED_REQUIRED_BACKEND: no actual CUDA/HIP/model/device, quantized payload, vision, parallel/state/restore or cancellation-cleanup evidence.",
@@ -238,14 +243,25 @@
"references/repo-provenance.md",
"references/source-identity.json"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Select the family-specific route and preserve the explicitly shallow Qwen coverage rather than transferring deeper GLM/DeepSeek/MiniMax evidence.",
+ "Separate shared weights, private conversation state, API history and native KV; count unified physical memory once.",
+ "Preserve exact token/image identities and completion boundaries before reuse, then test one falsifiable hypothesis without modifying slot headers to bypass compatibility."
+ ]
},
"applicability": {
- "scope": "Model-specific dispatch/residency, API/native state and qualification investigation. Common catalog placement is not CPU inference or cross-device certification."
+ "scope": "Model-specific dispatch/residency, API/native state and qualification investigation. Common catalog placement is not CPU inference or cross-device certification.",
+ "prerequisites": "Access to the selected model-family runtime/state/qualification contract; native changes need frozen correctness/comparison requirements and the matching authorized CUDA/HIP/model resources.",
+ "input_contract": "Full source revision, actual executable/model family, model and asset identities, representation, compiler/runtime/device, rendered workload, sampling policy, single/parallel route, exact tokens/images and relevant completion observations.",
+ "produced_artifact": "A route/identity/evidence-class report with invariant, proposed check, actual result, counter-signal, missing resource and next authorized action; missing native requirements remain INCONCLUSIVE.",
+ "failure_modes": "Transferring coverage through a common facade or quantization name, confusing API history with native KV/private state, double-counting unified RAM/VRAM, or repairing a slot header to appear compatible can invalidate diagnosis.",
+ "refusal_point": "When required CUDA/HIP/model evidence is unavailable, retain BLOCKED_REQUIRED_BACKEND and an unverified/INCONCLUSIVE conclusion; exit-zero skips or CPU simulations cannot clear it."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Select the family-specific route and preserve the explicitly shallow Qwen coverage rather than transferring deeper GLM/DeepSeek/MiniMax evidence.",
+ "Separate shared weights, private conversation state, API history and native KV; count unified physical memory once.",
+ "Preserve exact token/image identities and completion boundaries before reuse, then test one falsifiable hypothesis without modifying slot headers to bypass compatibility."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/repository-evolution-analysis.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/repository-evolution-analysis.json
index f8da7204d..e0eb40a17 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/repository-evolution-analysis.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/repository-evolution-analysis.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/methods/repository-evolution-analysis/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/methods/repository-evolution-analysis/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/methods/repository-evolution-analysis/SKILL.md",
"src/inferrsi/data/skill_library/methods/repository-evolution-analysis/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/rsi-evidence.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/rsi-evidence.json
index e3e9c23a3..a388c6c82 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/rsi-evidence.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/rsi-evidence.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/workflows/rsi-evidence/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/workflows/rsi-evidence/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/skill-index.json",
"src/inferrsi/data/skill_library/workflows/rsi-evidence/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/skill-evolution-playbook.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/skill-evolution-playbook.json
index 379fe83b5..aa4cc2218 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/skill-evolution-playbook.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/skill-evolution-playbook.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/methods/skill-evolution-playbook/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/methods/skill-evolution-playbook/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/methods/skill-evolution-playbook/SKILL.md",
"src/inferrsi/data/skill_library/methods/skill-evolution-playbook/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/splash.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/splash.json
index 9e6612249..ea0a93eb6 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/splash.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/splash.json
@@ -25,8 +25,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/repositories/splash/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/repositories/splash/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/repositories/splash/LICENSE",
"src/inferrsi/data/skill_library/repositories/splash/SKILL.md",
@@ -61,7 +61,7 @@
"src/inferrsi/data/skill_library/repositories/splash/LICENSE": "cfc7749b96f63bd31c3c42b5c471bf756814053e847c10f3eb003417bc523d30",
"src/inferrsi/data/skill_library/repositories/splash/SKILL.md": "619b87c282dcec4befa319085764fb53fa506a05c6a7112500bcca662f059ece",
"src/inferrsi/data/skill_library/repositories/splash/THIRD_PARTY_NOTICES": "2fedf1277bf732f8435c3e3be2798f52f25f7e2bd4155865cb366da55b88e50b",
- "src/inferrsi/data/skill_library/repositories/splash/metadata.json": "d4c679f474c31549156aad417e23c2802b9f4bc904472e5cb16916f2996b50ba",
+ "src/inferrsi/data/skill_library/repositories/splash/metadata.json": "9d0613387fdaee61de7e79146e2a896f19ac7f5c1d355f74afe767074f64dd2a",
"src/inferrsi/data/skill_library/repositories/splash/references/repo-provenance.md": "20e3462f1b6a8752eb0fa6f36e76867f142a49c1953f498f895e29c4fd4f0204",
"src/inferrsi/data/skill_library/repositories/splash/references/repo-routing-metadata.json": "287ae591f42949b8d33312659ec8303c47d5c73093c64c125edb1cf6ea8de8b9",
"src/inferrsi/data/skill_library/repositories/splash/references/source-identity.json": "2f778882ec440538a46f48332650412b446722246c8dc5690083f19da7dce61d",
@@ -189,7 +189,12 @@
"method": "User-approved native Metal/C++ and isolated source-Python adaptation; stable-ID lanes, integrated graph, assertion cases and create/verify stages. Required-backend block disables automatic admission."
},
"applicability": {
- "scope": "Pinned Splash representation, planning, state/serving and verification investigation; common discovery does not qualify a different Mac/model/backend."
+ "scope": "Pinned Splash representation, planning, state/serving and verification investigation; common discovery does not qualify a different Mac/model/backend.",
+ "prerequisites": "Access to the pinned representation/planning, serving-cache or qualification route; native checks need matching Apple/backend/model resources, an authorized workload and frozen semantic/measurement controls.",
+ "input_contract": "Source/build, macOS/device, target and compatible draft artifacts/formats, tokenizer/template/vision selection, context/cache/memory controls, KV/recurrent precision, ANE and verifier-row policy, concurrency/priority/sampling, quality controls and cache state.",
+ "produced_artifact": "An execution-identity report with observation, source-backed invariant, alternate explanation, smallest discriminating check, actual result and unresolved Metal/ANE/model/disk-state gap.",
+ "failure_modes": "Treating an API alias as new weights, equal bit width as loader/kernel compatibility, or a fixed seed as fixed cache/batch/adaptive-verification scheduling can invalidate the comparison; CPU cache/codec checks do not prove GPU state.",
+ "refusal_point": "Stop at unsupported formats, unavailable backend/model resources, unsafe service exposure or the authorized budget."
},
"limitations": [
"Required Metal/ANE/model/actual device and disk-state execution remains BLOCKED_REQUIRED_BACKEND; Python policy/codec checks do not clear it.",
@@ -201,14 +206,25 @@
"references/repo-provenance.md",
"references/source-identity.json"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Resolve representation, memory/cache owner, validity key and reclaim/rollback boundary before tuning a numeric control.",
+ "Freeze semantic and workload controls beyond the seed, and retain actual-path fallbacks, admission waits, evictions, cancellations and errors.",
+ "Inspect options without starting the normal serve path, which can download target/draft weights; installation and serving require separate task authority."
+ ]
},
"applicability": {
- "scope": "Pinned Splash representation, planning, state/serving and verification investigation; common discovery does not qualify a different Mac/model/backend."
+ "scope": "Pinned Splash representation, planning, state/serving and verification investigation; common discovery does not qualify a different Mac/model/backend.",
+ "prerequisites": "Access to the pinned representation/planning, serving-cache or qualification route; native checks need matching Apple/backend/model resources, an authorized workload and frozen semantic/measurement controls.",
+ "input_contract": "Source/build, macOS/device, target and compatible draft artifacts/formats, tokenizer/template/vision selection, context/cache/memory controls, KV/recurrent precision, ANE and verifier-row policy, concurrency/priority/sampling, quality controls and cache state.",
+ "produced_artifact": "An execution-identity report with observation, source-backed invariant, alternate explanation, smallest discriminating check, actual result and unresolved Metal/ANE/model/disk-state gap.",
+ "failure_modes": "Treating an API alias as new weights, equal bit width as loader/kernel compatibility, or a fixed seed as fixed cache/batch/adaptive-verification scheduling can invalidate the comparison; CPU cache/codec checks do not prove GPU state.",
+ "refusal_point": "Stop at unsupported formats, unavailable backend/model resources, unsafe service exposure or the authorized budget."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Resolve representation, memory/cache owner, validity key and reclaim/rollback boundary before tuning a numeric control.",
+ "Freeze semantic and workload controls beyond the seed, and retain actual-path fallbacks, admission waits, evictions, cancellations and errors.",
+ "Inspect options without starting the normal serve path, which can download target/draft weights; installation and serving require separate task authority."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/storage-io-measurement.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/storage-io-measurement.json
index 9e0474379..80446779c 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/storage-io-measurement.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/storage-io-measurement.json
@@ -24,8 +24,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/hardware/host/storage-io-measurement/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/hardware/host/storage-io-measurement/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/hardware/host/storage-io-measurement/SKILL.md",
"src/inferrsi/data/skill_library/hardware/host/storage-io-measurement/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-input-pipeline.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-input-pipeline.json
index bcc6324a9..2ba4373ba 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-input-pipeline.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-input-pipeline.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/methods/training-input-pipeline/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/methods/training-input-pipeline/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/methods/training-input-pipeline/SKILL.md",
"src/inferrsi/data/skill_library/methods/training-input-pipeline/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-resource-budget.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-resource-budget.json
index 0a23ae89f..bbd183b6b 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-resource-budget.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-resource-budget.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/methods/training-resource-budget/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/methods/training-resource-budget/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/methods/training-resource-budget/SKILL.md",
"src/inferrsi/data/skill_library/methods/training-resource-budget/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/video-diffusion-optimization.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/video-diffusion-optimization.json
index a03848459..9e43ad614 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/video-diffusion-optimization.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/video-diffusion-optimization.json
@@ -26,8 +26,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/methods/video-diffusion-optimization/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/methods/video-diffusion-optimization/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/methods/video-diffusion-optimization/SKILL.md",
"src/inferrsi/data/skill_library/methods/video-diffusion-optimization/assets/campaign.template.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/vllm-metal.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/vllm-metal.json
index ae0d8bcd1..ca08d0062 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/vllm-metal.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/vllm-metal.json
@@ -26,8 +26,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/repositories/vllm-metal/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/repositories/vllm-metal/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/repositories/vllm-metal/LICENSE",
"src/inferrsi/data/skill_library/repositories/vllm-metal/SKILL.md",
@@ -65,7 +65,7 @@
"source_files": {
"src/inferrsi/data/skill_library/repositories/vllm-metal/LICENSE": "c71d239df91726fc519c6eb72d318ec65820627232b2f796219e87dcf35d0ab4",
"src/inferrsi/data/skill_library/repositories/vllm-metal/SKILL.md": "5f83b1b989e8dc239a6f34a6b3bd63872533ecc1404483a69dabfb1865648439",
- "src/inferrsi/data/skill_library/repositories/vllm-metal/metadata.json": "63d51d1ace6b0ba503c28751a8ffbf9224038395f4146a94c775c628d8075470",
+ "src/inferrsi/data/skill_library/repositories/vllm-metal/metadata.json": "55652ec6de744d92044951e25ea89aa18c92b14846dd4b597444d773bb8c31e3",
"src/inferrsi/data/skill_library/repositories/vllm-metal/references/repo-provenance.md": "970bda4d1081704aa82c94d6544a98bbf479430c4fc2c82a1a82f206043aa72a",
"src/inferrsi/data/skill_library/repositories/vllm-metal/references/repo-routing-metadata.json": "230a2b383bcf2855af4de50374c41f11b0b0841268ae45e424c1c97f679b477a",
"src/inferrsi/data/skill_library/repositories/vllm-metal/references/scope-and-notices.md": "aaef98eaea4cff56f39852e1d131163ebc01334e79034b1890f67a2660727d3e",
@@ -185,7 +185,11 @@
"method": "Stable-ID manual author sequencing plus bounded cache author lane; integrated graph followed by assertion/helper/static and selected pure Python source tests; Apple native blocks retained."
},
"applicability": {
- "scope": "Selected plugin preparation, KV/offload ownership and speculative/deferred/dispatch evidence investigation. Common catalog scope does not certify CPU inference or cross-backend qualification."
+ "scope": "Selected plugin preparation, KV/offload ownership and speculative/deferred/dispatch evidence investigation. Common catalog scope does not certify CPU inference or cross-backend qualification.",
+ "prerequisites": "Access to the plugin-specific platform/ABI, cache-lifecycle and qualification contracts; actual engine changes require separate authority, resource ownership, frozen gates and cleanup, with a supported artifact pairing.",
+ "input_contract": "Source/package/runtime/native artifact identities, Python/OS/chip/device, backend/executor, checkpoint/tokenizer/draft identities, format/dtype, effective feature flags and workload.",
+ "produced_artifact": "An identity/route/invariant report with actual safe-check result, counter-signal and next authorized native check; partial CPU evidence stays separate from unverified transfers, sleep/wake, model and HTTP outcomes.",
+ "failure_modes": "Using lazy import success as native registration or ABI evidence, treating a feature switch as dispatch, or conflating cache capacity with physical residency and kernel time with HTTP time can overstate plugin behavior."
},
"limitations": [
"BLOCKED_REQUIRED_BACKEND: actual Apple MLX/Metal and MPS kernels/model loading, cache transfers/sleep/wake and real worker dispatch are not verified.",
@@ -198,14 +202,24 @@
"references/repo-provenance.md",
"references/source-identity.json"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Resolve platform registration, backend/runner and native ABI pairing before capacity tuning; a package visibility check establishes only its reported software facts.",
+ "Keep MLX and MPS, full-attention and recurrent state, logical cache capacity and physical residency as separate contracts.",
+ "Preserve failed paths and distinguish parser/API checks from real transfers, sleep/wake, model output and HTTP performance evidence."
+ ]
},
"applicability": {
- "scope": "Selected plugin preparation, KV/offload ownership and speculative/deferred/dispatch evidence investigation. Common catalog scope does not certify CPU inference or cross-backend qualification."
+ "scope": "Selected plugin preparation, KV/offload ownership and speculative/deferred/dispatch evidence investigation. Common catalog scope does not certify CPU inference or cross-backend qualification.",
+ "prerequisites": "Access to the plugin-specific platform/ABI, cache-lifecycle and qualification contracts; actual engine changes require separate authority, resource ownership, frozen gates and cleanup, with a supported artifact pairing.",
+ "input_contract": "Source/package/runtime/native artifact identities, Python/OS/chip/device, backend/executor, checkpoint/tokenizer/draft identities, format/dtype, effective feature flags and workload.",
+ "produced_artifact": "An identity/route/invariant report with actual safe-check result, counter-signal and next authorized native check; partial CPU evidence stays separate from unverified transfers, sleep/wake, model and HTTP outcomes.",
+ "failure_modes": "Using lazy import success as native registration or ABI evidence, treating a feature switch as dispatch, or conflating cache capacity with physical residency and kernel time with HTTP time can overstate plugin behavior."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Resolve platform registration, backend/runner and native ABI pairing before capacity tuning; a package visibility check establishes only its reported software facts.",
+ "Keep MLX and MPS, full-attention and recurrent state, logical cache capacity and physical residency as separate contracts.",
+ "Preserve failed paths and distinguish parser/API checks from real transfers, sleep/wake, model output and HTTP performance evidence."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/weight-loading-optimization.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/weight-loading-optimization.json
index 56b37fd49..f9a0c015a 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/weight-loading-optimization.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/weight-loading-optimization.json
@@ -29,8 +29,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/methods/weight-loading-optimization/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/methods/weight-loading-optimization/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/methods/weight-loading-optimization/SKILL.md",
"src/inferrsi/data/skill_library/methods/weight-loading-optimization/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-afd-p2p-exchanges.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-afd-p2p-exchanges.json
index 438723ba8..0b33bbe63 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-afd-p2p-exchanges.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-afd-p2p-exchanges.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/SKILL.md",
"src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/metadata.json",
@@ -42,7 +42,7 @@
"checkout_ref": "checkout:audit-afd-p2p-exchanges",
"source_files": {
"src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/SKILL.md": "a46110891eca860915e314d6921a22856d631453b22b17837fe30262d6780535",
- "src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/metadata.json": "4e6100dfb9fc7bc00b19915f4fd38ad509704f2a609c658adf4d632d4e05b6c1",
+ "src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/metadata.json": "ae8c7582265dd6902cc4172293fc709244cc915df9390672fde25687b9e79b4b",
"src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/references/afd-plugin-LICENSE.md": "0e8566ea362276878e1f38f97da49964dbea301c15d4aa9d727f97a4c39fd943",
"src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/references/source-notes.md": "5073d87450cb60d1b245308dff8807e73749c07de976d50f2fa26694fad81a72",
"src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
@@ -92,7 +92,12 @@
}
],
"applicability": {
- "scope": "Source-based CUDA P2pNccl rank/payload audit; selected eager dense-FFN path with compute_gate_on_attention=false. Repository identity afd_plugin; runtime/backend compatibility unverified."
+ "scope": "Source-based CUDA P2pNccl rank/payload audit; selected eager dense-FFN path with compute_gate_on_attention=false. Repository identity afd_plugin; runtime/backend compatibility unverified.",
+ "prerequisites": "Access to the pinned source ledger and caller-level evidence for the selected eager dense FFN path with Attention-side gating disabled; any proposed cross-stream reuse needs an established communicator completion contract.",
+ "input_contract": "Plugin and dependency revisions, connector and role, A/F worker counts, DP/PCP/TP coordinates, worker/device ranks, subgroup roster, stage, per-peer token counts, payload shape/dtype/device, transfer context and returned slices.",
+ "produced_artifact": "A bounded rank-namespace and per-peer payload ownership ledger, the earliest evidenced mismatch or missing edge, and a proposed verification plan.",
+ "failure_modes": "Inferring data peers from control senders, accepting equal total row counts despite a peer permutation, or treating an even-split fallback or Python return as proof of ownership/completion can produce a wrong exchange diagnosis.",
+ "refusal_point": "Stop on A < F, invalid role ranks, duplicate/missing assignments or an unresolved rank namespace; return the conflicting callers and ledger entries without inventing ranks or attributing the fault to networking."
},
"limitations": [
"Source-derived candidate; native exchange, model quality, backend compatibility, performance and agent efficacy not_run.",
@@ -107,14 +112,25 @@
"../../../methods/inference-benchmark-design/SKILL.md",
"../../../methods/dense-performance-evidence/SKILL.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Record role, world and subgroup-local ranks separately; reconstruct the actual subgroup roster before interpreting send/receive peers.",
+ "Track distinct row identities and stage-specific peer lengths through concatenation and splitting; use same-total swapped lengths as a countercheck.",
+ "Establish the last communicator/stream completion obligation before reusing or releasing a receive buffer."
+ ]
},
"applicability": {
- "scope": "Source-based CUDA P2pNccl rank/payload audit; selected eager dense-FFN path with compute_gate_on_attention=false. Repository identity afd_plugin; runtime/backend compatibility unverified."
+ "scope": "Source-based CUDA P2pNccl rank/payload audit; selected eager dense-FFN path with compute_gate_on_attention=false. Repository identity afd_plugin; runtime/backend compatibility unverified.",
+ "prerequisites": "Access to the pinned source ledger and caller-level evidence for the selected eager dense FFN path with Attention-side gating disabled; any proposed cross-stream reuse needs an established communicator completion contract.",
+ "input_contract": "Plugin and dependency revisions, connector and role, A/F worker counts, DP/PCP/TP coordinates, worker/device ranks, subgroup roster, stage, per-peer token counts, payload shape/dtype/device, transfer context and returned slices.",
+ "produced_artifact": "A bounded rank-namespace and per-peer payload ownership ledger, the earliest evidenced mismatch or missing edge, and a proposed verification plan.",
+ "failure_modes": "Inferring data peers from control senders, accepting equal total row counts despite a peer permutation, or treating an even-split fallback or Python return as proof of ownership/completion can produce a wrong exchange diagnosis.",
+ "refusal_point": "Stop on A < F, invalid role ranks, duplicate/missing assignments or an unresolved rank namespace; return the conflicting callers and ledger entries without inventing ranks or attributing the fault to networking."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Record role, world and subgroup-local ranks separately; reconstruct the actual subgroup roster before interpreting send/receive peers.",
+ "Track distinct row identities and stage-specific peer lengths through concatenation and splitting; use same-total swapped lengths as a countercheck.",
+ "Establish the last communicator/stream completion obligation before reusing or releasing a receive buffer."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-flare-session-lifecycle.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-flare-session-lifecycle.json
index d9f3e6d7d..347ee07bf 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-flare-session-lifecycle.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-flare-session-lifecycle.json
@@ -29,8 +29,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/LICENSE",
"src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/SKILL.md",
@@ -52,7 +52,7 @@
"source_files": {
"src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/LICENSE": "cfc7749b96f63bd31c3c42b5c471bf756814053e847c10f3eb003417bc523d30",
"src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/SKILL.md": "02a5f7c18e44201215ecec17c6ac96b27be9148bcc189df287f39d1b372b7721",
- "src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/metadata.json": "58716737446f46bd1de77aa323c5652f97277d44dd952c82b42399b2cb769b87",
+ "src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/metadata.json": "9c667c7735ebe4fb07247e098218f6b03a1c1fafa776f1372f8d948ea8ea756b",
"src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/references/diagnostics.md": "0908a0eae53a2e10775a8b6ee3e3944b2fa410174402b5a1a8c98eb674dd4b4d",
"src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/references/lifecycle.md": "107ebe3c22e5dba8007c3b8ea61149f10bc55c51b557a4230f3e5de859bb0f7c",
"src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/references/source-ledger.json": "61e51a97ffd7c1fa7866bdd440b024d0c5e0065635ac8d0ba53d2c69c7f9e470",
@@ -126,7 +126,12 @@
"method": "AREX scope/ground/construct/verify and repository evidence/usability methods adapted to InferRSI native owners and packaging; no DisCo runtime or managed import."
},
"applicability": {
- "scope": "Pinned Flare complete-session restore and admission audit for its NVIDIA SM120/SM121 research prototype. CUDA catalog scope is content classification, not device or backend qualification."
+ "scope": "Pinned Flare complete-session restore and admission audit for its NVIDIA SM120/SM121 research prototype. CUDA catalog scope is content classification, not device or backend qualification.",
+ "prerequisites": "Access to package/source identity and bounded restore traces, with preparation compatibility established before cache-state diagnosis. Running, freeing resources or changing configuration requires separate authority.",
+ "input_contract": "Execution ABI and prepared-contract version, ordinary/speculative configuration, model/layout identity, request IDs and consumed-token endpoints, GPU/host prefixes, continuation row, ticket/waiter identities, transfer events, fences and resource counts.",
+ "produced_artifact": "A checkpoint/ticket ownership table, five-outcome admission ledger and diagnosis identifying the earliest unsupported transition and smallest missing observation.",
+ "failure_modes": "A KV-only lookup hit can omit continuation state; admission can be mistaken for actual reuse; client cancellation can be mistaken for transfer completion; mixing absolute timestamps from client and native clocks can corrupt the diagnosis.",
+ "refusal_point": "Stop with an unresolved edge when package identity, layout, transfer completion or clock provenance is unknown."
},
"limitations": [
"Source-only candidate; native Flare/model/GPU/service and skill efficacy were not evaluated.",
@@ -139,14 +144,25 @@
"references/source-notes.md",
"references/source-ledger.json"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Pair the immutable KV parent chain with the complete continuation row at the same consumed-token endpoint before calling a checkpoint restorable.",
+ "Distinguish lookup outcomes, destination reservation, ticket creation, transfer completion, publication and actual reused-token evidence.",
+ "Follow scheduler-owned tickets through waiter cancellation and check write/use fences before allowing storage reuse."
+ ]
},
"applicability": {
- "scope": "Pinned Flare complete-session restore and admission audit for its NVIDIA SM120/SM121 research prototype. CUDA catalog scope is content classification, not device or backend qualification."
+ "scope": "Pinned Flare complete-session restore and admission audit for its NVIDIA SM120/SM121 research prototype. CUDA catalog scope is content classification, not device or backend qualification.",
+ "prerequisites": "Access to package/source identity and bounded restore traces, with preparation compatibility established before cache-state diagnosis. Running, freeing resources or changing configuration requires separate authority.",
+ "input_contract": "Execution ABI and prepared-contract version, ordinary/speculative configuration, model/layout identity, request IDs and consumed-token endpoints, GPU/host prefixes, continuation row, ticket/waiter identities, transfer events, fences and resource counts.",
+ "produced_artifact": "A checkpoint/ticket ownership table, five-outcome admission ledger and diagnosis identifying the earliest unsupported transition and smallest missing observation.",
+ "failure_modes": "A KV-only lookup hit can omit continuation state; admission can be mistaken for actual reuse; client cancellation can be mistaken for transfer completion; mixing absolute timestamps from client and native clocks can corrupt the diagnosis.",
+ "refusal_point": "Stop with an unresolved edge when package identity, layout, transfer completion or clock provenance is unknown."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Pair the immutable KV parent chain with the complete continuation row at the same consumed-token endpoint before calling a checkpoint restorable.",
+ "Distinguish lookup outcomes, destination reservation, ticket creation, transfer completion, publication and actual reused-token evidence.",
+ "Follow scheduler-owned tickets through waiter cancellation and check write/use fences before allowing storage reuse."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/diagnose-attention-bottlenecks.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/diagnose-attention-bottlenecks.json
index 9eca28b37..a10f57972 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/diagnose-attention-bottlenecks.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/diagnose-attention-bottlenecks.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/hardware/cuda/diagnose-attention-bottlenecks/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/hardware/cuda/diagnose-attention-bottlenecks/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/hardware/cuda/diagnose-attention-bottlenecks/SKILL.md",
"src/inferrsi/data/skill_library/hardware/cuda/diagnose-attention-bottlenecks/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/host-overhead-profiling.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/host-overhead-profiling.json
index a470ca646..279f931d2 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/host-overhead-profiling.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/host-overhead-profiling.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/hardware/cuda/host-overhead-profiling/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/hardware/cuda/host-overhead-profiling/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/hardware/cuda/host-overhead-profiling/SKILL.md",
"src/inferrsi/data/skill_library/hardware/cuda/host-overhead-profiling/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/reason-hopper-kernels.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/reason-hopper-kernels.json
index 058605427..32f754cbd 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/reason-hopper-kernels.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/reason-hopper-kernels.json
@@ -36,8 +36,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/bde70d82bb65cc2e5dcd0f9d33cda5d38091120b/src/inferrsi/data/skill_library/hardware/cuda/reason-hopper-kernels/SKILL.md",
- "commit": "bde70d82bb65cc2e5dcd0f9d33cda5d38091120b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a04c599c49465b67ea40a1f0de622706ab3da955/src/inferrsi/data/skill_library/hardware/cuda/reason-hopper-kernels/SKILL.md",
+ "commit": "a04c599c49465b67ea40a1f0de622706ab3da955",
"paths": [
"src/inferrsi/data/skill_library/hardware/cuda/reason-hopper-kernels/SKILL.md",
"src/inferrsi/data/skill_library/hardware/cuda/reason-hopper-kernels/metadata.json",
From 7b8995dcfa9a40029c16fbc892ed2cca1ea218db Mon Sep 17 00:00:00 2001
From: Dong1017 <55723162+Dong1017@users.noreply.github.com>
Date: Sat, 31 Oct 2026 12:19:40 +0800
Subject: [PATCH 03/11] ci: qualify the new knowledge payload without rewriting
historical evidence
Keep model-free qualification on a separately frozen current payload. Test framework-query contribution independently of incidental corpus ranking.
Agent-assisted-by: Codex / gpt-6-astra [effort=xhigh] (author)
Signed-off-by: Dong1017 <55723162+Dong1017@users.noreply.github.com>
---
.github/workflows/ci-ready.yml | 2 +-
.../kv-actor-qualification-20261031/README.md | 27 ++++++++++++
.../patch/README.md | 7 +++
.../plan/integrity.json | 7 +++
.../plan/qualification-plan.json | 42 ++++++++++++++++++
.../results/README.md | 6 +++
.../validator/README.md | 16 +++++++
tests/check_installed.py | 4 +-
tests/test_evidence_inventory.py | 4 +-
tests/test_qualify_kv_edit_actor.py | 44 ++++++++++++++++++-
tests/test_research.py | 12 ++++-
tools/evidence_inventory.json | 5 +++
12 files changed, 169 insertions(+), 7 deletions(-)
create mode 100644 evidence/kv-actor-qualification-20261031/README.md
create mode 100644 evidence/kv-actor-qualification-20261031/patch/README.md
create mode 100644 evidence/kv-actor-qualification-20261031/plan/integrity.json
create mode 100644 evidence/kv-actor-qualification-20261031/plan/qualification-plan.json
create mode 100644 evidence/kv-actor-qualification-20261031/results/README.md
create mode 100644 evidence/kv-actor-qualification-20261031/validator/README.md
diff --git a/.github/workflows/ci-ready.yml b/.github/workflows/ci-ready.yml
index 62b8ee333..4d2831a40 100644
--- a/.github/workflows/ci-ready.yml
+++ b/.github/workflows/ci-ready.yml
@@ -92,7 +92,7 @@ jobs:
run: |
env -i PATH=/usr/bin:/bin LANG=C.UTF-8 PYTHONPATH=src \
/usr/bin/python3 -m tools.qualify_kv_edit_actor \
- --plan evidence/kv-actor-qualification-20261030/plan/qualification-plan.json \
+ --plan evidence/kv-actor-qualification-20261031/plan/qualification-plan.json \
--revision "$GITHUB_SHA" --output "$RUNNER_TEMP/kv-actor-qualification"
- name: Retain separate scripted actor receipt including failures
if: always()
diff --git a/evidence/kv-actor-qualification-20261031/README.md b/evidence/kv-actor-qualification-20261031/README.md
new file mode 100644
index 000000000..770600da6
--- /dev/null
+++ b/evidence/kv-actor-qualification-20261031/README.md
@@ -0,0 +1,27 @@
+# Scripted KV actor qualification for the updated K payload
+
+This prospective software-control plan qualifies the complete K owner at
+`a04c599c49465b67ea40a1f0de622706ab3da955`. Batch 4 added source-supported
+applicability and recommendations to `metadata.json`; the procedure and source
+notes retain their previous bytes. Metadata is part of the actor's exposed
+knowledge, so this is a new payload identity even though its evidence and
+candidate maturity did not change.
+
+The [2026-10-30 plan](../kv-actor-qualification-20261030/plan/qualification-plan.json),
+its integrity manifest and retained receipts remain unchanged. They describe
+the earlier K bytes and cannot qualify this payload by inheritance. This plan
+retains the same complete three-file owner, task-plan digest, 1,800-lease
+workload, limits, scripted arms and qualification requirements. The existing
+qualifier still rejects changed bytes, expanded exposure and unavailable
+isolation; no production qualifier logic is changed.
+
+Ready must produce a fresh receipt and raw archive for the exact clean submitted
+revision, this plan and its actual A2/A3 exposure. Until that run completes,
+qualification remains not_run/INCONCLUSIVE. Existing tests and earlier receipts
+are not a hosted result for this version.
+
+This plan makes zero provider calls and changes no live authorization, actor
+manifest, charge identity, reservation or ledger. A future live plan must still
+match its separately authorized exposure and hosted qualification exactly.
+Scripted software qualification supplies no knowledge-efficacy, model-quality
+or production-performance result.
diff --git a/evidence/kv-actor-qualification-20261031/patch/README.md b/evidence/kv-actor-qualification-20261031/patch/README.md
new file mode 100644
index 000000000..ae41d6ded
--- /dev/null
+++ b/evidence/kv-actor-qualification-20261031/patch/README.md
@@ -0,0 +1,7 @@
+# Payload change
+
+The complete K owner retains its procedure and source notes. Its metadata adds
+source-supported applicability and owner advice at the plan's knowledge revision.
+The current Ready job and software tests select this new prospective plan.
+The earlier qualification bundle and all production qualifier, task-verifier,
+live-authorization and reservation logic remain unchanged.
diff --git a/evidence/kv-actor-qualification-20261031/plan/integrity.json b/evidence/kv-actor-qualification-20261031/plan/integrity.json
new file mode 100644
index 000000000..ff7785e8f
--- /dev/null
+++ b/evidence/kv-actor-qualification-20261031/plan/integrity.json
@@ -0,0 +1,7 @@
+{
+ "README.md": "4d2d85cc4afa8d44809d01fa34b2c7af588e0c819a5d435466aa79915231d93b",
+ "patch/README.md": "eca4ad73b58e1d924042f34ced0aabc858a04012f47145c164bb5031a91e02e8",
+ "plan/qualification-plan.json": "b21e414dcefbbb221607c0ddb9e90d943f40ad0266bfbf271818dc2e5601bd95",
+ "results/README.md": "8b2137b0cddb25c5aa8c33d107b6c288e2bf887e82c7ea9090891c2c2d1c0d39",
+ "validator/README.md": "9c635cd59e6707a5b34ebbc47d47a755d78ab8b9228d60813630a147f1263f17"
+}
diff --git a/evidence/kv-actor-qualification-20261031/plan/qualification-plan.json b/evidence/kv-actor-qualification-20261031/plan/qualification-plan.json
new file mode 100644
index 000000000..b2a9f1c71
--- /dev/null
+++ b/evidence/kv-actor-qualification-20261031/plan/qualification-plan.json
@@ -0,0 +1,42 @@
+{
+ "schema": "kv-actor-model-free-qualification-v1",
+ "scope": "Scripted actor/workbench/isolation qualification only; zero real model calls; no K efficacy result",
+ "accepted_scope": "https://github.com/InferMatrixLab/InferRSI/issues/248#issuecomment-6526000390",
+ "base_revision": "57bc805db9bb56995e2ae988830e1c52a14f85d3",
+ "task_plan_sha256": "60cc479686aa2ba3772f627cc5179a3c0dc6d215c0fdeb57547fd58dc30859ed",
+ "knowledge_revision": "a04c599c49465b67ea40a1f0de622706ab3da955",
+ "knowledge_root": "src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes",
+ "knowledge_files_sha256": {
+ "SKILL.md": "4c3f27459df239f9f4fc971c559d4c9ddfbfebd70507a8a93be8eae59a30b69d",
+ "metadata.json": "8ded6e60c562131616dc8cdcf0d7efc04aef1884885da0b30fb16f42d7433b7c",
+ "references/source-notes.md": "3665b034594fc8cc767c408002f4e16f8803369242405027c7f37ce6c9ecebe9"
+ },
+ "qualification_max_seconds": 480,
+ "descendant_probe_seconds": 2,
+ "visible_workload_count": 1800,
+ "visible_min_improvement_pct": 0,
+ "actor_limits": {
+ "max_source_bytes": 65536,
+ "max_feedback_bytes": 8192,
+ "max_writes": 8,
+ "max_checks": 8,
+ "max_benchmarks": 4,
+ "benchmark_repetitions": 2,
+ "max_seconds": 600
+ },
+ "client": {
+ "kind": "anthropic-messages-edit-check-v1",
+ "provider": "yibuapi",
+ "endpoint": "https://yibuapi.com/v1/messages",
+ "api_version": "2023-06-01",
+ "max_requests": 8,
+ "max_output_tokens": 4096,
+ "max_request_bytes": 262144,
+ "max_read_bytes": 65536,
+ "max_tool_calls": 32
+ },
+ "model": "claude-opus-5-5",
+ "input_token_ceiling": 1000000,
+ "retry_policy": "No automatic rerun; retain every qualification attempt separately",
+ "paid_execution": "Absent. Scripted response fixtures are not provider evidence or spending authority."
+}
diff --git a/evidence/kv-actor-qualification-20261031/results/README.md b/evidence/kv-actor-qualification-20261031/results/README.md
new file mode 100644
index 000000000..25f0151f5
--- /dev/null
+++ b/evidence/kv-actor-qualification-20261031/results/README.md
@@ -0,0 +1,6 @@
+# Results
+
+No hosted result for this prospective payload version is committed here. Ready
+retains exact-head qualification receipts and privacy-redacted raw archives as
+separate run/attempt artifacts, including failures. Earlier payload receipts and
+local unit tests do not qualify this version or establish knowledge efficacy.
diff --git a/evidence/kv-actor-qualification-20261031/validator/README.md b/evidence/kv-actor-qualification-20261031/validator/README.md
new file mode 100644
index 000000000..0bc1141fe
--- /dev/null
+++ b/evidence/kv-actor-qualification-20261031/validator/README.md
@@ -0,0 +1,16 @@
+# Unchanged software qualification contract
+
+The existing `tools/qualify_kv_edit_actor.py` validates the complete regular-file
+K owner and every pinned file digest. It reuses the independently frozen task
+plan, verifier and benchmark-output supplement under
+`evidence/kv-reclamation-20261030/` without changing their bytes or gates.
+
+A fresh hosted receipt must bind the exact clean revision, dependency closure,
+new plan digest, complete A2/A3 exposure and common actor contracts. It must
+retain both edit/check/measurement revisions, immutable final grading, runtime
+isolation and descendant cleanup. Timing gains are not software-control gates;
+unavailable requirements remain INCONCLUSIVE.
+
+The current plan is not an update to any live actor or paid execution approval.
+Live admission still requires its separately frozen manifests, matching hosted
+exposure, explicit authorization and existing reservation checks.
diff --git a/tests/check_installed.py b/tests/check_installed.py
index de58c3e09..e15715a06 100644
--- a/tests/check_installed.py
+++ b/tests/check_installed.py
@@ -293,8 +293,10 @@
assert research["profile"]["model"] == "MiniMax-H3"
assert research["target_validation"] == "unverified"
assert research["target"]["device"] == "M5 Mac Studio"
+framework_first = search_knowledge("mlx", limit=1)[0]["id"]
assert any(
- item["record"]["id"] == "inferrsi.common.skills.lazy_evaluation_stream_lifetimes"
+ item["record"]["id"] == framework_first
+ and {"group": "framework", "query": "mlx"} in item["matched_queries"]
for item in research["items"]
)
assert {item["group"] for item in research["queries"]} == {
diff --git a/tests/test_evidence_inventory.py b/tests/test_evidence_inventory.py
index 93adacf95..bb7747e48 100644
--- a/tests/test_evidence_inventory.py
+++ b/tests/test_evidence_inventory.py
@@ -145,6 +145,6 @@ def test_checked_in_inventory():
repo = Path(__file__).resolve().parents[1]
result = validate_inventory(repo)
assert result["errors"] == []
- assert len(result["bundles"]) == 18
- assert result["checked_hashes"] == 362
+ assert len(result["bundles"]) == 19
+ assert result["checked_hashes"] == 367
assert result["unavailable"] # Historical raw artifacts are not in this checkout.
diff --git a/tests/test_qualify_kv_edit_actor.py b/tests/test_qualify_kv_edit_actor.py
index 39497e612..5d77e5f31 100644
--- a/tests/test_qualify_kv_edit_actor.py
+++ b/tests/test_qualify_kv_edit_actor.py
@@ -9,7 +9,48 @@
from inferrsi.model import anthropic_edit_agent as edit
from tools import qualify_kv_edit_actor as qualify
-PLAN = qualify.ROOT / "evidence/kv-actor-qualification-20261030/plan/qualification-plan.json"
+PLAN = qualify.ROOT / "evidence/kv-actor-qualification-20261031/plan/qualification-plan.json"
+HISTORICAL = qualify.ROOT / "evidence/kv-actor-qualification-20261030"
+
+
+def test_historical_qualification_bundle_is_retained_byte_for_byte():
+ integrity = HISTORICAL / "plan/integrity.json"
+ assert qualify.task.file_hash(integrity) == (
+ "d6c018da4cf1cde265d9dce36cb4368d75953d2291acb1ef48ddb67d09d89336"
+ )
+ for name, digest in json.loads(integrity.read_text()).items():
+ assert qualify.task.file_hash(HISTORICAL / name) == digest
+
+
+def test_new_payload_plan_preserves_the_software_contract():
+ current = qualify.checked_plan(PLAN)
+ historical = json.loads((HISTORICAL / "plan/qualification-plan.json").read_text())
+ identity_fields = {"base_revision", "knowledge_revision", "knowledge_files_sha256"}
+ assert {k: v for k, v in current.items() if k not in identity_fields} == {
+ k: v for k, v in historical.items() if k not in identity_fields
+ }
+ assert current["knowledge_revision"] == "a04c599c49465b67ea40a1f0de622706ab3da955"
+ changed = {
+ name for name, digest in current["knowledge_files_sha256"].items()
+ if historical["knowledge_files_sha256"][name] != digest
+ }
+ assert changed == {"metadata.json"}
+
+
+def test_current_plan_rejects_metadata_drift(tmp_path, monkeypatch):
+ plan = qualify.checked_plan(PLAN)
+ owner = tmp_path / plan["knowledge_root"]
+ for name in plan["knowledge_files_sha256"]:
+ source = qualify.ROOT / plan["knowledge_root"] / name
+ target = owner / name
+ target.parent.mkdir(parents=True, exist_ok=True)
+ target.write_bytes(source.read_bytes())
+ monkeypatch.setattr(qualify, "ROOT", tmp_path)
+ qualify.checked_plan(PLAN)
+ metadata = owner / "metadata.json"
+ metadata.write_bytes(metadata.read_bytes() + b"\n")
+ with pytest.raises(ValueError, match="Pinned K bytes changed"):
+ qualify.checked_plan(PLAN)
def identity():
@@ -150,6 +191,7 @@ def test_existing_ready_job_only_adds_secret_free_separate_receipt():
assert job["runs-on"] == "ubuntu-22.04" and job["permissions"] == {"contents": "read"}
commands = "\n".join(step.get("run", "") for step in job["steps"])
assert "tools.qualify_kv_edit_actor" in commands and "env -i" in commands
+ assert f"--plan {PLAN.relative_to(qualify.ROOT).as_posix()}" in commands
assert "secrets." not in commands and "send_message" not in commands
assert "--privileged" not in commands and "sysctl" not in commands
uploads = [s["with"] for s in job["steps"] if "upload-artifact" in s.get("uses", "")]
diff --git a/tests/test_research.py b/tests/test_research.py
index f1dfa0951..3da4ba48e 100644
--- a/tests/test_research.py
+++ b/tests/test_research.py
@@ -27,7 +27,13 @@ def test_h3_retrieves_framework_and_task_knowledge_with_source_boundaries():
items = result["items"]
identities = [item["record"]["id"] for item in items]
assert len(identities) == len(set(identities)) <= 12
- assert "inferrsi.common.skills.lazy_evaluation_stream_lifetimes" in identities
+ # Corpus additions can reorder matches; the independent framework query still contributes.
+ framework_first = search_knowledge("mlx", limit=1)[0]["id"]
+ assert any(
+ item["record"]["id"] == framework_first
+ and {"group": "framework", "query": "mlx"} in item["matched_queries"]
+ for item in items
+ )
assert set(result["profile"]["knowledge_ids"]) <= set(identities)
assert any("diffusion" in item["record"]["scenario"] for item in items)
assert any(
@@ -169,8 +175,10 @@ def test_public_cli_exposes_model_profile_and_keeps_filters(capsys):
assert result["profile"]["sources"]
assert result["target"]["device"] == "M5 Mac Studio"
assert result["target_validation"] == "unverified"
+ framework_first = search_knowledge("mlx", limit=1)[0]["id"]
assert any(
- item["record"]["id"] == "inferrsi.common.skills.lazy_evaluation_stream_lifetimes"
+ item["record"]["id"] == framework_first
+ and {"group": "framework", "query": "mlx"} in item["matched_queries"]
for item in result["items"]
)
assert main([
diff --git a/tools/evidence_inventory.json b/tools/evidence_inventory.json
index 0b0676692..57d691efb 100644
--- a/tools/evidence_inventory.json
+++ b/tools/evidence_inventory.json
@@ -1,6 +1,11 @@
{
"version": 1,
"bundles": {
+ "kv-actor-qualification-20261031": {
+ "kind": "campaign",
+ "manifests": [{"path": "plan/integrity.json", "schema": "bundle-sha256"}],
+ "external": ["Fresh exact-revision hosted actor receipts for the updated K payload are retained as separate CI artifacts; this plan makes no provider calls and grants no live authorization."]
+ },
"kv-actor-qualification-20261030": {
"kind": "campaign",
"manifests": [{"path": "plan/integrity.json", "schema": "bundle-sha256"}],
From 40db800a96a27356192d0946b91ff5cc264f0859 Mon Sep 17 00:00:00 2001
From: Dong1017 <55723162+Dong1017@users.noreply.github.com>
Date: Sat, 31 Oct 2026 12:27:04 +0800
Subject: [PATCH 04/11] docs: regenerate skill projections from published
source history
Keep the canonical source snapshot reachable as a preceding commit.
Agent-assisted-by: Codex / gpt-6-astra [effort=xhigh] (author)
Signed-off-by: Dong1017 <55723162+Dong1017@users.noreply.github.com>
---
docs/skill-coverage.md | 62 +++++++++----------
.../repo_packs/inferrsi/knowledge/README.md | 4 +-
.../ascend-aclgraph-capture-replay.json | 4 +-
.../ascend/skills/ascend-device-health.json | 4 +-
.../skills/ascend-distributed-parallel.json | 4 +-
.../ascend/skills/ascend-graph-mode.json | 4 +-
.../skills/ascend-model-adaptation.json | 4 +-
.../skills/ascend-operator-profiling.json | 4 +-
.../ascend/skills/ascend-operator-tuning.json | 4 +-
.../skills/ascend-optimization-casebook.json | 4 +-
.../skills/ascend-performance-analysis.json | 4 +-
.../ascend-video-diffusion-optimization.json | 4 +-
.../ascend/skills/ascendc-tiling-design.json | 4 +-
.../skills/cann-attribution-overlap.json | 4 +-
.../torchnpu-device-memory-lifecycle.json | 4 +-
...torchnpu-pinned-host-memory-lifecycle.json | 4 +-
.../skills/assess-serving-preemption.json | 4 +-
.../assess-split-backward-schedule.json | 4 +-
.../skills/audit-fixed-workload-claim.json | 4 +-
.../audit-h3-metal-fasth3-semantics.json | 32 +++++++---
.../skills/audit-h3-runtime-residency.json | 32 +++++++---
.../skills/audit-llama-runtime-state.json | 32 +++++++---
.../audit-recurrent-kv-finalization.json | 32 +++++++---
.../common/skills/audit-sd-cpp-residency.json | 32 +++++++---
.../audit-sglang-diffusion-lifetimes.json | 32 +++++++---
.../skills/audit-vllm-kv-lifetimes.json | 35 ++++++++---
.../bound-windowed-attention-replay.json | 30 ++++++---
.../budget-speculative-verification.json | 30 ++++++---
.../skills/calibrate-serving-simulation.json | 4 +-
.../skills/cgroup-resource-controls.json | 4 +-
.../checkpoint-recovery-consistency.json | 4 +-
.../skills/choose-parallel-training-plan.json | 4 +-
.../skills/cite-check-downstream-pr.json | 4 +-
.../common/skills/claim-contract.json | 4 +-
.../knowledge/common/skills/comfy-ui.json | 32 +++++++---
.../common/skills/competitor-scan.json | 4 +-
.../skills/dcp-async-save-performance.json | 35 ++++++++---
.../skills/dense-performance-evidence.json | 4 +-
.../diagnose-vllm-admission-pressure.json | 35 ++++++++---
.../skills/diffusers-invocation-state.json | 32 +++++++---
.../common/skills/draft-acceptance-issue.json | 4 +-
.../knowledge/common/skills/dwarfstar.json | 32 +++++++---
.../evaluate-elastic-serving-placement.json | 4 +-
.../skills/generation-cache-lifecycle.json | 32 +++++++---
.../common/skills/host-memory-management.json | 4 +-
.../skills/host-overhead-optimization.json | 4 +-
.../knowledge/common/skills/husky.json | 32 +++++++---
.../skills/inference-benchmark-design.json | 4 +-
.../inference-optimization-casebook.json | 4 +-
.../common/skills/kernel-writing-guide.json | 4 +-
.../common/skills/kv-tier-management.json | 33 +++++++---
.../lazy-evaluation-stream-lifetimes.json | 32 +++++++---
.../knowledge/common/skills/lily.json | 32 +++++++---
.../knowledge/common/skills/lithos-metal.json | 30 ++++++---
.../knowledge/common/skills/local-dream.json | 30 ++++++---
.../knowledge/common/skills/mistral-rs.json | 30 ++++++---
.../knowledge/common/skills/mlc-llm.json | 30 ++++++---
.../knowledge/common/skills/mlx-swift-lm.json | 30 ++++++---
.../common/skills/npu-skill-factory.json | 4 +-
.../skills/optimization-search-policy.json | 4 +-
.../common/skills/os-process-placement.json | 4 +-
.../skills/performance-ceiling-analysis.json | 4 +-
.../skills/plan-activation-residency.json | 4 +-
.../knowledge/common/skills/project-maya.json | 32 +++++++---
.../skills/repository-evolution-analysis.json | 4 +-
.../knowledge/common/skills/rsi-evidence.json | 4 +-
.../skills/skill-evolution-playbook.json | 4 +-
.../knowledge/common/skills/splash.json | 32 +++++++---
.../common/skills/storage-io-measurement.json | 4 +-
.../skills/training-input-pipeline.json | 4 +-
.../skills/training-resource-budget.json | 4 +-
.../skills/video-diffusion-optimization.json | 4 +-
.../knowledge/common/skills/vllm-metal.json | 30 ++++++---
.../skills/weight-loading-optimization.json | 4 +-
.../cuda/skills/audit-afd-p2p-exchanges.json | 32 +++++++---
.../skills/audit-flare-session-lifecycle.json | 32 +++++++---
.../diagnose-attention-bottlenecks.json | 4 +-
.../cuda/skills/host-overhead-profiling.json | 4 +-
.../cuda/skills/reason-hopper-kernels.json | 4 +-
79 files changed, 823 insertions(+), 357 deletions(-)
diff --git a/docs/skill-coverage.md b/docs/skill-coverage.md
index 7261443c3..9d36d9eab 100644
--- a/docs/skill-coverage.md
+++ b/docs/skill-coverage.md
@@ -4,7 +4,7 @@ Generated from the registry, the skill sidecars and the two maintained
matrices. **Do not edit by hand**; regenerate after any registry, skill,
matrix or frozen-plan change.
-- Input digest: `sha256:bf6ac4fe724e1f91f7373ed7352852919749aa9cb1e53e8d27e68e3e1df4b8ff`
+- Input digest: `sha256:8409aafa0406d6443db8b39586410559ad676d141afd43cb0d527e328d608a16`
- Regenerate: `python tools/build_skill_coverage.py`
- Verify without writing: `python tools/build_skill_coverage.py --check`
@@ -79,40 +79,40 @@ optimization is fully covered or that any procedure has measured benefit.
| Skill | Backend | Declared task | Applicability | Conditions | SearchSpace layer / stage | Reference Matrix | State | Gaps |
|---|---|---|---|---|---|---|---|---|
-| [dcp-async-save-performance](../src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/SKILL.md) | common | Diagnose training-visible checkpoint pauses using PyTorch Distributed Checkpoint asynchronous saving and staging ... | **unknown** | — | 1 Accelerator · Diagnose / Configure / Optimize
2 RAM/Host · Observe / Profile / Configure / Optimize / Verify
3 SSD · Observe / Profile / Configure / Optimize / Verify | 1 · Systems sources (C1 / DCP async-save-only approval — pytorch/pytorch @ ...) | candidate · not_run · 1 readset(s) | no-applicability, no-eval |
-| [diffusers-invocation-state](../src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/SKILL.md) | common | Trace Diffusers pipeline-wrapper and scheduler invocation state when repeated calls, from_pipe reuse or interruption ... | FluxPipeline and FlowMatchEulerDiscreteScheduler invocation ownership at this source pin ... | — | 2 RAM/Host †guidance: Diffusers invocation guidance (source-derived) | Diffusers invocation source extraction (huggingface/diffusers @ `aea7b70be0f594bcdc15a6f2677ee2ca85e49e8e` ...) | candidate · not_run · 1 readset(s) | no-eval |
-| [lazy-evaluation-stream-lifetimes](../src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/SKILL.md) | common | Diagnose MLX lazy evaluation, cross-stream ordering and array-buffer lifetimes when timing, memory use or asynchronous ... | MLX evaluation/stream/buffer diagnosis at the pinned implementation; qualify the actual ... | — | 1 Accelerator †guidance: MLX framework guidance (source-derived) | MLX framework source extraction (ml-explore/mlx @ `bff10220c742907f0c68e55ff174aaea56f74e3d` (MIT)) | candidate · not_run · 1 readset(s) | no-eval |
+| [dcp-async-save-performance](../src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/SKILL.md) | common | Diagnose training-visible checkpoint pauses using PyTorch Distributed Checkpoint asynchronous saving and staging ... | Source-supported diagnosis of training-visible PyTorch DCP save pauses ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator · Diagnose / Configure / Optimize
2 RAM/Host · Observe / Profile / Configure / Optimize / Verify
3 SSD · Observe / Profile / Configure / Optimize / Verify | 1 · Systems sources (C1 / DCP async-save-only approval — pytorch/pytorch @ ...) | candidate · not_run · 1 readset(s) | no-eval |
+| [diffusers-invocation-state](../src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/SKILL.md) | common | Trace Diffusers pipeline-wrapper and scheduler invocation state when repeated calls, from_pipe reuse or interruption ... | FluxPipeline and FlowMatchEulerDiscreteScheduler invocation ownership at this source pin ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 2 RAM/Host †guidance: Diffusers invocation guidance (source-derived) | Diffusers invocation source extraction (huggingface/diffusers @ `aea7b70be0f594bcdc15a6f2677ee2ca85e49e8e` ...) | candidate · not_run · 1 readset(s) | no-eval |
+| [lazy-evaluation-stream-lifetimes](../src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/SKILL.md) | common | Diagnose MLX lazy evaluation, cross-stream ordering and array-buffer lifetimes when timing, memory use or asynchronous ... | MLX evaluation/stream/buffer diagnosis at the pinned implementation; qualify the actual ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator †guidance: MLX framework guidance (source-derived) | MLX framework source extraction (ml-explore/mlx @ `bff10220c742907f0c68e55ff174aaea56f74e3d` (MIT)) | candidate · not_run · 1 readset(s) | no-eval |
## Registered: Repositories (26)
| Skill | Backend | Declared task | Applicability | Conditions | SearchSpace layer / stage | Reference Matrix | State | Gaps |
|---|---|---|---|---|---|---|---|---|
-| [audit-afd-p2p-exchanges](../src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/SKILL.md) | cuda | Audit CUDA P2pNccl Attention/FFN rank namespaces and per-stage, per-peer hidden-state ownership when uneven worker ... | Source-based CUDA P2pNccl rank/payload audit; selected eager dense-FFN path with ... | — | 1 Accelerator · Diagnose / Configure / Verify | 1 · Systems sources (vllm-project/afd-plugin @ `df8d6a0c403c3ca1b4ed9cf1687e7204908bebec` ...) | candidate · not_run · 1 readset(s) | no-eval |
-| [audit-flare-session-lifecycle](../src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/SKILL.md) | cuda | Audit Flare complete prefix/session restoration, inbound transfer tickets, publication/use/retirement fences and ... | Pinned Flare complete-session restore and admission audit for its NVIDIA SM120/SM121 ... | — | 1 Accelerator †guidance: Flare preparation and complete-session source contracts
2 RAM/Host †guidance: Flare preparation and complete-session source contracts
3 SSD †guidance: Flare preparation and complete-session source contracts | Flare preparation and complete-session source knowledge (NVIDIA/flare-local-inference → [NVIDIA/local-inference](https://github ...) | candidate · not_run · 2 readset(s) | no-eval |
-| [audit-h3-metal-fasth3-semantics](../src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/SKILL.md) | common | Audit h3-metal FastH3 effective configuration, checkpoint tensor and schedule semantics, partial VSA tiles, sparse ... | h3-metal FastH3 requested/effective configuration, checkpoint transformation and ... | — | 1 Accelerator · Observe / Diagnose / Verify | 1 · Systems sources (zuiho-kai/h3-metal @ `4ba47a3eabdbebf801ac6664b31f8ec168bc57a8` ...) | candidate · not_run · 3 readset(s) | no-eval |
-| [audit-h3-runtime-residency](../src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/SKILL.md) | common | Audit MiniMax-H3 generation in h3.c on Apple Silicon by tracing conditioning, AdaLN precompute, DiT residency, Metal ... | MiniMax-H3 joint audio/video C/Metal runtime audit: raw checkpoint compatibility, input ... | — | 1 Accelerator · Observe / Diagnose / Verify
2 RAM/Host · Observe / Diagnose / Verify
3 SSD · Diagnose / Configure / Verify | 1 · Systems sources (antirez/h3.c @ `8974cc055ea9c02fcd14cc27dfda3e1027c05153` (MIT ...) | candidate · not_run · 3 readset(s) | no-eval |
-| [audit-llama-runtime-state](../src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/SKILL.md) | common | Audit llama.cpp context memory, partial decode progress, graph reuse and server cancellation when a retry, stale ... | Pinned llama.cpp context-memory, graph-reuse and server task-lifecycle diagnosis ... | — | 1 Accelerator · Diagnose / Verify
2 RAM/Host · Diagnose / Verify | 1 · Systems sources (ggml-org/llama.cpp @ `707616feb5f379e99ff7442680a55b4a1704c7c9` (MIT ...) | candidate · not_run · 1 readset(s) | no-eval |
-| [audit-recurrent-kv-finalization](../src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/SKILL.md) | common | Audit vLLM-RLT synchronous early-exit KV propagation, later depth-specific attention reads, and request storage ... | Pinned vLLM-RLT native Ouro synchronous scheduling with host KVMetadata: LAST_EXITED ... | — | — | 1 · Systems sources (InferMatrixLab/vllm-rlt (legacy ThinkFlowLab/vllm-rlt) @ ...) | candidate · not_run · 2 readset(s) | no-SearchSpace-layer, no-eval |
-| [audit-sd-cpp-residency](../src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/SKILL.md) | common | Audit stable-diffusion.cpp image-generation memory pressure by separating catalog freshness and weight-byte delivery ... | Source-grounded audit of native image-generation weight/runtime residency, optional ... | — | 1 Accelerator · Observe / Diagnose / Verify
2 RAM/Host · Observe / Diagnose / Verify
3 SSD · Diagnose / Verify | 1 · Systems sources (leejet/stable-diffusion.cpp @ `b13a48219d5cdf408edcb93da6cd69b612ed03d ...) | candidate · not_run · 2 readset(s) | no-eval |
-| [audit-sglang-diffusion-lifetimes](../src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/SKILL.md) | common | Diagnose SGLang multimodal_gen stage state, component residency and request timeout/deletion when grouped outputs ... | Bounded monolithic multimodal_gen composed-pipeline state and lifecycle diagnosis; actual ... | — | 1 Accelerator · Diagnose / Verify
2 RAM/Host · Diagnose / Verify | 1 · Systems sources (sgl-project/sglang @ `df13d5dce31abfeff0e701a62e8c39a136758e6c` ...) | candidate · not_run · 1 readset(s) | no-eval |
-| [audit-vllm-kv-lifetimes](../src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/SKILL.md) | common | Audit vLLM V1 asynchronous KV load/save completion, block retention, preemption, and receive-failure contracts when ... | **unknown** | — | 1 Accelerator · Observe / Diagnose / Verify | 1 · Systems sources (vllm-project/vllm @ `7f730b4cea9a66520a9200265928ebe016b48a39` ...) | candidate · not_run · 1 readset(s) | no-applicability, no-eval |
-| [bound-windowed-attention-replay](../src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/SKILL.md) | common | Bound replay or prefill query rows across a producer/consumer boundary whose retained inputs and separately preserved ... | Decoders with an explicit producer/consumer boundary and a demonstrated dependency ... | — | 1 Accelerator †guidance: DeepInfer bounded source guidance
2 RAM/Host †guidance: DeepInfer bounded source guidance | DeepInfer source extraction (deepseek-ai/DeepInfer @ `8b16149741a4935e1dec8cea27853f273b374c5b` ...) | candidate · not_run · 2 readset(s) | no-eval |
-| [budget-speculative-verification](../src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/SKILL.md) | common | Decide how many draft tokens to verify per decode step when acceptance varies by request and position. Use when an ... | Engines with multi-token speculative decoding whose verification forward cost grows with ... | — | 1 Accelerator †guidance: DeepInfer bounded source guidance | DeepInfer source extraction (deepseek-ai/DeepInfer @ `8b16149741a4935e1dec8cea27853f273b374c5b` ...) | candidate · not_run · 2 readset(s) | no-eval |
-| [comfy-ui](../src/inferrsi/data/skill_library/repositories/comfy-ui/SKILL.md) | common | Audit ComfyUI workflow-output cache validity, lazy dependencies and model-weight residency as separate state contracts ... | Source-grounded ComfyUI workflow-output cache and model-weight ownership audit; confirm ... | — | 1 Accelerator · Diagnose / Configure / Verify
2 RAM/Host · Diagnose / Configure / Verify | 1 · Systems sources (Comfy-Org/ComfyUI @ `b1a247c9cf9dbfda4b2c0a0c1d50950667eeaa0c` (GNU ...) | candidate · not_run · 1 readset(s) | no-eval |
-| [diagnose-vllm-admission-pressure](../src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/SKILL.md) | common | Diagnose vLLM V1 admission stalls and KV-pressure preemption by separating token, input, active-request ... | **unknown** | — | 1 Accelerator · Observe / Diagnose / Configure / Optimize / Verify | 1 · Systems sources (vllm-project/vllm @ `7f730b4cea9a66520a9200265928ebe016b48a39` ...) | candidate · not_run · 14 readset(s) | no-applicability, no-eval |
-| [dwarfstar](../src/inferrsi/data/skill_library/repositories/dwarfstar/SKILL.md) | common | Investigate DwarfStar model fit, expert SSD streaming, session and vision cache correctness, speculative sampling and ... | Pinned DwarfStar source investigation and experiment planning; not arbitrary GGUF support ... | — | — | DwarfStar portable AREX extraction (antirez/ds4 @ `ee7e8405232c8971000cc7ba2b945b00bf6d20d3`; AREX-Skill ...) | candidate · partial_cpu_only · 2 readset(s) | no-SearchSpace-layer, no-eval |
-| [generation-cache-lifecycle](../src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/SKILL.md) | common | Trace MLX-LM prefill, generation batching, prompt-cache reuse and cooperative stopping when request state, cache rows ... | MLX-LM generation, batch-row and prefix-cache diagnosis; model/backend/runtime ... | — | 2 RAM/Host †guidance: MLX-LM application guidance (source-derived) | MLX-LM application source extraction (ml-explore/mlx-lm @ `ab7f7048dc341ced2d84b1887eeb17d60b0d1b02` (MIT)) | candidate · not_run · 1 readset(s) | no-eval |
-| [husky](../src/inferrsi/data/skill_library/repositories/husky/SKILL.md) | common | Review Husky's official-document claims as research hypotheses, separating author reports, implementation evidence ... | Review Husky document claims and specify missing evidence; common catalog scope is not ... | — | — | Husky document-only research adaptation (Official https://husky.underdog.ai/; displayed2026-09-20 ...) | candidate · not_run · 2 readset(s) | no-SearchSpace-layer, no-eval |
-| [kv-tier-management](../src/inferrsi/data/skill_library/repositories/lmcache/kv-tier-management/SKILL.md) | common | Manage runtime KV cache tiers (GPU → CPU RAM → local disk) for LLM serving: decide when KV is stored, retained ... | **unknown** | — | 2 RAM/Host · Configure / Optimize
3 SSD · Optimize | 1 · Systems sources (LMCache/LMCache @ c3b875a) | candidate · not_run · 0 readset(s) | no-applicability, no-eval |
-| [lily](../src/inferrsi/data/skill_library/repositories/lily/SKILL.md) | common | Investigate Lily's exact Qwen3.6-35B-A3B MLX-affine-Q4 contract, Metal state and kernels, strict greedy text API ... | Exact Lily source/checkpoint/layout/API/cache/qualification investigation; common catalog ... | — | — | Lily portable AREX extraction (perplexityai/pplx-garden `lily/` @ `1ed972ed3f0bd5616c997c9507c25616c6 ...) | candidate · not_run · 2 readset(s) | no-SearchSpace-layer, no-eval |
-| [lithos-metal](../src/inferrsi/data/skill_library/repositories/lithos-metal/SKILL.md) | common | Inspect Lithos Metal compilation and fusion, GPU-resident generation and DSpark state, or local serving and ... | Pinned Lithos Metal 0.1.2 source/API contracts; Apple silicon Metal runtime; model and ... | — | 1 Accelerator · Diagnose / Configure / Optimize / Verify | 1 · Systems sources (lithos-ai/lithos-metal @ `6653b7df5f2c4392ec7b038960f2796b1499923f` ...) | candidate · not_run · 1 readset(s) | no-eval |
-| [local-dream](../src/inferrsi/data/skill_library/repositories/local-dream/SKILL.md) | common | Audit Local Dream Android desired-versus-serving process state, remote exposure, native backend and DiT packaging/ABI ... | Selected Android app/process and native diffusion boundary contracts; common is a ... | — | 2 RAM/Host †guidance: Local Dream Android diffusion source contracts | Local Dream Android diffusion boundary source (xororz/local-dream @ `db162df4d9bb6efc6595250423d8df26241f4c27` ...) | candidate · not_run · 2 readset(s) | no-eval |
-| [mistral-rs](../src/inferrsi/data/skill_library/repositories/mistral-rs/SKILL.md) | common | Investigate mistral.rs Metal paged-attention decode layouts, prefill coordinates and workspace, actual-model prewarm ... | Selected native Metal paged decode/prefill layout/metadata/workspace and ... | — | 1 Accelerator †guidance: mistral.rs Metal paged-attention source contracts
2 RAM/Host †guidance: mistral.rs Metal paged-attention source contracts | mistral.rs Metal paged-attention extraction (EricLBuehler/mistral.rs @ `9459891e7697238225885518119682ac8209a612` ...) | candidate · not_run · 2 readset(s) | no-eval |
-| [mlc-llm](../src/inferrsi/data/skill_library/repositories/mlc-llm/SKILL.md) | common | Investigate MLC LLM compiled Metal artifact/schema, Apple target and pinned TVM module/storage/stream contracts, and ... | Selected artifact/Apple target/pinned TVM Metal consumer and KV configuration source ... | — | 1 Accelerator †guidance: MLC LLM artifact, target and runtime source contracts
2 RAM/Host †guidance: MLC LLM artifact, target and runtime source contracts | MLC LLM compiled Metal and runtime contracts (mlc-ai/mlc-llm @ `8978ea9aec626d64096aaa92fbc89bda2e3ea802`; TVM ...) | candidate · not_run · 2 readset(s) | no-eval |
-| [mlx-swift-lm](../src/inferrsi/data/skill_library/repositories/mlx-swift-lm/SKILL.md) | common | Investigate MLX Swift LM generation ownership, cancellation and settled task completion, and recurrent GatedDelta Metal ... | Selected Swift ownership/cancellation/settling and actual GatedDelta Metal/ops contracts ... | — | 1 Accelerator †guidance: MLX Swift LM ownership and recurrent Metal source contracts
2 RAM/Host †guidance: MLX Swift LM ownership and recurrent Metal source contracts | MLX Swift LM lifecycle and recurrent Metal extraction (ml-explore/mlx-swift-lm @ `3d066d3a0ffdc7e70c476a3d31488ea62dfdd0ac` ...) | candidate · not_run · 2 readset(s) | no-eval |
-| [project-maya](../src/inferrsi/data/skill_library/repositories/project-maya/SKILL.md) | common | Investigate Project Maya model-family dispatch, representation and memory tiers, API versus native state, speculative ... | Model-specific dispatch/residency, API/native state and qualification investigation ... | — | 1 Accelerator †guidance: Project Maya model-specific source contracts
2 RAM/Host †guidance: Project Maya model-specific source contracts
3 SSD †guidance: Project Maya model-specific source contracts | Project Maya source and CPU-slice extraction (mw00/project-maya @ `249f9a87d1a9d72bec2b684d704b0aa9296b1c9c` ...) | candidate · partial_cpu_only · 2 readset(s) | no-eval |
-| [splash](../src/inferrsi/data/skill_library/repositories/splash/SKILL.md) | common | Investigate Splash model-format compatibility, automatic memory planning, Metal and ANE execution, DFlash2 ... | Pinned Splash representation, planning, state/serving and verification investigation ... | — | — | Splash portable AREX extraction (incoai/splash @ `39a97385f8560cf8a93972cec9b0a963c9195250`; AREX ...) | candidate · partial_cpu_only · 2 readset(s) | no-SearchSpace-layer, no-eval |
-| [vllm-metal](../src/inferrsi/data/skill_library/repositories/vllm-metal/SKILL.md) | common | Investigate vllm-metal Apple inference through platform and native-ABI selection, paged KV/offload ownership ... | Selected plugin preparation, KV/offload ownership and speculative/deferred/dispatch ... | — | 1 Accelerator †guidance: vLLM Metal selected source contracts
2 RAM/Host †guidance: vLLM Metal selected source contracts
3 SSD †guidance: vLLM Metal selected source contracts | vLLM Metal portable AREX source and CPU extraction (vllm-project/vllm-metal @ `e7df7a6ff215f99c7842552d230fdbc709835934` ...) | candidate · partial_cpu_only · 2 readset(s) | no-eval |
+| [audit-afd-p2p-exchanges](../src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/SKILL.md) | cuda | Audit CUDA P2pNccl Attention/FFN rank namespaces and per-stage, per-peer hidden-state ownership when uneven worker ... | Source-based CUDA P2pNccl rank/payload audit; selected eager dense-FFN path with ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator · Diagnose / Configure / Verify | 1 · Systems sources (vllm-project/afd-plugin @ `df8d6a0c403c3ca1b4ed9cf1687e7204908bebec` ...) | candidate · not_run · 1 readset(s) | no-eval |
+| [audit-flare-session-lifecycle](../src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/SKILL.md) | cuda | Audit Flare complete prefix/session restoration, inbound transfer tickets, publication/use/retirement fences and ... | Pinned Flare complete-session restore and admission audit for its NVIDIA SM120/SM121 ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator †guidance: Flare preparation and complete-session source contracts
2 RAM/Host †guidance: Flare preparation and complete-session source contracts
3 SSD †guidance: Flare preparation and complete-session source contracts | Flare preparation and complete-session source knowledge (NVIDIA/flare-local-inference → [NVIDIA/local-inference](https://github ...) | candidate · not_run · 2 readset(s) | no-eval |
+| [audit-h3-metal-fasth3-semantics](../src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/SKILL.md) | common | Audit h3-metal FastH3 effective configuration, checkpoint tensor and schedule semantics, partial VSA tiles, sparse ... | h3-metal FastH3 requested/effective configuration, checkpoint transformation and ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator · Observe / Diagnose / Verify | 1 · Systems sources (zuiho-kai/h3-metal @ `4ba47a3eabdbebf801ac6664b31f8ec168bc57a8` ...) | candidate · not_run · 3 readset(s) | no-eval |
+| [audit-h3-runtime-residency](../src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/SKILL.md) | common | Audit MiniMax-H3 generation in h3.c on Apple Silicon by tracing conditioning, AdaLN precompute, DiT residency, Metal ... | MiniMax-H3 joint audio/video C/Metal runtime audit: raw checkpoint compatibility, input ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator · Observe / Diagnose / Verify
2 RAM/Host · Observe / Diagnose / Verify
3 SSD · Diagnose / Configure / Verify | 1 · Systems sources (antirez/h3.c @ `8974cc055ea9c02fcd14cc27dfda3e1027c05153` (MIT ...) | candidate · not_run · 3 readset(s) | no-eval |
+| [audit-llama-runtime-state](../src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/SKILL.md) | common | Audit llama.cpp context memory, partial decode progress, graph reuse and server cancellation when a retry, stale ... | Pinned llama.cpp context-memory, graph-reuse and server task-lifecycle diagnosis ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator · Diagnose / Verify
2 RAM/Host · Diagnose / Verify | 1 · Systems sources (ggml-org/llama.cpp @ `707616feb5f379e99ff7442680a55b4a1704c7c9` (MIT ...) | candidate · not_run · 1 readset(s) | no-eval |
+| [audit-recurrent-kv-finalization](../src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/SKILL.md) | common | Audit vLLM-RLT synchronous early-exit KV propagation, later depth-specific attention reads, and request storage ... | Pinned vLLM-RLT native Ouro synchronous scheduling with host KVMetadata: LAST_EXITED ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | — | 1 · Systems sources (InferMatrixLab/vllm-rlt (legacy ThinkFlowLab/vllm-rlt) @ ...) | candidate · not_run · 2 readset(s) | no-SearchSpace-layer, no-eval |
+| [audit-sd-cpp-residency](../src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/SKILL.md) | common | Audit stable-diffusion.cpp image-generation memory pressure by separating catalog freshness and weight-byte delivery ... | Source-grounded audit of native image-generation weight/runtime residency, optional ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator · Observe / Diagnose / Verify
2 RAM/Host · Observe / Diagnose / Verify
3 SSD · Diagnose / Verify | 1 · Systems sources (leejet/stable-diffusion.cpp @ `b13a48219d5cdf408edcb93da6cd69b612ed03d ...) | candidate · not_run · 2 readset(s) | no-eval |
+| [audit-sglang-diffusion-lifetimes](../src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/SKILL.md) | common | Diagnose SGLang multimodal_gen stage state, component residency and request timeout/deletion when grouped outputs ... | Bounded monolithic multimodal_gen composed-pipeline state and lifecycle diagnosis; actual ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator · Diagnose / Verify
2 RAM/Host · Diagnose / Verify | 1 · Systems sources (sgl-project/sglang @ `df13d5dce31abfeff0e701a62e8c39a136758e6c` ...) | candidate · not_run · 1 readset(s) | no-eval |
+| [audit-vllm-kv-lifetimes](../src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/SKILL.md) | common | Audit vLLM V1 asynchronous KV load/save completion, block retention, preemption, and receive-failure contracts when ... | Pinned vLLM V1 scheduler/connector ownership audit for asynchronous KV load/save ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator · Observe / Diagnose / Verify | 1 · Systems sources (vllm-project/vllm @ `7f730b4cea9a66520a9200265928ebe016b48a39` ...) | candidate · not_run · 1 readset(s) | no-eval |
+| [bound-windowed-attention-replay](../src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/SKILL.md) | common | Bound replay or prefill query rows across a producer/consumer boundary whose retained inputs and separately preserved ... | Decoders with an explicit producer/consumer boundary and a demonstrated dependency ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator †guidance: DeepInfer bounded source guidance
2 RAM/Host †guidance: DeepInfer bounded source guidance | DeepInfer source extraction (deepseek-ai/DeepInfer @ `8b16149741a4935e1dec8cea27853f273b374c5b` ...) | candidate · not_run · 2 readset(s) | no-eval |
+| [budget-speculative-verification](../src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/SKILL.md) | common | Decide how many draft tokens to verify per decode step when acceptance varies by request and position. Use when an ... | Engines with multi-token speculative decoding whose verification forward cost grows with ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator †guidance: DeepInfer bounded source guidance | DeepInfer source extraction (deepseek-ai/DeepInfer @ `8b16149741a4935e1dec8cea27853f273b374c5b` ...) | candidate · not_run · 2 readset(s) | no-eval |
+| [comfy-ui](../src/inferrsi/data/skill_library/repositories/comfy-ui/SKILL.md) | common | Audit ComfyUI workflow-output cache validity, lazy dependencies and model-weight residency as separate state contracts ... | Source-grounded ComfyUI workflow-output cache and model-weight ownership audit; confirm ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator · Diagnose / Configure / Verify
2 RAM/Host · Diagnose / Configure / Verify | 1 · Systems sources (Comfy-Org/ComfyUI @ `b1a247c9cf9dbfda4b2c0a0c1d50950667eeaa0c` (GNU ...) | candidate · not_run · 1 readset(s) | no-eval |
+| [diagnose-vllm-admission-pressure](../src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/SKILL.md) | common | Diagnose vLLM V1 admission stalls and KV-pressure preemption by separating token, input, active-request ... | Pinned vLLM V1 branch-based diagnosis of queued requests, low scheduled-token counts and ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator · Observe / Diagnose / Configure / Optimize / Verify | 1 · Systems sources (vllm-project/vllm @ `7f730b4cea9a66520a9200265928ebe016b48a39` ...) | candidate · not_run · 14 readset(s) | no-eval |
+| [dwarfstar](../src/inferrsi/data/skill_library/repositories/dwarfstar/SKILL.md) | common | Investigate DwarfStar model fit, expert SSD streaming, session and vision cache correctness, speculative sampling and ... | Pinned DwarfStar source investigation and experiment planning; not arbitrary GGUF support ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | — | DwarfStar portable AREX extraction (antirez/ds4 @ `ee7e8405232c8971000cc7ba2b945b00bf6d20d3`; AREX-Skill ...) | candidate · partial_cpu_only · 2 readset(s) | no-SearchSpace-layer, no-eval |
+| [generation-cache-lifecycle](../src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/SKILL.md) | common | Trace MLX-LM prefill, generation batching, prompt-cache reuse and cooperative stopping when request state, cache rows ... | MLX-LM generation, batch-row and prefix-cache diagnosis; model/backend/runtime ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 2 RAM/Host †guidance: MLX-LM application guidance (source-derived) | MLX-LM application source extraction (ml-explore/mlx-lm @ `ab7f7048dc341ced2d84b1887eeb17d60b0d1b02` (MIT)) | candidate · not_run · 1 readset(s) | no-eval |
+| [husky](../src/inferrsi/data/skill_library/repositories/husky/SKILL.md) | common | Review Husky's official-document claims as research hypotheses, separating author reports, implementation evidence ... | Review Husky document claims and specify missing evidence; common catalog scope is not ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | — | Husky document-only research adaptation (Official https://husky.underdog.ai/; displayed2026-09-20 ...) | candidate · not_run · 2 readset(s) | no-SearchSpace-layer, no-eval |
+| [kv-tier-management](../src/inferrsi/data/skill_library/repositories/lmcache/kv-tier-management/SKILL.md) | common | Manage runtime KV cache tiers (GPU → CPU RAM → local disk) for LLM serving: decide when KV is stored, retained ... | LMCache multiprocess runtime KV tier policy across engine GPU cache, host L1 and ... | failure_modes, input_contract, produced_artifact
3 advice | 2 RAM/Host · Configure / Optimize
3 SSD · Optimize | 1 · Systems sources (LMCache/LMCache @ c3b875a) | candidate · not_run · 0 readset(s) | no-eval |
+| [lily](../src/inferrsi/data/skill_library/repositories/lily/SKILL.md) | common | Investigate Lily's exact Qwen3.6-35B-A3B MLX-affine-Q4 contract, Metal state and kernels, strict greedy text API ... | Exact Lily source/checkpoint/layout/API/cache/qualification investigation; common catalog ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | — | Lily portable AREX extraction (perplexityai/pplx-garden `lily/` @ `1ed972ed3f0bd5616c997c9507c25616c6 ...) | candidate · not_run · 2 readset(s) | no-SearchSpace-layer, no-eval |
+| [lithos-metal](../src/inferrsi/data/skill_library/repositories/lithos-metal/SKILL.md) | common | Inspect Lithos Metal compilation and fusion, GPU-resident generation and DSpark state, or local serving and ... | Pinned Lithos Metal 0.1.2 source/API contracts; Apple silicon Metal runtime; model and ... | failure_modes, input_contract, produced_artifact
3 advice | 1 Accelerator · Diagnose / Configure / Optimize / Verify | 1 · Systems sources (lithos-ai/lithos-metal @ `6653b7df5f2c4392ec7b038960f2796b1499923f` ...) | candidate · not_run · 1 readset(s) | no-eval |
+| [local-dream](../src/inferrsi/data/skill_library/repositories/local-dream/SKILL.md) | common | Audit Local Dream Android desired-versus-serving process state, remote exposure, native backend and DiT packaging/ABI ... | Selected Android app/process and native diffusion boundary contracts; common is a ... | failure_modes, input_contract, produced_artifact
3 advice | 2 RAM/Host †guidance: Local Dream Android diffusion source contracts | Local Dream Android diffusion boundary source (xororz/local-dream @ `db162df4d9bb6efc6595250423d8df26241f4c27` ...) | candidate · not_run · 2 readset(s) | no-eval |
+| [mistral-rs](../src/inferrsi/data/skill_library/repositories/mistral-rs/SKILL.md) | common | Investigate mistral.rs Metal paged-attention decode layouts, prefill coordinates and workspace, actual-model prewarm ... | Selected native Metal paged decode/prefill layout/metadata/workspace and ... | failure_modes, input_contract, produced_artifact
3 advice | 1 Accelerator †guidance: mistral.rs Metal paged-attention source contracts
2 RAM/Host †guidance: mistral.rs Metal paged-attention source contracts | mistral.rs Metal paged-attention extraction (EricLBuehler/mistral.rs @ `9459891e7697238225885518119682ac8209a612` ...) | candidate · not_run · 2 readset(s) | no-eval |
+| [mlc-llm](../src/inferrsi/data/skill_library/repositories/mlc-llm/SKILL.md) | common | Investigate MLC LLM compiled Metal artifact/schema, Apple target and pinned TVM module/storage/stream contracts, and ... | Selected artifact/Apple target/pinned TVM Metal consumer and KV configuration source ... | failure_modes, input_contract, produced_artifact
3 advice | 1 Accelerator †guidance: MLC LLM artifact, target and runtime source contracts
2 RAM/Host †guidance: MLC LLM artifact, target and runtime source contracts | MLC LLM compiled Metal and runtime contracts (mlc-ai/mlc-llm @ `8978ea9aec626d64096aaa92fbc89bda2e3ea802`; TVM ...) | candidate · not_run · 2 readset(s) | no-eval |
+| [mlx-swift-lm](../src/inferrsi/data/skill_library/repositories/mlx-swift-lm/SKILL.md) | common | Investigate MLX Swift LM generation ownership, cancellation and settled task completion, and recurrent GatedDelta Metal ... | Selected Swift ownership/cancellation/settling and actual GatedDelta Metal/ops contracts ... | failure_modes, input_contract, produced_artifact
3 advice | 1 Accelerator †guidance: MLX Swift LM ownership and recurrent Metal source contracts
2 RAM/Host †guidance: MLX Swift LM ownership and recurrent Metal source contracts | MLX Swift LM lifecycle and recurrent Metal extraction (ml-explore/mlx-swift-lm @ `3d066d3a0ffdc7e70c476a3d31488ea62dfdd0ac` ...) | candidate · not_run · 2 readset(s) | no-eval |
+| [project-maya](../src/inferrsi/data/skill_library/repositories/project-maya/SKILL.md) | common | Investigate Project Maya model-family dispatch, representation and memory tiers, API versus native state, speculative ... | Model-specific dispatch/residency, API/native state and qualification investigation ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator †guidance: Project Maya model-specific source contracts
2 RAM/Host †guidance: Project Maya model-specific source contracts
3 SSD †guidance: Project Maya model-specific source contracts | Project Maya source and CPU-slice extraction (mw00/project-maya @ `249f9a87d1a9d72bec2b684d704b0aa9296b1c9c` ...) | candidate · partial_cpu_only · 2 readset(s) | no-eval |
+| [splash](../src/inferrsi/data/skill_library/repositories/splash/SKILL.md) | common | Investigate Splash model-format compatibility, automatic memory planning, Metal and ANE execution, DFlash2 ... | Pinned Splash representation, planning, state/serving and verification investigation ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | — | Splash portable AREX extraction (incoai/splash @ `39a97385f8560cf8a93972cec9b0a963c9195250`; AREX ...) | candidate · partial_cpu_only · 2 readset(s) | no-SearchSpace-layer, no-eval |
+| [vllm-metal](../src/inferrsi/data/skill_library/repositories/vllm-metal/SKILL.md) | common | Investigate vllm-metal Apple inference through platform and native-ABI selection, paged KV/offload ownership ... | Selected plugin preparation, KV/offload ownership and speculative/deferred/dispatch ... | failure_modes, input_contract, produced_artifact
3 advice | 1 Accelerator †guidance: vLLM Metal selected source contracts
2 RAM/Host †guidance: vLLM Metal selected source contracts
3 SSD †guidance: vLLM Metal selected source contracts | vLLM Metal portable AREX source and CPU extraction (vllm-project/vllm-metal @ `e7df7a6ff215f99c7842552d230fdbc709835934` ...) | candidate · partial_cpu_only · 2 readset(s) | no-eval |
## Registered: Methods (20)
@@ -221,7 +221,7 @@ an entrypoint routes to procedures below it rather than adding one.
| `no-SearchSpace-layer` | 15: ascend-video-diffusion-optimization, audit-fixed-workload-claim, audit-recurrent-kv-finalization, cite-check-downstream-pr, competitor-scan, draft-acceptance-issue, dwarfstar, husky, inference-benchmark-design, lily, npu-skill-factory, repository-evolution-analysis, skill-evolution-playbook, splash, video-diffusion-optimization |
| `no-Reference-Matrix-row` | 8: ascend-video-diffusion-optimization, audit-fixed-workload-claim, cite-check-downstream-pr, claim-contract, competitor-scan, draft-acceptance-issue, rsi-evidence, video-diffusion-optimization |
| `no-sidecar` | 0: — |
-| `no-applicability` | 41: as listed in the Gaps column above |
+| `no-applicability` | 37: as listed in the Gaps column above |
| `no-eval` | 75: as listed in the Gaps column above |
These are inventory gaps, not defects. A workflow skill may legitimately
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/README.md b/src/inferrsi/data/repo_packs/inferrsi/knowledge/README.md
index 50464fea2..8ef93522c 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/README.md
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/README.md
@@ -15,7 +15,7 @@ These are knowledge abstracts (`--kind knowledge`), with namespaced package
IDs distinct from explicitly selected `checkout_ref` skills. Ascend/CUDA
procedures use their corresponding backend; shared methods use `common`.
-Source snapshot: [c5e13944c06a8a03ab7015a8933981a983e42186](https://github.com/InferMatrixLab/InferRSI/tree/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library).
+Source snapshot: [e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb](https://github.com/InferMatrixLab/InferRSI/tree/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library).
```bash
inferrsi knowledge search 'profiling' --repo-type inferrsi --limit 50
@@ -29,7 +29,7 @@ contain the current input paths and bytes (Git checkout filters are respected).
```bash
python tools/build_skill_index.py --source-revision
-python tools/build_skill_index.py --source-revision c5e13944c06a8a03ab7015a8933981a983e42186 --check
+python tools/build_skill_index.py --source-revision e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb --check
```
`--check` detects missing, stale and unexpected generated records without writing.
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-aclgraph-capture-replay.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-aclgraph-capture-replay.json
index 15c7f3ff8..c4aad3d20 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-aclgraph-capture-replay.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-aclgraph-capture-replay.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/hardware/ascend/ascend-aclgraph-capture-replay/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/hardware/ascend/ascend-aclgraph-capture-replay/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-aclgraph-capture-replay/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-aclgraph-capture-replay/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-device-health.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-device-health.json
index 93260a6e7..e9a1baecc 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-device-health.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-device-health.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/hardware/ascend/ascend-device-health/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/hardware/ascend/ascend-device-health/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-device-health/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-device-health/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-distributed-parallel.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-distributed-parallel.json
index ef68579bd..b6cb1ac4c 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-distributed-parallel.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-distributed-parallel.json
@@ -27,8 +27,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/hardware/ascend/ascend-distributed-parallel/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/hardware/ascend/ascend-distributed-parallel/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-distributed-parallel/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-distributed-parallel/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-graph-mode.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-graph-mode.json
index 36f92b57d..fbed32757 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-graph-mode.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-graph-mode.json
@@ -26,8 +26,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/hardware/ascend/ascend-graph-mode/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/hardware/ascend/ascend-graph-mode/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-graph-mode/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-graph-mode/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-model-adaptation.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-model-adaptation.json
index a0d5c8c59..11716cf6b 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-model-adaptation.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-model-adaptation.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/hardware/ascend/ascend-model-adaptation/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/hardware/ascend/ascend-model-adaptation/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-model-adaptation/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-model-adaptation/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-profiling.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-profiling.json
index 5f746c46d..9f9b62917 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-profiling.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-profiling.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-profiling/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-profiling/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-profiling/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-profiling/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-tuning.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-tuning.json
index f8ceae7c8..acf92eaaa 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-tuning.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-tuning.json
@@ -24,8 +24,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-tuning/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-tuning/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-tuning/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-tuning/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-optimization-casebook.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-optimization-casebook.json
index 92c99dd63..c670f72a1 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-optimization-casebook.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-optimization-casebook.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/hardware/ascend/ascend-optimization-casebook/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/hardware/ascend/ascend-optimization-casebook/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-optimization-casebook/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-optimization-casebook/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-performance-analysis.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-performance-analysis.json
index 2b70e686b..6dee31309 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-performance-analysis.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-performance-analysis.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/hardware/ascend/ascend-performance-analysis/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/hardware/ascend/ascend-performance-analysis/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-performance-analysis/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-performance-analysis/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-video-diffusion-optimization.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-video-diffusion-optimization.json
index 20c058d35..8795d5a9c 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-video-diffusion-optimization.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-video-diffusion-optimization.json
@@ -26,8 +26,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/hardware/ascend/ascend-video-diffusion-optimization/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/hardware/ascend/ascend-video-diffusion-optimization/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-video-diffusion-optimization/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-video-diffusion-optimization/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascendc-tiling-design.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascendc-tiling-design.json
index 5be88df8b..6723825ea 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascendc-tiling-design.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascendc-tiling-design.json
@@ -24,8 +24,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/hardware/ascend/ascendc-tiling-design/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/hardware/ascend/ascendc-tiling-design/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascendc-tiling-design/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascendc-tiling-design/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/cann-attribution-overlap.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/cann-attribution-overlap.json
index 3aa47ab27..71542827f 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/cann-attribution-overlap.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/cann-attribution-overlap.json
@@ -24,8 +24,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/hardware/ascend/cann-attribution-overlap/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/hardware/ascend/cann-attribution-overlap/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/cann-attribution-overlap/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/cann-attribution-overlap/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-device-memory-lifecycle.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-device-memory-lifecycle.json
index 9b074d7fa..0cc17230b 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-device-memory-lifecycle.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-device-memory-lifecycle.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/hardware/ascend/torchnpu-device-memory-lifecycle/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/hardware/ascend/torchnpu-device-memory-lifecycle/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/torchnpu-device-memory-lifecycle/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/torchnpu-device-memory-lifecycle/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-pinned-host-memory-lifecycle.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-pinned-host-memory-lifecycle.json
index 6fde6678f..92fb282ca 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-pinned-host-memory-lifecycle.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-pinned-host-memory-lifecycle.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/hardware/ascend/torchnpu-pinned-host-memory-lifecycle/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/hardware/ascend/torchnpu-pinned-host-memory-lifecycle/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/torchnpu-pinned-host-memory-lifecycle/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/torchnpu-pinned-host-memory-lifecycle/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-serving-preemption.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-serving-preemption.json
index 721060d94..66594aec7 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-serving-preemption.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-serving-preemption.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/methods/assess-serving-preemption/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/methods/assess-serving-preemption/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/methods/assess-serving-preemption/SKILL.md",
"src/inferrsi/data/skill_library/methods/assess-serving-preemption/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-split-backward-schedule.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-split-backward-schedule.json
index 7756c1301..d312f891d 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-split-backward-schedule.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-split-backward-schedule.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/methods/assess-split-backward-schedule/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/methods/assess-split-backward-schedule/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/methods/assess-split-backward-schedule/SKILL.md",
"src/inferrsi/data/skill_library/methods/assess-split-backward-schedule/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-fixed-workload-claim.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-fixed-workload-claim.json
index a5357f135..d1fdfe4a2 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-fixed-workload-claim.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-fixed-workload-claim.json
@@ -20,8 +20,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/methods/audit-fixed-workload-claim/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/methods/audit-fixed-workload-claim/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/methods/audit-fixed-workload-claim/SKILL.md",
"src/inferrsi/data/skill_library/methods/audit-fixed-workload-claim/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-metal-fasth3-semantics.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-metal-fasth3-semantics.json
index ecbb2d324..1755f4066 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-metal-fasth3-semantics.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-metal-fasth3-semantics.json
@@ -29,8 +29,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/LICENSE",
"src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/SKILL.md",
@@ -56,7 +56,7 @@
"source_files": {
"src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/LICENSE": "663e51edcd9a3a8f858018ce6924ad2895a24c269c26c39ad75325bcd0f7e5bc",
"src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/SKILL.md": "e5205de6eba52b3d1693666af755640697717e52ee7834e1ae30e4071b208d54",
- "src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/metadata.json": "86a67bb56110c3a9e191e0ff576194162e960cbddca0a99c593ea7aef8df8a4e",
+ "src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/metadata.json": "f1681981db4f9077ee27ff71571dceecd69daf25136bd374b7d9ad3b59762a6a",
"src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/references/THIRD_PARTY_NOTICES.md": "a0d77db7d2aca0eda811ab3c386cc32b3143f2c2609a042cf4642954f0f6fb09",
"src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/references/preflight.md": "5cf745005914c74b18a09f127d1d02ce0cb64f6cd7899501ce3cb40938c42a4c",
"src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/references/repo-provenance.md": "6a4c14207223dcc2584f7133bd14ff4590c2258fc2bd44f3721b6bf63e3e35cf",
@@ -148,7 +148,12 @@
"method": "Bounded fork-specific source audit, bundled selected-declaration helper and assertion-backed cases. Native Apple/Metal/model evidence remains blocked; no numerical/performance/skill-efficacy claim."
},
"applicability": {
- "scope": "h3-metal FastH3 requested/effective configuration, checkpoint transformation and schedule/RoPE semantics, VSA prefix/partial tiles and dispatch/fallback audit. Common catalog placement is discovery, not CPU inference or Apple qualification."
+ "scope": "h3-metal FastH3 requested/effective configuration, checkpoint transformation and schedule/RoPE semantics, VSA prefix/partial tiles and dispatch/fallback audit. Common catalog placement is discovery, not CPU inference or Apple qualification.",
+ "prerequisites": "Access to the fork-specific source and semantic contracts; native checks additionally need their matching Metal/model resources and separate execution authority.",
+ "input_contract": "Source revision/patches, requested parameters and explicitly captured environment controls, checkpoint path/key-family/config identity, requested/aligned frames, latent dimensions, packed prefix segments and sparse-pipeline availability or unknown.",
+ "produced_artifact": "A ledger of requested and normalized parameters, ambient overrides, source invariants, counterexamples, observed/unknown facts and the smallest discriminating next check.",
+ "failure_modes": "Ignoring ambient overrides, mixing checkpoint identity switches, using global rather than per-head Q/K/V concatenation, pooling padded boundary rows, or double-adding coarse attention can invalidate the semantic audit.",
+ "refusal_point": "Stop on conflicting configuration, mixed checkpoint identity, missing resources or the authorized boundary; name the missing native test/fixture and leave it blocked/not_run."
},
"limitations": [
"BLOCKED_REQUIRED_BACKEND: native Apple/Metal, actual FastH3 checkpoint/schedule/sparse parity, audio/video quality, target memory fit, performance and service evidence not_run.",
@@ -163,14 +168,25 @@
"references/source-identity.json",
"references/repo-provenance.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Record requested parameters, normalized values and later environment overrides separately, including requested versus aligned frame counts.",
+ "Audit checkpoint transformations and schedule consumers beyond tensor names/shapes; tiny permutation fixtures can falsify mappings but do not establish model parity.",
+ "Resolve sparse eligibility and the dense-fine-plus-gated-coarse fallback explicitly; preserve unknown pipeline availability."
+ ]
},
"applicability": {
- "scope": "h3-metal FastH3 requested/effective configuration, checkpoint transformation and schedule/RoPE semantics, VSA prefix/partial tiles and dispatch/fallback audit. Common catalog placement is discovery, not CPU inference or Apple qualification."
+ "scope": "h3-metal FastH3 requested/effective configuration, checkpoint transformation and schedule/RoPE semantics, VSA prefix/partial tiles and dispatch/fallback audit. Common catalog placement is discovery, not CPU inference or Apple qualification.",
+ "prerequisites": "Access to the fork-specific source and semantic contracts; native checks additionally need their matching Metal/model resources and separate execution authority.",
+ "input_contract": "Source revision/patches, requested parameters and explicitly captured environment controls, checkpoint path/key-family/config identity, requested/aligned frames, latent dimensions, packed prefix segments and sparse-pipeline availability or unknown.",
+ "produced_artifact": "A ledger of requested and normalized parameters, ambient overrides, source invariants, counterexamples, observed/unknown facts and the smallest discriminating next check.",
+ "failure_modes": "Ignoring ambient overrides, mixing checkpoint identity switches, using global rather than per-head Q/K/V concatenation, pooling padded boundary rows, or double-adding coarse attention can invalidate the semantic audit.",
+ "refusal_point": "Stop on conflicting configuration, mixed checkpoint identity, missing resources or the authorized boundary; name the missing native test/fixture and leave it blocked/not_run."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Record requested parameters, normalized values and later environment overrides separately, including requested versus aligned frame counts.",
+ "Audit checkpoint transformations and schedule consumers beyond tensor names/shapes; tiny permutation fixtures can falsify mappings but do not establish model parity.",
+ "Resolve sparse eligibility and the dense-fine-plus-gated-coarse fallback explicitly; preserve unknown pipeline availability."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-runtime-residency.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-runtime-residency.json
index 5506ae49e..637f682f2 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-runtime-residency.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-runtime-residency.json
@@ -30,8 +30,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/LICENSE",
"src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/SKILL.md",
@@ -57,7 +57,7 @@
"source_files": {
"src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/LICENSE": "663e51edcd9a3a8f858018ce6924ad2895a24c269c26c39ad75325bcd0f7e5bc",
"src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/SKILL.md": "fdb4d9483c6efa836f9e59617da605fa9c2186e11c2041b05d5c81850ba0475d",
- "src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/metadata.json": "fe626a77af3f084f36fd73ffdb6f67976e7dceeafb94ba22078ef03c3a96cfae",
+ "src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/metadata.json": "952bd2947b68e56d249f10798e028d191a8ffbdb2f123775be66946d3e2ef77e",
"src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/references/THIRD_PARTY_NOTICES.md": "a0d77db7d2aca0eda811ab3c386cc32b3143f2c2609a042cf4642954f0f6fb09",
"src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/references/minimax-h3-apple.md": "57ce01754ba96c378f13a91305ef3d418b926f99b2753f0af99315572bce063a",
"src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/references/minimax-h3-weight-inventory.json": "fbaacaeec340d3ce893556ee864405a0746af4d13530a38cb3d5aa9d60929271",
@@ -177,7 +177,12 @@
"method": "Manual pinned-source distillation informed by the existing Repo-To-Skill ledger; no AREX/DisCo execution or behavioral efficacy claim."
},
"applicability": {
- "scope": "MiniMax-H3 joint audio/video C/Metal runtime audit: raw checkpoint compatibility, input encoders, schedule precompute, phased/cached residency, numerical controls and SSD streaming on Apple Silicon. Common catalog placement is a discovery classification, not cross-backend qualification."
+ "scope": "MiniMax-H3 joint audio/video C/Metal runtime audit: raw checkpoint compatibility, input encoders, schedule precompute, phased/cached residency, numerical controls and SSD streaming on Apple Silicon. Common catalog placement is a discovery classification, not cross-backend qualification.",
+ "prerequisites": "Access to the pinned source architecture and selected component/format inventory; native checks need matching model, fixture and backend resources within an authorized experiment budget.",
+ "input_contract": "Runtime revision/patches, macOS/SDK/GPU and memory budget, FL2VA/Ref2VA selection, component formats, prompt and ordered media, requested/aligned frames, canvases, steps/seed, layer/reuse/token/precision policies and preview/interactive-cache state.",
+ "produced_artifact": "A phase/component/owner ledger with resident weights, runtime/host buffers, overlap, release/reuse boundaries and observations/unknowns, followed by a discriminating check, actual result and remaining gap.",
+ "failure_modes": "Estimating fit from model-file bytes, confusing runtime int8 with incompatible MLX bundles, ignoring preview/cache overlap, or reducing DiT residency to address a VAE/host-output peak can produce a wrong capacity diagnosis.",
+ "refusal_point": "Stop at unavailable model/fixture/backend resources or the accepted experiment budget; missing target evidence remains unverified/INCONCLUSIVE."
},
"limitations": [
"Native h3.c/Metal execution, target peak memory, audio/video quality, performance, service behavior and skill efficacy were not run.",
@@ -191,14 +196,25 @@
"references/source-identity.json",
"references/minimax-h3-weight-inventory.json"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Resolve the actual conditioning and output components before counting residency; do not add alternative checkpoints or count shared files twice.",
+ "Trace one-shot, preview and interactive ownership separately, including AdaLN schedule dependencies and prepared-cache reset boundaries.",
+ "Classify approximation and precision controls before comparison, then measure the relevant phase with tensor, physical/compressed/swap and storage observations."
+ ]
},
"applicability": {
- "scope": "MiniMax-H3 joint audio/video C/Metal runtime audit: raw checkpoint compatibility, input encoders, schedule precompute, phased/cached residency, numerical controls and SSD streaming on Apple Silicon. Common catalog placement is a discovery classification, not cross-backend qualification."
+ "scope": "MiniMax-H3 joint audio/video C/Metal runtime audit: raw checkpoint compatibility, input encoders, schedule precompute, phased/cached residency, numerical controls and SSD streaming on Apple Silicon. Common catalog placement is a discovery classification, not cross-backend qualification.",
+ "prerequisites": "Access to the pinned source architecture and selected component/format inventory; native checks need matching model, fixture and backend resources within an authorized experiment budget.",
+ "input_contract": "Runtime revision/patches, macOS/SDK/GPU and memory budget, FL2VA/Ref2VA selection, component formats, prompt and ordered media, requested/aligned frames, canvases, steps/seed, layer/reuse/token/precision policies and preview/interactive-cache state.",
+ "produced_artifact": "A phase/component/owner ledger with resident weights, runtime/host buffers, overlap, release/reuse boundaries and observations/unknowns, followed by a discriminating check, actual result and remaining gap.",
+ "failure_modes": "Estimating fit from model-file bytes, confusing runtime int8 with incompatible MLX bundles, ignoring preview/cache overlap, or reducing DiT residency to address a VAE/host-output peak can produce a wrong capacity diagnosis.",
+ "refusal_point": "Stop at unavailable model/fixture/backend resources or the accepted experiment budget; missing target evidence remains unverified/INCONCLUSIVE."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Resolve the actual conditioning and output components before counting residency; do not add alternative checkpoints or count shared files twice.",
+ "Trace one-shot, preview and interactive ownership separately, including AdaLN schedule dependencies and prepared-cache reset boundaries.",
+ "Classify approximation and precision controls before comparison, then measure the relevant phase with tensor, physical/compressed/swap and storage observations."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-llama-runtime-state.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-llama-runtime-state.json
index ea5171357..cbcacf0f3 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-llama-runtime-state.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-llama-runtime-state.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/SKILL.md",
"src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/metadata.json",
@@ -41,7 +41,7 @@
"checkout_ref": "checkout:audit-llama-runtime-state",
"source_files": {
"src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/SKILL.md": "95e81b6bc096a954fa73139ea5bc001d1efa6c99747161da62f012b494e5c013",
- "src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/metadata.json": "1d30b31dc8352fc6a8d7b8abbef8118c840e47e5b90795dae0c758bf9717278c",
+ "src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/metadata.json": "22f1d01b48ea6e2b2ef2deeb20c1404ce8b586eb18442b5adb1390310b4b7503",
"src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/references/source-notes.md": "3d12a401e9decb55a2f4f3b2bfa0d360ef5110f63a0469dfe285654e78054190",
"src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
},
@@ -94,7 +94,12 @@
}
],
"applicability": {
- "scope": "Pinned llama.cpp context-memory, graph-reuse and server task-lifecycle diagnosis; concrete model/backend behavior requires native validation."
+ "scope": "Pinned llama.cpp context-memory, graph-reuse and server task-lifecycle diagnosis; concrete model/backend behavior requires native validation.",
+ "prerequisites": "Pinned llama.cpp source or a concrete runtime trace, access to the relevant memory/graph/server ownership evidence and an identified caller recovery policy that remains in force.",
+ "input_contract": "Source/build and model revisions, memory type, actual backend/device and execution mode, context/batch/ubatch parameters, sequence IDs/positions/output flags and return status; graph cases need dispatch evidence, server cases need correlated HTTP/task/slot/cache logs.",
+ "produced_artifact": "An owner/event ledger of ubatch progress, retained memory, graph-gate decisions, input readiness and slot release/retention, with the first supported broken obligation or smallest missing observation.",
+ "failure_modes": "Assuming whole-batch rollback after partial decode failure, treating token count as a complete graph-reuse key, or equating posted cancellation with released slots can hide retained state or misdiagnose correct deferred work.",
+ "refusal_point": "Stop before reset or repair if revision, memory type, actual backend, sequence ownership or partial progress is unknown; escalate unsupported rollback or unresolved in-flight work."
},
"limitations": [
"Source-derived candidate; native runtime and efficacy validation not run.",
@@ -104,14 +109,25 @@
"provenance_readsets": [
"references/source-notes.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Reconcile per-sequence retained positions with completed ubatches before retrying a failed logical batch.",
+ "Check both graph allow_reuse and input can_reuse gates, then establish copy readiness independently of graph identity.",
+ "Trace cancellation posting, normal task processing, slot release and prompt retention as separate events."
+ ]
},
"applicability": {
- "scope": "Pinned llama.cpp context-memory, graph-reuse and server task-lifecycle diagnosis; concrete model/backend behavior requires native validation."
+ "scope": "Pinned llama.cpp context-memory, graph-reuse and server task-lifecycle diagnosis; concrete model/backend behavior requires native validation.",
+ "prerequisites": "Pinned llama.cpp source or a concrete runtime trace, access to the relevant memory/graph/server ownership evidence and an identified caller recovery policy that remains in force.",
+ "input_contract": "Source/build and model revisions, memory type, actual backend/device and execution mode, context/batch/ubatch parameters, sequence IDs/positions/output flags and return status; graph cases need dispatch evidence, server cases need correlated HTTP/task/slot/cache logs.",
+ "produced_artifact": "An owner/event ledger of ubatch progress, retained memory, graph-gate decisions, input readiness and slot release/retention, with the first supported broken obligation or smallest missing observation.",
+ "failure_modes": "Assuming whole-batch rollback after partial decode failure, treating token count as a complete graph-reuse key, or equating posted cancellation with released slots can hide retained state or misdiagnose correct deferred work.",
+ "refusal_point": "Stop before reset or repair if revision, memory type, actual backend, sequence ownership or partial progress is unknown; escalate unsupported rollback or unresolved in-flight work."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Reconcile per-sequence retained positions with completed ubatches before retrying a failed logical batch.",
+ "Check both graph allow_reuse and input can_reuse gates, then establish copy readiness independently of graph identity.",
+ "Trace cancellation posting, normal task processing, slot release and prompt retention as separate events."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-recurrent-kv-finalization.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-recurrent-kv-finalization.json
index 21373ea43..6ec452006 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-recurrent-kv-finalization.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-recurrent-kv-finalization.json
@@ -26,8 +26,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/LICENSE",
"src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/SKILL.md",
@@ -48,7 +48,7 @@
"source_files": {
"src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/LICENSE": "cfc7749b96f63bd31c3c42b5c471bf756814053e847c10f3eb003417bc523d30",
"src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/SKILL.md": "a468d75f25e77b96e6fb5376692f7a89c0464cc00d4df035f1dc6cc923f713cc",
- "src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/metadata.json": "db0b4593d7a6fbb898fdf11a1710b529abbade847ea57319843669b375b68fc8",
+ "src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/metadata.json": "dfd605f8f31740565063fad0785dea3e7751b8deea4d43e1a66ca30d92ae901f",
"src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/references/NOTICE.md": "cca794656897ecdc37f983f7f23c4cd12306dbe3c9cd2f54fece6340753ae4cf",
"src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/references/source-identity.json": "5e52db0a0bda4673b9aac46a3fc035b824780def72921fee6796edb2ed9fca65",
"src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/references/source-notes.md": "0956f59cd9a1fac27da53228dd60c51a07b87639a43c09bfab296c4aeeceaab9",
@@ -112,7 +112,12 @@
}
],
"applicability": {
- "scope": "Pinned vLLM-RLT native Ouro synchronous scheduling with host KVMetadata: LAST_EXITED propagation versus SHARED, later depth-specific Torch-reference attention and normal release/wait/free order. No speculation, connector, preemption, prefix sharing, split/packed prefill or graph/static-buffer path qualification."
+ "scope": "Pinned vLLM-RLT native Ouro synchronous scheduling with host KVMetadata: LAST_EXITED propagation versus SHARED, later depth-specific Torch-reference attention and normal release/wait/free order. No speculation, connector, preemption, prefix sharing, split/packed prefill or graph/static-buffer path qualification.",
+ "prerequisites": "The selected native Ouro synchronous path uses host KVMetadata, no speculation/connector/preemption, atomic prefill and disabled graph/static-buffer/packed-prefill paths; changed execution conditions need a new bounded check.",
+ "input_contract": "Repository revision, model/configuration, recurrent-layer set, loop count, cache layout and block size, dtype/device, attention backend and effective flags, plus the traced token/depth/layer and its next consumer.",
+ "produced_artifact": "A bounded token/finalization/consumer/retirement trace separating source call order, observed data and written validity, completion evidence and remaining native model/backend/performance gaps.",
+ "failure_modes": "Substituting mark_finalized for K/V propagation, accepting allocated pages as a complete written prefix, applying LAST_EXITED copy expectations to SHARED, or treating event recording as completion can invalidate the ownership diagnosis.",
+ "refusal_point": "On a failed prerequisite or complete-prefix check, retain the first failing token/depth/layer and trace the missing operation. Stop the conclusion when flags or metadata ownership leave the selected path."
},
"limitations": [
"Source-derived candidate; native CPU/CUDA, model correctness/quality, performance and agent efficacy are not validated.",
@@ -123,14 +128,25 @@
"references/source-notes.md",
"references/source-identity.json"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Distinguish copied K/V values, logical written positions and allocated capacity; follow the later depth-specific attention consumer.",
+ "Preserve executed shallower values during LAST_EXITED propagation and use SHARED as a separate one-plane control.",
+ "Follow the latest request event through runner release before scheduler cache release; page reuse is not evidence of erased old bytes."
+ ]
},
"applicability": {
- "scope": "Pinned vLLM-RLT native Ouro synchronous scheduling with host KVMetadata: LAST_EXITED propagation versus SHARED, later depth-specific Torch-reference attention and normal release/wait/free order. No speculation, connector, preemption, prefix sharing, split/packed prefill or graph/static-buffer path qualification."
+ "scope": "Pinned vLLM-RLT native Ouro synchronous scheduling with host KVMetadata: LAST_EXITED propagation versus SHARED, later depth-specific Torch-reference attention and normal release/wait/free order. No speculation, connector, preemption, prefix sharing, split/packed prefill or graph/static-buffer path qualification.",
+ "prerequisites": "The selected native Ouro synchronous path uses host KVMetadata, no speculation/connector/preemption, atomic prefill and disabled graph/static-buffer/packed-prefill paths; changed execution conditions need a new bounded check.",
+ "input_contract": "Repository revision, model/configuration, recurrent-layer set, loop count, cache layout and block size, dtype/device, attention backend and effective flags, plus the traced token/depth/layer and its next consumer.",
+ "produced_artifact": "A bounded token/finalization/consumer/retirement trace separating source call order, observed data and written validity, completion evidence and remaining native model/backend/performance gaps.",
+ "failure_modes": "Substituting mark_finalized for K/V propagation, accepting allocated pages as a complete written prefix, applying LAST_EXITED copy expectations to SHARED, or treating event recording as completion can invalidate the ownership diagnosis.",
+ "refusal_point": "On a failed prerequisite or complete-prefix check, retain the first failing token/depth/layer and trace the missing operation. Stop the conclusion when flags or metadata ownership leave the selected path."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Distinguish copied K/V values, logical written positions and allocated capacity; follow the later depth-specific attention consumer.",
+ "Preserve executed shallower values during LAST_EXITED propagation and use SHARED as a separate one-plane control.",
+ "Follow the latest request event through runner release before scheduler cache release; page reuse is not evidence of erased old bytes."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sd-cpp-residency.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sd-cpp-residency.json
index b2bde92e2..5b331e56a 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sd-cpp-residency.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sd-cpp-residency.json
@@ -25,8 +25,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/LICENSE",
"src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/SKILL.md",
@@ -49,7 +49,7 @@
"source_files": {
"src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/LICENSE": "b53fa08f515cb5a6fff7b9fd8fcd0961a4b80df29d74372d25e1f9171aa042ee",
"src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/SKILL.md": "4c04d39812fb566e3ecb5807c170d5c32848eb2b93a1f0fedcf66ba9a9c92860",
- "src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/metadata.json": "13ab6d874c4e3135be80ef7c3cb6bfd28fb9121012e0536831ec3705ccb8c5eb",
+ "src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/metadata.json": "412c649b16526038ee0299f8b2a610bb7f120eaab25cd83575809fe03b604b2e",
"src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/references/loader-storage-boundaries.md": "c1ec047aa37fca6b374f9bfd750e8f30d45af1d48fe728db23873c0b84ad94e8",
"src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/references/phase-residency.md": "7b537cb972347dd68e303960f142892346d8b19b884e4d08016ea36a23bdaa7a",
"src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/references/source-identity.json": "af4830370b6eddc2d176a9ae14da4ab56a90b0fe048a01a89ee051df4365d8e9",
@@ -117,7 +117,12 @@
}
],
"applicability": {
- "scope": "Source-grounded audit of native image-generation weight/runtime residency, optional segment prefetch and bounded non-ZIP safetensors catalog/read/staging ownership; confirm the actual source, GGML, backend, model and executing path independently."
+ "scope": "Source-grounded audit of native image-generation weight/runtime residency, optional segment prefetch and bounded non-ZIP safetensors catalog/read/staging ownership; confirm the actual source, GGML, backend, model and executing path independently.",
+ "prerequisites": "Access to the pinned source/storage boundaries and evidence of the actual image runner/backend dispatch; native capacity or prefetch checks require matching GGML, model and backend resources.",
+ "input_contract": "Runtime/GGML revisions and selection, compiled/loaded backends and devices, model/components and numerical image settings, effective memory/prefetch/placement settings and actual dispatch; retain available phase allocations, cache/workspace, transfer events and completed-image counts.",
+ "produced_artifact": "A phase-by-runner/backend/device ownership table for stored and resident weights, retained runtime, pending allocations, pin/prefetch ownership and release/eviction, with one discriminating check and its actual status.",
+ "failure_modes": "Treating max_vram as proof of segmentation, unpin as physical freeing, or configured prefetch as completed useful overlap can misidentify the capacity cause; missing observations cannot be read as zero.",
+ "refusal_point": "Stop when the cause is resolved or matching GGML/model/backend resources are unavailable; invoking this audit does not authorize dependency installation, model downloads, policy changes or device provisioning."
},
"limitations": [
"Source-derived candidate; native inference, allocator behavior, image quality and performance were not run.",
@@ -132,14 +137,25 @@
"references/source-notes.md",
"references/loader-storage-boundaries.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Separate parameter storage, compute-resident weights, retained workspace/output caches and pending allocations for each runner and phase.",
+ "Verify the actual monolithic, segmented or multi-device path before interpreting a memory budget or prefetch setting.",
+ "For a policy comparison, hold model, inputs, numerical settings and device fixed; check completed images and quality before latency, and separate startup, first and warm requests."
+ ]
},
"applicability": {
- "scope": "Source-grounded audit of native image-generation weight/runtime residency, optional segment prefetch and bounded non-ZIP safetensors catalog/read/staging ownership; confirm the actual source, GGML, backend, model and executing path independently."
+ "scope": "Source-grounded audit of native image-generation weight/runtime residency, optional segment prefetch and bounded non-ZIP safetensors catalog/read/staging ownership; confirm the actual source, GGML, backend, model and executing path independently.",
+ "prerequisites": "Access to the pinned source/storage boundaries and evidence of the actual image runner/backend dispatch; native capacity or prefetch checks require matching GGML, model and backend resources.",
+ "input_contract": "Runtime/GGML revisions and selection, compiled/loaded backends and devices, model/components and numerical image settings, effective memory/prefetch/placement settings and actual dispatch; retain available phase allocations, cache/workspace, transfer events and completed-image counts.",
+ "produced_artifact": "A phase-by-runner/backend/device ownership table for stored and resident weights, retained runtime, pending allocations, pin/prefetch ownership and release/eviction, with one discriminating check and its actual status.",
+ "failure_modes": "Treating max_vram as proof of segmentation, unpin as physical freeing, or configured prefetch as completed useful overlap can misidentify the capacity cause; missing observations cannot be read as zero.",
+ "refusal_point": "Stop when the cause is resolved or matching GGML/model/backend resources are unavailable; invoking this audit does not authorize dependency installation, model downloads, policy changes or device provisioning."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Separate parameter storage, compute-resident weights, retained workspace/output caches and pending allocations for each runner and phase.",
+ "Verify the actual monolithic, segmented or multi-device path before interpreting a memory budget or prefetch setting.",
+ "For a policy comparison, hold model, inputs, numerical settings and device fixed; check completed images and quality before latency, and separate startup, first and warm requests."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sglang-diffusion-lifetimes.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sglang-diffusion-lifetimes.json
index a76263984..19b7de83b 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sglang-diffusion-lifetimes.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sglang-diffusion-lifetimes.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/LICENSE",
"src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/SKILL.md",
@@ -43,7 +43,7 @@
"source_files": {
"src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/LICENSE": "1495e1e757ef4d0925a2350563cf5754bb23c51701a8ec4fb3c5cdcbedae6747",
"src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/SKILL.md": "b0a4ba4751087126143ef86983798deb0ed00e24b6ff06bf6ef76a9df6e274f2",
- "src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/metadata.json": "6ed57dfb4f25ecb34b81f2188b055e05d172ba2581e6a3b605b814b2f733a4d6",
+ "src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/metadata.json": "e2cb137978960d29b3644967a29dab930c768d3453bce8aff2aad96fd90a56d9",
"src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/references/source-notes.md": "d0fcfd1177ae2f351041c5b2abba7f5121cec133006d9e4dca8f74ddb567332d",
"src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
},
@@ -101,7 +101,12 @@
}
],
"applicability": {
- "scope": "Bounded monolithic multimodal_gen composed-pipeline state and lifecycle diagnosis; actual backend/model behavior requires native validation."
+ "scope": "Bounded monolithic multimodal_gen composed-pipeline state and lifecycle diagnosis; actual backend/model behavior requires native validation.",
+ "prerequisites": "Pinned source or a concrete trace of the selected composed-pipeline executor and installed residency manager, with the applicable monolithic route identified.",
+ "input_contract": "Source/build, backend and executor/pipeline class, model revision, modality/shape, steps/scheduler/seed/output count, grouped/serial route, ranks, placement, warmup/save settings and symptom; timeout cases need correlated request/job and client/scheduler/worker events.",
+ "produced_artifact": "A version-bound owner/event ledger, first unresolved boundary, causal hypothesis, smallest real discriminating check, exact results and missing dependencies.",
+ "failure_modes": "Assuming all executors share the SyncExecutor contract, treating closed component use as eviction, assuming grouped schedulers are isolated, or reading socket close/job deletion as stopped device compute can misdiagnose ownership.",
+ "refusal_point": "Stop performance claims when the executing path, cleanup or output quality is unresolved; retain the failure and a known baseline rather than inferring a blanket leak or safety conclusion."
},
"limitations": [
"Source-derived candidate; upstream runtime and efficacy validation not run.",
@@ -111,14 +116,25 @@
"provenance_readsets": [
"references/source-notes.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Trace the actual executor and separate request, stage, component-use, module, scheduler, finalizer and allocator-cache lifetimes.",
+ "Check scheduler aliases and declared output-copy policies before introducing overlap; compare real serial/grouped outputs with fixed timesteps and seeds.",
+ "Correlate worker, finalizer and device completion after client timeout or deletion; preserve an unknown execution outcome until that evidence exists."
+ ]
},
"applicability": {
- "scope": "Bounded monolithic multimodal_gen composed-pipeline state and lifecycle diagnosis; actual backend/model behavior requires native validation."
+ "scope": "Bounded monolithic multimodal_gen composed-pipeline state and lifecycle diagnosis; actual backend/model behavior requires native validation.",
+ "prerequisites": "Pinned source or a concrete trace of the selected composed-pipeline executor and installed residency manager, with the applicable monolithic route identified.",
+ "input_contract": "Source/build, backend and executor/pipeline class, model revision, modality/shape, steps/scheduler/seed/output count, grouped/serial route, ranks, placement, warmup/save settings and symptom; timeout cases need correlated request/job and client/scheduler/worker events.",
+ "produced_artifact": "A version-bound owner/event ledger, first unresolved boundary, causal hypothesis, smallest real discriminating check, exact results and missing dependencies.",
+ "failure_modes": "Assuming all executors share the SyncExecutor contract, treating closed component use as eviction, assuming grouped schedulers are isolated, or reading socket close/job deletion as stopped device compute can misdiagnose ownership.",
+ "refusal_point": "Stop performance claims when the executing path, cleanup or output quality is unresolved; retain the failure and a known baseline rather than inferring a blanket leak or safety conclusion."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Trace the actual executor and separate request, stage, component-use, module, scheduler, finalizer and allocator-cache lifetimes.",
+ "Check scheduler aliases and declared output-copy policies before introducing overlap; compare real serial/grouped outputs with fixed timesteps and seeds.",
+ "Correlate worker, finalizer and device completion after client timeout or deletion; preserve an unknown execution outcome until that evidence exists."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-vllm-kv-lifetimes.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-vllm-kv-lifetimes.json
index 9173d33e8..7deddd33d 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-vllm-kv-lifetimes.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-vllm-kv-lifetimes.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/SKILL.md",
"src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/metadata.json",
@@ -39,7 +39,7 @@
"checkout_ref": "checkout:audit-vllm-kv-lifetimes",
"source_files": {
"src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/SKILL.md": "4c3f27459df239f9f4fc971c559d4c9ddfbfebd70507a8a93be8eae59a30b69d",
- "src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/metadata.json": "b43fc3a7ccae59c664fcdf9ff77962edcc07c2a3df8b59324db551642bd23e23",
+ "src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/metadata.json": "8ded6e60c562131616dc8cdcf0d7efc04aef1884885da0b30fb16f42d7433b7c",
"src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/references/source-notes.md": "3665b034594fc8cc767c408002f4e16f8803369242405027c7f37ce6c9ecebe9",
"src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
},
@@ -78,12 +78,33 @@
"provenance_readsets": [
"references/source-notes.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "applicability": {
+ "scope": "Pinned vLLM V1 scheduler/connector ownership audit for asynchronous KV load/save completion, retained blocks, preemption and failed receives; concrete connector conformance is not established by the interface.",
+ "prerequisites": "A concrete connector and worker/model-runner call trace at identified revisions, with request/block events sufficient to compare actual callers against the pinned scheduler/interface contracts.",
+ "input_contract": "Engine/connector revision and class, producer/consumer roles, KV-group count/block namespace, failure policy, concurrent-batch and graph modes; correlate bounded request IDs with allocation, matched tokens, load/save events, receive failures, preemption, finish and pool return.",
+ "produced_artifact": "An ownership/event ledger and bounded contract-conformance report naming the earliest evidenced broken obligation, owner, affected requests/blocks and smallest missing observation, with a proposed targeted regression check.",
+ "failure_modes": "Equating reserved tokens with usable KV, finished responses with reusable blocks, or send completion with scheduler write-fence completion can hide unsafe reuse; missing failed-receive completion can leave waits unreleased.",
+ "refusal_point": "Escalate absent concrete caller evidence, ambiguous block namespaces, unsupported layouts or revision drift; stop before inferring a backend fix, freeing retained blocks or claiming a reproduced bug."
+ },
+ "recommendations": [
+ "Trace lookup, allocation, worker execution and completion propagation before changing request state; repeated or undecided lookup and ordinary admission waits can be valid.",
+ "Check delayed-free ownership and scheduler in-flight-write fences separately, including empty-step save/metadata ordering in the real worker.",
+ "Keep receive failure and finished_recving correlated, and use request-level failure evidence for multi-group layouts while preserving the existing recovery policy."
+ ]
+ },
+ "applicability": {
+ "scope": "Pinned vLLM V1 scheduler/connector ownership audit for asynchronous KV load/save completion, retained blocks, preemption and failed receives; concrete connector conformance is not established by the interface.",
+ "prerequisites": "A concrete connector and worker/model-runner call trace at identified revisions, with request/block events sufficient to compare actual callers against the pinned scheduler/interface contracts.",
+ "input_contract": "Engine/connector revision and class, producer/consumer roles, KV-group count/block namespace, failure policy, concurrent-batch and graph modes; correlate bounded request IDs with allocation, matched tokens, load/save events, receive failures, preemption, finish and pool return.",
+ "produced_artifact": "An ownership/event ledger and bounded contract-conformance report naming the earliest evidenced broken obligation, owner, affected requests/blocks and smallest missing observation, with a proposed targeted regression check.",
+ "failure_modes": "Equating reserved tokens with usable KV, finished responses with reusable blocks, or send completion with scheduler write-fence completion can hide unsafe reuse; missing failed-receive completion can leave waits unreleased.",
+ "refusal_point": "Escalate absent concrete caller evidence, ambiguous block namespaces, unsupported layouts or revision drift; stop before inferring a backend fix, freeing retained blocks or claiming a reproduced bug."
},
- "applicability": {},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Trace lookup, allocation, worker execution and completion propagation before changing request state; repeated or undecided lookup and ordinary admission waits can be valid.",
+ "Check delayed-free ownership and scheduler in-flight-write fences separately, including empty-step save/metadata ordering in the real worker.",
+ "Keep receive failure and finished_recving correlated, and use request-level failure evidence for multi-group layouts while preserving the existing recovery policy."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/bound-windowed-attention-replay.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/bound-windowed-attention-replay.json
index b27a542ce..9ac8dc741 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/bound-windowed-attention-replay.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/bound-windowed-attention-replay.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/SKILL.md",
"src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/metadata.json",
@@ -42,7 +42,7 @@
"checkout_ref": "checkout:bound-windowed-attention-replay",
"source_files": {
"src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/SKILL.md": "4703beb6e5194ef2ff2c0b5ec69f64e35417a4bc0dfb50a10c80a920b77ef3b5",
- "src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/metadata.json": "014be24d0e669f4acbc0717dc323eec784026cf185186619d5a19d6d84496131",
+ "src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/metadata.json": "c68254b1a96a8834ae46c7a1d9035856fa4783315b219aa95e31461bd8ac58c9",
"src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/references/source-notes.md": "fd917bd5ac80bee9c9f4525f8e0300f6d67dc144e2a16632af42a67f3d53d098",
"src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/references/validation.md": "60e71e70c5519bcdf6c171eec2be2f342e8801bedd746d6005f11adc646dfaff",
"src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
@@ -90,7 +90,11 @@
"applicability": {
"scope": "Decoders with an explicit producer/consumer boundary and a demonstrated dependency closure for compact query rows, including separately preserved historical state",
"prerequisites": "A derived dependency closure; a full-length KV write at the boundary; known rollback depth; control over metadata construction",
- "hardware": "Backend-independent retain arithmetic; window sizes, boundary index and ring lengths come from the model config and do not transfer"
+ "hardware": "Backend-independent retain arithmetic; window sizes, boundary index and ring lengths come from the model config and do not transfer",
+ "input_contract": "The actual producer/consumer split and transitive dependency closure, boundary index, window and rollback depth, consumer KV sharing/compression layout and separately preserved historical keys.",
+ "produced_artifact": "A dependency listing, enforced retain length, compact query/position/store-slot/visibility metadata, a separate historical-key retention statement and a diagnostic equivalence-reference mode.",
+ "failure_modes": "Applying window-plus-rollback without dependency closure, discarding full-layout state, clipping away the sampling row, or comparing across preemption can invalidate replay; compact queries do not imply historical keys were dropped.",
+ "refusal_point": "Stop if required consumer state is unavailable without a preserved full-length copy, the boundary representation cannot be captured, or comparison preemption moves the anchor. Do not use for clipped-region outputs or required bit-identical preemption replay."
},
"limitations": [
"Source-derived candidate; no equivalence run was performed.",
@@ -104,16 +108,26 @@
"references/source-notes.md",
"references/validation.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Derive every consumer dependency before computing keep; preserve the required full-layout representation separately from the compact query rows.",
+ "Use the visibility-only diagnostic mode to isolate equivalence from row gathering, and retain one sampling row for a fully clipped chunk.",
+ "Check metadata at chunk/store boundaries, actual prompt-logprob handling and zero preemption during equivalence comparison."
+ ]
},
"applicability": {
"scope": "Decoders with an explicit producer/consumer boundary and a demonstrated dependency closure for compact query rows, including separately preserved historical state",
"prerequisites": "A derived dependency closure; a full-length KV write at the boundary; known rollback depth; control over metadata construction",
- "hardware": "Backend-independent retain arithmetic; window sizes, boundary index and ring lengths come from the model config and do not transfer"
+ "hardware": "Backend-independent retain arithmetic; window sizes, boundary index and ring lengths come from the model config and do not transfer",
+ "input_contract": "The actual producer/consumer split and transitive dependency closure, boundary index, window and rollback depth, consumer KV sharing/compression layout and separately preserved historical keys.",
+ "produced_artifact": "A dependency listing, enforced retain length, compact query/position/store-slot/visibility metadata, a separate historical-key retention statement and a diagnostic equivalence-reference mode.",
+ "failure_modes": "Applying window-plus-rollback without dependency closure, discarding full-layout state, clipping away the sampling row, or comparing across preemption can invalidate replay; compact queries do not imply historical keys were dropped.",
+ "refusal_point": "Stop if required consumer state is unavailable without a preserved full-length copy, the boundary representation cannot be captured, or comparison preemption moves the anchor. Do not use for clipped-region outputs or required bit-identical preemption replay."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Derive every consumer dependency before computing keep; preserve the required full-layout representation separately from the compact query rows.",
+ "Use the visibility-only diagnostic mode to isolate equivalence from row gathering, and retain one sampling row for a fully clipped chunk.",
+ "Check metadata at chunk/store boundaries, actual prompt-logprob handling and zero preemption during equivalence comparison."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/budget-speculative-verification.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/budget-speculative-verification.json
index 29288ee06..f419ee351 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/budget-speculative-verification.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/budget-speculative-verification.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/SKILL.md",
"src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/metadata.json",
@@ -42,7 +42,7 @@
"checkout_ref": "checkout:budget-speculative-verification",
"source_files": {
"src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/SKILL.md": "f157562d1edc905cb4fc07246567e3c5b844efb9fbf5fe39f0d40553a1b1a307",
- "src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/metadata.json": "4d68bf4c1767eaee12ade963bbd551d0e61d1bcd81a984bb02fd3aae46965919",
+ "src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/metadata.json": "1aa80a3109072c1ec211f38dbeb9e9cb9edb56aeac187946b43e65dd7502e46f",
"src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/references/source-notes.md": "6fa955022f16cd598f44fd6521901d1cb868b06cec545cabe2b82769a641b5d6",
"src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/references/validation.md": "673d93f6700052974b340fab25b3438e0e6344960569ac9abec7d7720f662288",
"src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
@@ -88,7 +88,11 @@
"applicability": {
"scope": "Engines with multi-token speculative decoding whose verification forward cost grows with query rows",
"prerequisites": "A calibrated per-position acceptance signal; a measured (rows, step_time) curve; a handle on metadata construction",
- "hardware": "Backend-independent procedure; the reference implementation targets NVIDIA SM100 and Ascend 950 only"
+ "hardware": "Backend-independent procedure; the reference implementation targets NVIDIA SM100 and Ascend 950 only",
+ "input_contract": "Draft width and row representation, the defined individual/conditional versus cumulative acceptance semantics, verify rows/time curve, sampler cost and acceptance round actually available at schedule time.",
+ "produced_artifact": "A per-step verification row budget, per-request query lengths from one through draft_width plus one, cumulative acceptance vectors and predicted-versus-realized acceptance error logs.",
+ "failure_modes": "Confusing verification savings with reduced drafting, misreading confidence semantics or available rounds, allowing later-position confidence to select an earlier token, or generating non-prefix grants can invalidate the budget or sampling claim.",
+ "refusal_point": "Stop if unthrottled drafting or sampler cost dominates, the signal cannot be made cumulative, or no round supplies acceptance estimates; the procedure also does not apply without a per-step cost model."
},
"limitations": [
"Source-derived candidate; no local efficacy validation and no reproduced gain.",
@@ -103,16 +107,26 @@
"references/source-notes.md",
"references/validation.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Measure draft, verify and sampler costs separately and fit the verify curve for graph sizes that will actually replay.",
+ "Preserve the one-row floor and prefix grants, and audit CPU/GPU allocation timing against the per-position decision-independence contract.",
+ "Report verification rows, accepted tokens, total step time and throughput separately; use an independently specified sampling oracle rather than requiring unchanged accepted length."
+ ]
},
"applicability": {
"scope": "Engines with multi-token speculative decoding whose verification forward cost grows with query rows",
"prerequisites": "A calibrated per-position acceptance signal; a measured (rows, step_time) curve; a handle on metadata construction",
- "hardware": "Backend-independent procedure; the reference implementation targets NVIDIA SM100 and Ascend 950 only"
+ "hardware": "Backend-independent procedure; the reference implementation targets NVIDIA SM100 and Ascend 950 only",
+ "input_contract": "Draft width and row representation, the defined individual/conditional versus cumulative acceptance semantics, verify rows/time curve, sampler cost and acceptance round actually available at schedule time.",
+ "produced_artifact": "A per-step verification row budget, per-request query lengths from one through draft_width plus one, cumulative acceptance vectors and predicted-versus-realized acceptance error logs.",
+ "failure_modes": "Confusing verification savings with reduced drafting, misreading confidence semantics or available rounds, allowing later-position confidence to select an earlier token, or generating non-prefix grants can invalidate the budget or sampling claim.",
+ "refusal_point": "Stop if unthrottled drafting or sampler cost dominates, the signal cannot be made cumulative, or no round supplies acceptance estimates; the procedure also does not apply without a per-step cost model."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Measure draft, verify and sampler costs separately and fit the verify curve for graph sizes that will actually replay.",
+ "Preserve the one-row floor and prefix grants, and audit CPU/GPU allocation timing against the per-position decision-independence contract.",
+ "Report verification rows, accepted tokens, total step time and throughput separately; use an independently specified sampling oracle rather than requiring unchanged accepted length."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/calibrate-serving-simulation.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/calibrate-serving-simulation.json
index a9920f438..8ec449df8 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/calibrate-serving-simulation.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/calibrate-serving-simulation.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/methods/calibrate-serving-simulation/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/methods/calibrate-serving-simulation/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/methods/calibrate-serving-simulation/SKILL.md",
"src/inferrsi/data/skill_library/methods/calibrate-serving-simulation/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cgroup-resource-controls.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cgroup-resource-controls.json
index 7169a9d63..6b7388dc0 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cgroup-resource-controls.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cgroup-resource-controls.json
@@ -25,8 +25,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/hardware/host/cgroup-resource-controls/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/hardware/host/cgroup-resource-controls/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/hardware/host/cgroup-resource-controls/SKILL.md",
"src/inferrsi/data/skill_library/hardware/host/cgroup-resource-controls/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/checkpoint-recovery-consistency.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/checkpoint-recovery-consistency.json
index 59160a87d..0583daf4a 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/checkpoint-recovery-consistency.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/checkpoint-recovery-consistency.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/methods/checkpoint-recovery-consistency/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/methods/checkpoint-recovery-consistency/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/methods/checkpoint-recovery-consistency/SKILL.md",
"src/inferrsi/data/skill_library/methods/checkpoint-recovery-consistency/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/choose-parallel-training-plan.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/choose-parallel-training-plan.json
index 4149eb57b..67d8f7fa8 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/choose-parallel-training-plan.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/choose-parallel-training-plan.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/methods/choose-parallel-training-plan/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/methods/choose-parallel-training-plan/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/methods/choose-parallel-training-plan/SKILL.md",
"src/inferrsi/data/skill_library/methods/choose-parallel-training-plan/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cite-check-downstream-pr.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cite-check-downstream-pr.json
index 066b9cfd9..4d0cd5072 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cite-check-downstream-pr.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cite-check-downstream-pr.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/workflows/cite-check-downstream-pr/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/workflows/cite-check-downstream-pr/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/skill-index.json",
"src/inferrsi/data/skill_library/workflows/cite-check-downstream-pr/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/claim-contract.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/claim-contract.json
index 1626bcd8a..fb38b7757 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/claim-contract.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/claim-contract.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/workflows/claim-contract/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/workflows/claim-contract/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/skill-index.json",
"src/inferrsi/data/skill_library/workflows/claim-contract/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/comfy-ui.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/comfy-ui.json
index 1b15338b2..045391f46 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/comfy-ui.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/comfy-ui.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/repositories/comfy-ui/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/repositories/comfy-ui/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/repositories/comfy-ui/LICENSE",
"src/inferrsi/data/skill_library/repositories/comfy-ui/SKILL.md",
@@ -45,7 +45,7 @@
"source_files": {
"src/inferrsi/data/skill_library/repositories/comfy-ui/LICENSE": "3972dc9744f6499f0f9b2dbf76696f2ae7ad8af9b23dde66d6af86c9dfb36986",
"src/inferrsi/data/skill_library/repositories/comfy-ui/SKILL.md": "c993bd6ff8e9b5f82f39f175ec8242341136e4ee0808af414e1d94ad3e196c5e",
- "src/inferrsi/data/skill_library/repositories/comfy-ui/metadata.json": "74742851a002a51c208362cdcd2a59fdab5d1734d8492b1f93fbec3cf21e0841",
+ "src/inferrsi/data/skill_library/repositories/comfy-ui/metadata.json": "81ad1c73b12426e4bf7961889f4b9a1c92739c6be18380402f7866e021cac67b",
"src/inferrsi/data/skill_library/repositories/comfy-ui/references/graph-cache-invalidation.md": "614c9a4e6943d24d7d7be7d632682ce8b1dcc239124a37d9dc1d487113772eac",
"src/inferrsi/data/skill_library/repositories/comfy-ui/references/model-residency.md": "014c68386e7cdfdb7b4f566836e8b2ff61bd5a5a506e7b4723f83fe80c5d4de7",
"src/inferrsi/data/skill_library/repositories/comfy-ui/references/repo-provenance.md": "ca996ef515af42a3ce304747fa3f12a31c00533deaace5ba2ce21aefa0a62e2e",
@@ -104,7 +104,12 @@
}
],
"applicability": {
- "scope": "Source-grounded ComfyUI workflow-output cache and model-weight ownership audit; confirm the actual revision, node registry, cache caller and static/dynamic patcher before transfer."
+ "scope": "Source-grounded ComfyUI workflow-output cache and model-weight ownership audit; confirm the actual revision, node registry, cache caller and static/dynamic patcher before transfer.",
+ "prerequisites": "Access to the pinned provenance and the actual node registry/cache caller; residency questions additionally require the concrete patcher and effective dynamic/static mode. Native memory conclusions need the actual backend.",
+ "input_contract": "Exact revision, API-format prompt/output targets, relevant node classes/constants/links/hidden inputs and external state, selected cache caller/mode and patcher settings, plus available execution/load/memory/timing observations.",
+ "produced_artifact": "A state/owner/key/lifetime/consumer ledger with invalidation or release events, evidence and unresolved boundaries, followed by a discriminating check and its actual result.",
+ "failure_modes": "Treating identical JSON as proof of unchanged external inputs, confusing eviction with key validity, or equating workflow cache hits with device-resident model weights can explain the wrong state contract.",
+ "refusal_point": "Stop when the cause/action is supported or required environment, backend or model resources are unavailable; do not manufacture a hit, hardware result or performance threshold."
},
"limitations": [
"Source-derived candidate; no local executor-cache, device-residency or performance qualification.",
@@ -114,14 +119,25 @@
"provenance_readsets": [
"references/repo-provenance.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Draw the affected dependency cone with declared, lazy and dynamically expanded links, then mark unfingerprinted external state unresolved.",
+ "Separate output/object cache validity and retention from patch identity, partial unload, full detach and allocator-visible release.",
+ "Change one test variable while fixing prompt, output target, model, backend, precision, seed and cache mode; check values and executed nodes before timing."
+ ]
},
"applicability": {
- "scope": "Source-grounded ComfyUI workflow-output cache and model-weight ownership audit; confirm the actual revision, node registry, cache caller and static/dynamic patcher before transfer."
+ "scope": "Source-grounded ComfyUI workflow-output cache and model-weight ownership audit; confirm the actual revision, node registry, cache caller and static/dynamic patcher before transfer.",
+ "prerequisites": "Access to the pinned provenance and the actual node registry/cache caller; residency questions additionally require the concrete patcher and effective dynamic/static mode. Native memory conclusions need the actual backend.",
+ "input_contract": "Exact revision, API-format prompt/output targets, relevant node classes/constants/links/hidden inputs and external state, selected cache caller/mode and patcher settings, plus available execution/load/memory/timing observations.",
+ "produced_artifact": "A state/owner/key/lifetime/consumer ledger with invalidation or release events, evidence and unresolved boundaries, followed by a discriminating check and its actual result.",
+ "failure_modes": "Treating identical JSON as proof of unchanged external inputs, confusing eviction with key validity, or equating workflow cache hits with device-resident model weights can explain the wrong state contract.",
+ "refusal_point": "Stop when the cause/action is supported or required environment, backend or model resources are unavailable; do not manufacture a hit, hardware result or performance threshold."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Draw the affected dependency cone with declared, lazy and dynamically expanded links, then mark unfingerprinted external state unresolved.",
+ "Separate output/object cache validity and retention from patch identity, partial unload, full detach and allocator-visible release.",
+ "Change one test variable while fixing prompt, output target, model, backend, precision, seed and cache mode; check values and executed nodes before timing."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/competitor-scan.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/competitor-scan.json
index 515fc89aa..d86367b79 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/competitor-scan.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/competitor-scan.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/workflows/competitor-scan/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/workflows/competitor-scan/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/skill-index.json",
"src/inferrsi/data/skill_library/workflows/competitor-scan/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dcp-async-save-performance.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dcp-async-save-performance.json
index ac50b9ca3..4887bf1ab 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dcp-async-save-performance.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dcp-async-save-performance.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/SKILL.md",
"src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/metadata.json",
@@ -41,7 +41,7 @@
"checkout_ref": "checkout:dcp-async-save-performance",
"source_files": {
"src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/SKILL.md": "645139ee0c88ec865cad952e24cda35f4648d4b777b732c9b7d274b6c7830845",
- "src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/metadata.json": "db7ad368394ccbfdeff6a16e3423c300d51305b1f95ff5c3936df91738a8cc35",
+ "src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/metadata.json": "49764fa454403fb38891d82ee63de1e07a90d3ff6a3cd28550aee7ea34cbdcb0",
"src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/references/sources.md": "22afe4f4d77f441096cc657d55d763473bfcbea8fb531c8e493fe49ab84955f7",
"src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
},
@@ -71,12 +71,33 @@
"provenance_readsets": [
"references/sources.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "applicability": {
+ "scope": "Source-supported diagnosis of training-visible PyTorch DCP save pauses, snapshot/staging/upload lifetimes and bounded CPU staging-buffer reuse; distributed recovery validity and backend durability need separate evidence.",
+ "prerequisites": "Known correspondence between the installed PyTorch API and source/version notes, identified checkpointed-state mutation boundaries and an initial one-outstanding-save policy before overlap is considered.",
+ "input_contract": "Installed PyTorch revision, stager, storage writer and thread/process executor, checkpoint contents, ranks per host and state-mutation boundaries; performance assessment needs caller/staging/upload waits and host/resource observations.",
+ "produced_artifact": "A completion/lifetime diagnosis and smallest justified staging/cache/backpressure lever, with separate caller-pause, staging, upload and step-time measurements proposed for authorized validation.",
+ "failure_modes": "Mutating source state before staging completes, overwriting cached snapshots before prior upload completes, ignoring caller-side state_dict work or treating async return as durability can corrupt the snapshot or misstate the pause.",
+ "refusal_point": "Reject staging overlap that races checkpointed-state mutation; do not infer backend durability from asynchronous return or writer completion."
+ },
+ "recommendations": [
+ "Place the staging wait before the earliest mutation of any checkpointed state, including forward-mutated buffers rather than only optimizer.step.",
+ "Wait for the previous upload before reusing a stager's CPU storage; observe the final upload before closing caller-owned staging resources.",
+ "Measure checkpoint and intervening steps with the same state/cadence, retaining first-use and cached-buffer costs plus per-host memory and I/O pressure."
+ ]
+ },
+ "applicability": {
+ "scope": "Source-supported diagnosis of training-visible PyTorch DCP save pauses, snapshot/staging/upload lifetimes and bounded CPU staging-buffer reuse; distributed recovery validity and backend durability need separate evidence.",
+ "prerequisites": "Known correspondence between the installed PyTorch API and source/version notes, identified checkpointed-state mutation boundaries and an initial one-outstanding-save policy before overlap is considered.",
+ "input_contract": "Installed PyTorch revision, stager, storage writer and thread/process executor, checkpoint contents, ranks per host and state-mutation boundaries; performance assessment needs caller/staging/upload waits and host/resource observations.",
+ "produced_artifact": "A completion/lifetime diagnosis and smallest justified staging/cache/backpressure lever, with separate caller-pause, staging, upload and step-time measurements proposed for authorized validation.",
+ "failure_modes": "Mutating source state before staging completes, overwriting cached snapshots before prior upload completes, ignoring caller-side state_dict work or treating async return as durability can corrupt the snapshot or misstate the pause.",
+ "refusal_point": "Reject staging overlap that races checkpointed-state mutation; do not infer backend durability from asynchronous return or writer completion."
},
- "applicability": {},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Place the staging wait before the earliest mutation of any checkpointed state, including forward-mutated buffers rather than only optimizer.step.",
+ "Wait for the previous upload before reusing a stager's CPU storage; observe the final upload before closing caller-owned staging resources.",
+ "Measure checkpoint and intervening steps with the same state/cadence, retaining first-use and cached-buffer costs plus per-host memory and I/O pressure."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dense-performance-evidence.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dense-performance-evidence.json
index 2f9845e02..1a93ad976 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dense-performance-evidence.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dense-performance-evidence.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/methods/dense-performance-evidence/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/methods/dense-performance-evidence/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/methods/dense-performance-evidence/SKILL.md",
"src/inferrsi/data/skill_library/methods/dense-performance-evidence/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diagnose-vllm-admission-pressure.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diagnose-vllm-admission-pressure.json
index ef7a40054..dafaa94d3 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diagnose-vllm-admission-pressure.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diagnose-vllm-admission-pressure.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/SKILL.md",
"src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/metadata.json",
@@ -52,7 +52,7 @@
"checkout_ref": "checkout:diagnose-vllm-admission-pressure",
"source_files": {
"src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/SKILL.md": "d99fc28d48127d3bc627f2a0cab6f3f16158a77da49428838a08865adaea9905",
- "src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/metadata.json": "78459416734f61a8faa819d9a1a3ad7b9a985e8bf8e149813c604d15be6bb92f",
+ "src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/metadata.json": "a4a4c4b7645852bce4abafde8e4128baacbdf1bc2a69362f7c3120db597fbe40",
"src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/references/papers-20261015/INTEGRATION_NOTES.md": "82b24be82451ee121892f61575d0cf4ecb9920813a25ccb26cce2b45dda6d8b3",
"src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/references/papers-20261015/cases/migration-is-not-free-memory.md": "2ac168f2725cba7fc92dfcc67c01bf2258dff6a2dbe1ddb063bc60c175d82d0a",
"src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/references/papers-20261015/cases/parallelism-change-with-state-cost.md": "5accc81086e4500d66f894ef6af9951e1cbb76941f6c631aab5c5561a786d8d8",
@@ -106,12 +106,33 @@
"references/papers-20261015/references/orca-readset.md",
"references/source-notes.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "applicability": {
+ "scope": "Pinned vLLM V1 branch-based diagnosis of queued requests, low scheduled-token counts and KV-pressure preemption, separating admission/token/input/allocation/connector-wait constraints.",
+ "prerequisites": "The deployed scheduler must be mapped to the pinned branches, with effective configuration and a bounded request/step trace; another revision requires re-establishing that correspondence.",
+ "input_contract": "Deployed revision, scheduler class/configuration, connector role, KV layout, policy and speculative/encoder/concurrent-batch features; per-step queues, tokens, both budgets, allocation results, victims, transfer/free state and timing.",
+ "produced_artifact": "A per-step accounting ledger, observed limiting branch and competing explanation, missing instrumentation and one conditional configuration/progress hypothesis with predicted effects and trade-offs.",
+ "failure_modes": "Inferring memory exhaustion from a queue or high KV usage, conflating token/input budgets, ignoring pending frees or reading async token accounting as load completion can attribute admission stalls to the wrong branch.",
+ "refusal_point": "When required state is missing, leave the diagnosis unresolved and request the smallest trace; escalate allocation or connector behavior outside the readset and stop before live mutation or performance claims."
+ },
+ "recommendations": [
+ "Trace the first actual blocking or skip branch and record running versus admission caps and token versus input budgets separately.",
+ "Inspect deferred frees and request-free eligibility before blaming preemption victim selection, then reconcile restored budgets for an already-scheduled victim.",
+ "Reconstruct waiting admission independently, including queue precedence and async prefix states; change no setting without a branch-specific hypothesis."
+ ]
+ },
+ "applicability": {
+ "scope": "Pinned vLLM V1 branch-based diagnosis of queued requests, low scheduled-token counts and KV-pressure preemption, separating admission/token/input/allocation/connector-wait constraints.",
+ "prerequisites": "The deployed scheduler must be mapped to the pinned branches, with effective configuration and a bounded request/step trace; another revision requires re-establishing that correspondence.",
+ "input_contract": "Deployed revision, scheduler class/configuration, connector role, KV layout, policy and speculative/encoder/concurrent-batch features; per-step queues, tokens, both budgets, allocation results, victims, transfer/free state and timing.",
+ "produced_artifact": "A per-step accounting ledger, observed limiting branch and competing explanation, missing instrumentation and one conditional configuration/progress hypothesis with predicted effects and trade-offs.",
+ "failure_modes": "Inferring memory exhaustion from a queue or high KV usage, conflating token/input budgets, ignoring pending frees or reading async token accounting as load completion can attribute admission stalls to the wrong branch.",
+ "refusal_point": "When required state is missing, leave the diagnosis unresolved and request the smallest trace; escalate allocation or connector behavior outside the readset and stop before live mutation or performance claims."
},
- "applicability": {},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Trace the first actual blocking or skip branch and record running versus admission caps and token versus input budgets separately.",
+ "Inspect deferred frees and request-free eligibility before blaming preemption victim selection, then reconcile restored budgets for an already-scheduled victim.",
+ "Reconstruct waiting admission independently, including queue precedence and async prefix states; change no setting without a branch-specific hypothesis."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diffusers-invocation-state.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diffusers-invocation-state.json
index b01184e87..3da90d49b 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diffusers-invocation-state.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diffusers-invocation-state.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/SKILL.md",
"src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/metadata.json",
@@ -42,7 +42,7 @@
"checkout_ref": "checkout:diffusers-invocation-state",
"source_files": {
"src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/SKILL.md": "7581b56336d07213a87d01477793ae1a8acf0b7cf876fce36c9a761ee32c1ff4",
- "src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/metadata.json": "68ac7f64b7ab947fc2db664bc3990f102c1cc5539b42350aa5818d0bbf17d86c",
+ "src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/metadata.json": "66f8d20c1644aba964adfff3b439bfd3ecd3875885191e6ce10721bc50877b1a",
"src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/references/ownership.md": "40688cc44183027efbf2fca8d3548c299cdf2e33fef54f76ab3549353eb9bd1b",
"src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/references/source-map.md": "10562a75683bd9bd07ec2e37a1cbd0f93f2534d6f85b55a566adb2d97485215f",
"src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/references/validation.md": "5a0b854b87d091d7d505b0bdd5c03330052507a851ab6edf35b93a6675f2d4f7",
@@ -84,7 +84,12 @@
}
],
"applicability": {
- "scope": "FluxPipeline and FlowMatchEulerDiscreteScheduler invocation ownership at this source pin; inspect the actual caller contract."
+ "scope": "FluxPipeline and FlowMatchEulerDiscreteScheduler invocation ownership at this source pin; inspect the actual caller contract.",
+ "prerequisites": "Access to the pinned ownership path and an identified caller admission/serialization contract; other pipeline or scheduler classes need their own contracts.",
+ "input_contract": "Diffusers/PyTorch revisions, pipeline/scheduler classes and object identities, construction/from_pipe path, actual components, scheduler config and resolved timesteps/sigmas, shape/dtype, callback/interrupt timing, output/cleanup path and overlapping intervals.",
+ "produced_artifact": "An object-identity/ownership map and call-to-schedule-reset/step/interrupt/output/cleanup trace, the caller admission contract, one next check and actual evidence status.",
+ "failure_modes": "Assuming from_pipe deep-copies scheduler state, passing an enumeration index instead of a scheduler timestep, or treating cooperative iteration skipping as kernel cancellation/exception cleanup can misidentify invocation ownership.",
+ "refusal_point": "Stop when the ownership question is resolved or a required fixture/backend is absent; do not infer service-wide race or serialization behavior from this readset."
},
"limitations": [
"Source-derived candidate; no installed API, native scheduler, model, concurrency or service validation.",
@@ -94,14 +99,25 @@
"provenance_readsets": [
"references/source-map.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Check wrapper and scheduler identities explicitly and follow the effective schedule reset and step-index progression.",
+ "Trace interruption through its loop-boundary consumer and normal decode/cleanup tail while preserving the caller's existing shared-state serialization.",
+ "Use a real scheduler CPU check only for its scheduler cycle; keep pipeline reuse/interruption and serving claims tied to matching fixtures and execution evidence."
+ ]
},
"applicability": {
- "scope": "FluxPipeline and FlowMatchEulerDiscreteScheduler invocation ownership at this source pin; inspect the actual caller contract."
+ "scope": "FluxPipeline and FlowMatchEulerDiscreteScheduler invocation ownership at this source pin; inspect the actual caller contract.",
+ "prerequisites": "Access to the pinned ownership path and an identified caller admission/serialization contract; other pipeline or scheduler classes need their own contracts.",
+ "input_contract": "Diffusers/PyTorch revisions, pipeline/scheduler classes and object identities, construction/from_pipe path, actual components, scheduler config and resolved timesteps/sigmas, shape/dtype, callback/interrupt timing, output/cleanup path and overlapping intervals.",
+ "produced_artifact": "An object-identity/ownership map and call-to-schedule-reset/step/interrupt/output/cleanup trace, the caller admission contract, one next check and actual evidence status.",
+ "failure_modes": "Assuming from_pipe deep-copies scheduler state, passing an enumeration index instead of a scheduler timestep, or treating cooperative iteration skipping as kernel cancellation/exception cleanup can misidentify invocation ownership.",
+ "refusal_point": "Stop when the ownership question is resolved or a required fixture/backend is absent; do not infer service-wide race or serialization behavior from this readset."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Check wrapper and scheduler identities explicitly and follow the effective schedule reset and step-index progression.",
+ "Trace interruption through its loop-boundary consumer and normal decode/cleanup tail while preserving the caller's existing shared-state serialization.",
+ "Use a real scheduler CPU check only for its scheduler cycle; keep pipeline reuse/interruption and serving claims tied to matching fixtures and execution evidence."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/draft-acceptance-issue.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/draft-acceptance-issue.json
index 00d84e827..984ba5344 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/draft-acceptance-issue.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/draft-acceptance-issue.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/workflows/draft-acceptance-issue/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/workflows/draft-acceptance-issue/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/skill-index.json",
"src/inferrsi/data/skill_library/workflows/draft-acceptance-issue/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dwarfstar.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dwarfstar.json
index f8984451a..aeddd1d88 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dwarfstar.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dwarfstar.json
@@ -25,8 +25,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/repositories/dwarfstar/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/repositories/dwarfstar/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/repositories/dwarfstar/LICENSE",
"src/inferrsi/data/skill_library/repositories/dwarfstar/SKILL.md",
@@ -57,7 +57,7 @@
"source_files": {
"src/inferrsi/data/skill_library/repositories/dwarfstar/LICENSE": "4a8a1358f5b048627d71e4988de0bafd386f4988991929629e6cb9e29b7b9b2f",
"src/inferrsi/data/skill_library/repositories/dwarfstar/SKILL.md": "a04a252ecdcd4710a3273595f4bd1f3eb8e24d5bb1bfe005f058a78388254dc9",
- "src/inferrsi/data/skill_library/repositories/dwarfstar/metadata.json": "e1f0c2dc3d7ea9278039c2e8c3e8ca4f8e3aba33504b7784d628e903f0b5fc8c",
+ "src/inferrsi/data/skill_library/repositories/dwarfstar/metadata.json": "c883f6bc05e585f072cf03f734b7a296a65842c7921d3afcfe89e1977405fbd2",
"src/inferrsi/data/skill_library/repositories/dwarfstar/references/repo-provenance.md": "7d72f9c85ee60fe003f5f5a585b52e9989ebcedca625b584fa39a976d4433998",
"src/inferrsi/data/skill_library/repositories/dwarfstar/references/repo-routing-metadata.json": "8e195b420b67151932af5f435c284f49677cc7863e650c83cfa9e762fd00db0b",
"src/inferrsi/data/skill_library/repositories/dwarfstar/references/source-identity.json": "a38fe07d003accd83b7626d7e189f7b49b4b73e7a223e4d441b0e33add5d9c91",
@@ -154,7 +154,12 @@
"method": "User-approved native C adaptation and partial drafting; manual stable-ID subskill orchestration and integrated create/verify stages. Required-backend blocks disable automatic admission."
},
"applicability": {
- "scope": "Pinned DwarfStar source investigation and experiment planning; not arbitrary GGUF support or cross-backend qualification."
+ "scope": "Pinned DwarfStar source investigation and experiment planning; not arbitrary GGUF support or cross-backend qualification.",
+ "prerequisites": "Access to the pinned native-source contract and relevant memory/state/qualification route; native verification needs a compatible build, matching hardware/weights and task-specific execution authority.",
+ "input_contract": "Engine revision/patches, actual model artifact/representation, selected backend/build and topology, memory/storage budget, context/chunk/session count, image inputs, sampling/speculation policy, cache state and quality/service requirements.",
+ "produced_artifact": "A bounded executing-tuple report with observed signal, source-backed mechanism, alternate explanation, proposed falsifier, actual check outcome and unresolved backend/model/service gap.",
+ "failure_modes": "Inferring model support from flags, aliases, file mapping or quantization names can select the wrong graph; treating CPU bookkeeping or parser checks as native backend/service qualification can overstate the result.",
+ "refusal_point": "Stop at unavailable hardware/weights, unsafe exposure or the authorized budget; retain memory-guard refusals rather than disabling the guard to claim fit."
},
"limitations": [
"Metal/CUDA/ROCm inference, actual model output, SSD runtime I/O, GPU recurrent rollback, physical distributed behavior, service quality/performance and skill efficacy remain unverified.",
@@ -166,14 +171,25 @@
"references/repo-provenance.md",
"references/source-identity.json"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Choose the memory, serving-state or qualification route by the observed symptom and identify the owner, validity key and release/rollback boundary.",
+ "Freeze comparison controls, test the smallest invariant first and retain fallback messages, cancellations, refusals and unavailable fixtures.",
+ "Base model support on the actual loaded artifact and graph; use native backend evidence for model, tensor-parallel or service conclusions."
+ ]
},
"applicability": {
- "scope": "Pinned DwarfStar source investigation and experiment planning; not arbitrary GGUF support or cross-backend qualification."
+ "scope": "Pinned DwarfStar source investigation and experiment planning; not arbitrary GGUF support or cross-backend qualification.",
+ "prerequisites": "Access to the pinned native-source contract and relevant memory/state/qualification route; native verification needs a compatible build, matching hardware/weights and task-specific execution authority.",
+ "input_contract": "Engine revision/patches, actual model artifact/representation, selected backend/build and topology, memory/storage budget, context/chunk/session count, image inputs, sampling/speculation policy, cache state and quality/service requirements.",
+ "produced_artifact": "A bounded executing-tuple report with observed signal, source-backed mechanism, alternate explanation, proposed falsifier, actual check outcome and unresolved backend/model/service gap.",
+ "failure_modes": "Inferring model support from flags, aliases, file mapping or quantization names can select the wrong graph; treating CPU bookkeeping or parser checks as native backend/service qualification can overstate the result.",
+ "refusal_point": "Stop at unavailable hardware/weights, unsafe exposure or the authorized budget; retain memory-guard refusals rather than disabling the guard to claim fit."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Choose the memory, serving-state or qualification route by the observed symptom and identify the owner, validity key and release/rollback boundary.",
+ "Freeze comparison controls, test the smallest invariant first and retain fallback messages, cancellations, refusals and unavailable fixtures.",
+ "Base model support on the actual loaded artifact and graph; use native backend evidence for model, tensor-parallel or service conclusions."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/evaluate-elastic-serving-placement.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/evaluate-elastic-serving-placement.json
index d4b8c68d9..f82d83914 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/evaluate-elastic-serving-placement.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/evaluate-elastic-serving-placement.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/methods/evaluate-elastic-serving-placement/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/methods/evaluate-elastic-serving-placement/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/methods/evaluate-elastic-serving-placement/SKILL.md",
"src/inferrsi/data/skill_library/methods/evaluate-elastic-serving-placement/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/generation-cache-lifecycle.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/generation-cache-lifecycle.json
index bc1749570..20aa41858 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/generation-cache-lifecycle.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/generation-cache-lifecycle.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/SKILL.md",
"src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/metadata.json",
@@ -43,7 +43,7 @@
"checkout_ref": "checkout:generation-cache-lifecycle",
"source_files": {
"src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/SKILL.md": "bf1177cb9985c12a9bf11edb64771018e1e314c4713920de8fd4a989ff70115f",
- "src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/metadata.json": "d86cecb615e93ba0f6be09d1ead102cd5a17068c07e1be2549ac5cd8f7519279",
+ "src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/metadata.json": "c2e30a2a3365dd27cac7916dc26aa356f1277d07ebee891c173ee37b77111fa1",
"src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/references/cache-stop.md": "bba6b5c73261065309cbcf11b1e591eef48916f0c79ceaaf74d98a9fa96b0992",
"src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/references/generation-batch.md": "f16d3c57d45e63bcaffd3871aa41c76972ee11f7dbf1e12b45d9b170303cad9b",
"src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/references/source-map.md": "91d26181f246307f2113f94b12ac9587d689b654dcd216e9f14b11cf5ebb650c",
@@ -89,7 +89,12 @@
}
],
"applicability": {
- "scope": "MLX-LM generation, batch-row and prefix-cache diagnosis; model/backend/runtime compatibility requires separate qualification."
+ "scope": "MLX-LM generation, batch-row and prefix-cache diagnosis; model/backend/runtime compatibility requires separate qualification.",
+ "prerequisites": "Access to the independently pinned MLX-LM generation/cache contracts and the selected entry path; model, backend and runtime compatibility is separately established for any native claim.",
+ "input_contract": "MLX-LM/MLX revisions and builds, backend/device, model/tokenizer/adapter/draft identities, entry point, tokenized prompts, cache classes, batching/prefill limits, per-request sampling/processors/stops, seed and KV quantization; failure traces include UIDs, stages, lengths, terminal reason and next-request result.",
+ "produced_artifact": "An entry-path decision, UID/stage/cache-row table, prefix identity and cache-class assumptions, terminal/cleanup trace and actual evidence status with a targeted verification plan.",
+ "failure_modes": "Ignoring server batching gates, filtering only some per-request row state, reusing a text-only prefix key or treating a stop flag as immediate MLX work cancellation can corrupt lifecycle reasoning.",
+ "refusal_point": "Stop at the first unsupported contract or missing required fixture/backend; retain a targeted plan without inferring paged allocation, radix sharing or instantaneous prefill interruption."
},
"limitations": [
"Source-derived candidate; no local native-backend, model, service or efficacy validation.",
@@ -98,14 +103,25 @@
"provenance_readsets": [
"references/source-map.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Resolve the library versus server entry and its batching gates before transferring BatchGenerator behavior.",
+ "Keep UID, caches, tokens, samplers, processors, lengths, stop matchers and pending outputs aligned under the same removal selection, including prompt-stage requests.",
+ "Check exact token/model/adapter/draft identity and cache-type copy/trim semantics, then trace cooperative removal through survivor and next-request behavior."
+ ]
},
"applicability": {
- "scope": "MLX-LM generation, batch-row and prefix-cache diagnosis; model/backend/runtime compatibility requires separate qualification."
+ "scope": "MLX-LM generation, batch-row and prefix-cache diagnosis; model/backend/runtime compatibility requires separate qualification.",
+ "prerequisites": "Access to the independently pinned MLX-LM generation/cache contracts and the selected entry path; model, backend and runtime compatibility is separately established for any native claim.",
+ "input_contract": "MLX-LM/MLX revisions and builds, backend/device, model/tokenizer/adapter/draft identities, entry point, tokenized prompts, cache classes, batching/prefill limits, per-request sampling/processors/stops, seed and KV quantization; failure traces include UIDs, stages, lengths, terminal reason and next-request result.",
+ "produced_artifact": "An entry-path decision, UID/stage/cache-row table, prefix identity and cache-class assumptions, terminal/cleanup trace and actual evidence status with a targeted verification plan.",
+ "failure_modes": "Ignoring server batching gates, filtering only some per-request row state, reusing a text-only prefix key or treating a stop flag as immediate MLX work cancellation can corrupt lifecycle reasoning.",
+ "refusal_point": "Stop at the first unsupported contract or missing required fixture/backend; retain a targeted plan without inferring paged allocation, radix sharing or instantaneous prefill interruption."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Resolve the library versus server entry and its batching gates before transferring BatchGenerator behavior.",
+ "Keep UID, caches, tokens, samplers, processors, lengths, stop matchers and pending outputs aligned under the same removal selection, including prompt-stage requests.",
+ "Check exact token/model/adapter/draft identity and cache-type copy/trim semantics, then trace cooperative removal through survivor and next-request behavior."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-memory-management.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-memory-management.json
index 18d9ae726..2eb884e41 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-memory-management.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-memory-management.json
@@ -30,8 +30,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/hardware/host/host-memory-management/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/hardware/host/host-memory-management/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/hardware/host/host-memory-management/SKILL.md",
"src/inferrsi/data/skill_library/hardware/host/host-memory-management/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-overhead-optimization.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-overhead-optimization.json
index 053898b1b..55f647afb 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-overhead-optimization.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-overhead-optimization.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/hardware/host/host-overhead-optimization/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/hardware/host/host-overhead-optimization/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/hardware/host/host-overhead-optimization/SKILL.md",
"src/inferrsi/data/skill_library/hardware/host/host-overhead-optimization/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/husky.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/husky.json
index 7cf060d13..57713b47c 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/husky.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/husky.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/repositories/husky/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/repositories/husky/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/repositories/husky/SKILL.md",
"src/inferrsi/data/skill_library/repositories/husky/metadata.json",
@@ -44,7 +44,7 @@
"checkout_ref": "checkout:husky",
"source_files": {
"src/inferrsi/data/skill_library/repositories/husky/SKILL.md": "ac86afdabf9adffb684cb635971934438724bdfde8628ec549727ea23c40cf3b",
- "src/inferrsi/data/skill_library/repositories/husky/metadata.json": "9b720bc6fcb65b6dbddba36d6187b4bee86346a1bc633f55a75bd43dbe4fa04e",
+ "src/inferrsi/data/skill_library/repositories/husky/metadata.json": "b1cb0a5be02d1170125bfc3e16c044b9c1463a611a14d36fac40638524fc2e52",
"src/inferrsi/data/skill_library/repositories/husky/references/author-report.md": "4c1351ca42b2535a7c03c4da87706040650ef10d86eadcdb496e881a147fc0f7",
"src/inferrsi/data/skill_library/repositories/husky/references/repo-provenance.md": "37de912e2e2a36b44233560e6a8368b994428bff587ed4f04b1cb5fbaba0a9e7",
"src/inferrsi/data/skill_library/repositories/husky/references/source-availability.json": "d3523e53c89d4aee5bfaa17db84501aa4c4eddd59bcee486df5d48d3aa48173d",
@@ -105,7 +105,12 @@
"method": "Single coherent research workflow, bounded source summary, original verification questions and offline schema checks. Repository/environment/native stages remain unavailable."
},
"applicability": {
- "scope": "Review Husky document claims and specify missing evidence; common catalog scope is not inference backend compatibility."
+ "scope": "Review Husky document claims and specify missing evidence; common catalog scope is not inference backend compatibility.",
+ "prerequisites": "Access to the bounded official-document inventory and verification protocol; implementation and native reproduction remain unavailable until separately supplied and authenticated.",
+ "input_contract": "The exact claim, source/view date, workload and comparison intent, supporting evidence classes and available implementation/model/output/measurement artifacts; unknown identity, licensing and availability remain explicit.",
+ "produced_artifact": "A reviewable claim/evidence request recording source availability, mechanism hypothesis, counter-signal, missing identities/artifacts, proposed falsifying check and actual result or bounded INCONCLUSIVE conclusion.",
+ "failure_modes": "Treating a public-page reading as code inspection, proposed extensions as shipped behavior, summed independent ratios as one gain, or a structurally valid self-reported card as authentic evidence can overstate qualification.",
+ "refusal_point": "Stop at missing source/receipts, unsafe or private data, unsupported resources or the authorized budget; unavailable artifacts yield an unverified/INCONCLUSIVE research conclusion."
},
"limitations": [
"No canonical public repository, implementation revision, retained receipts or source-code license was identified; unknowns remain unknown.",
@@ -119,14 +124,25 @@
"references/repo-provenance.md",
"references/source-availability.json"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Separate author reports, inspected implementation, executed experiments and new interpretations, including negative findings and future work.",
+ "Request immutable identities and artifacts, then state one mechanism, competing explanation and smallest ablation with frozen semantic requirements.",
+ "Use the offline card auditor only to identify structural gaps; it authenticates neither artifacts nor execution and leaves qualification unadjudicated."
+ ]
},
"applicability": {
- "scope": "Review Husky document claims and specify missing evidence; common catalog scope is not inference backend compatibility."
+ "scope": "Review Husky document claims and specify missing evidence; common catalog scope is not inference backend compatibility.",
+ "prerequisites": "Access to the bounded official-document inventory and verification protocol; implementation and native reproduction remain unavailable until separately supplied and authenticated.",
+ "input_contract": "The exact claim, source/view date, workload and comparison intent, supporting evidence classes and available implementation/model/output/measurement artifacts; unknown identity, licensing and availability remain explicit.",
+ "produced_artifact": "A reviewable claim/evidence request recording source availability, mechanism hypothesis, counter-signal, missing identities/artifacts, proposed falsifying check and actual result or bounded INCONCLUSIVE conclusion.",
+ "failure_modes": "Treating a public-page reading as code inspection, proposed extensions as shipped behavior, summed independent ratios as one gain, or a structurally valid self-reported card as authentic evidence can overstate qualification.",
+ "refusal_point": "Stop at missing source/receipts, unsafe or private data, unsupported resources or the authorized budget; unavailable artifacts yield an unverified/INCONCLUSIVE research conclusion."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Separate author reports, inspected implementation, executed experiments and new interpretations, including negative findings and future work.",
+ "Request immutable identities and artifacts, then state one mechanism, competing explanation and smallest ablation with frozen semantic requirements.",
+ "Use the offline card auditor only to identify structural gaps; it authenticates neither artifacts nor execution and leaves qualification unadjudicated."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-benchmark-design.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-benchmark-design.json
index e38e0536a..67c3eb326 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-benchmark-design.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-benchmark-design.json
@@ -30,8 +30,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/methods/inference-benchmark-design/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/methods/inference-benchmark-design/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/methods/inference-benchmark-design/SKILL.md",
"src/inferrsi/data/skill_library/methods/inference-benchmark-design/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-optimization-casebook.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-optimization-casebook.json
index 726c8467f..ebf094485 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-optimization-casebook.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-optimization-casebook.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/methods/inference-optimization-casebook/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/methods/inference-optimization-casebook/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/methods/inference-optimization-casebook/SKILL.md",
"src/inferrsi/data/skill_library/methods/inference-optimization-casebook/cases/admission-drain-occupancy.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kernel-writing-guide.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kernel-writing-guide.json
index 571129076..8c9881839 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kernel-writing-guide.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kernel-writing-guide.json
@@ -39,8 +39,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/methods/kernel-writing-guide/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/methods/kernel-writing-guide/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/methods/kernel-writing-guide/SKILL.md",
"src/inferrsi/data/skill_library/methods/kernel-writing-guide/cases/causal-conv1d-dispatch-state.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kv-tier-management.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kv-tier-management.json
index 427697464..5a1aba13c 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kv-tier-management.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kv-tier-management.json
@@ -24,8 +24,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/repositories/lmcache/kv-tier-management/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/repositories/lmcache/kv-tier-management/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/repositories/lmcache/kv-tier-management/SKILL.md",
"src/inferrsi/data/skill_library/repositories/lmcache/kv-tier-management/metadata.json",
@@ -41,7 +41,7 @@
"checkout_ref": "checkout:kv-tier-management",
"source_files": {
"src/inferrsi/data/skill_library/repositories/lmcache/kv-tier-management/SKILL.md": "e4b78f368d70c8abb74a935acb910414015f1aeec1b50f7a295e836822ec4b27",
- "src/inferrsi/data/skill_library/repositories/lmcache/kv-tier-management/metadata.json": "c221189b8d85fac3bd41ef1c7cd560f1aebc8cfa0381c182d1d571a7169be829",
+ "src/inferrsi/data/skill_library/repositories/lmcache/kv-tier-management/metadata.json": "85f7242b83334af11f57799875738b498bfe4680a51f36727792dfbe91806704",
"src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
},
"skill_metadata": {
@@ -69,12 +69,31 @@
"Original typed source inventory is not normalized; retain source identities and limits in the procedure/readsets."
],
"provenance_readsets": [],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "applicability": {
+ "scope": "LMCache multiprocess runtime KV tier policy across engine GPU cache, host L1 and configured L2 adapters, covering store/retention, prefetch admission, eviction and RAM pressure; startup loading and checkpoint recovery have separate owners.",
+ "prerequisites": "The deployment uses the inspected MP-mode semantics rather than single-process knobs; selected GDS/DAX/raw-block paths need their own platform and adapter prerequisites. Lazy offload additionally requires vLLM prefix caching.",
+ "input_contract": "MP versus single-process mode, store/retention policy and rationale, L1 manager capacities and eviction settings, L2 adapter capacities/eviction, chunk-size distribution and prefetch in-flight setting, together with token-level hit and pressure counters.",
+ "produced_artifact": "A mode- and adapter-specific tier-policy choice with retention rationale, an ordered RAM-release plan and a verification record separating L1/L2 token hits, prefetch pressure, store lag and per-manager usage.",
+ "failure_modes": "Mixing MP and in-process controls, treating key-count eviction as a byte fraction for unequal chunks, omitting fs capacity management, or increasing count-based prefetch admission without tracking L1 pressure can make the policy or memory claim unusable."
+ },
+ "recommendations": [
+ "Choose store retention and prefetched-key retention separately; use skip_l1 with noop only for buffer-only operation and account for failed stores remaining in L1.",
+ "Tune the count-based prefetch in-flight limit against L1 allocation failures and in-flight bytes; eager or warm prefetch can raise pressure before useful reuse.",
+ "Record per-adapter capacity/eviction and actual chunk-size uniformity; report token-level L1/L2 hits and sum per-l1_tag bytes rather than ratios."
+ ]
+ },
+ "applicability": {
+ "scope": "LMCache multiprocess runtime KV tier policy across engine GPU cache, host L1 and configured L2 adapters, covering store/retention, prefetch admission, eviction and RAM pressure; startup loading and checkpoint recovery have separate owners.",
+ "prerequisites": "The deployment uses the inspected MP-mode semantics rather than single-process knobs; selected GDS/DAX/raw-block paths need their own platform and adapter prerequisites. Lazy offload additionally requires vLLM prefix caching.",
+ "input_contract": "MP versus single-process mode, store/retention policy and rationale, L1 manager capacities and eviction settings, L2 adapter capacities/eviction, chunk-size distribution and prefetch in-flight setting, together with token-level hit and pressure counters.",
+ "produced_artifact": "A mode- and adapter-specific tier-policy choice with retention rationale, an ordered RAM-release plan and a verification record separating L1/L2 token hits, prefetch pressure, store lag and per-manager usage.",
+ "failure_modes": "Mixing MP and in-process controls, treating key-count eviction as a byte fraction for unequal chunks, omitting fs capacity management, or increasing count-based prefetch admission without tracking L1 pressure can make the policy or memory claim unusable."
},
- "applicability": {},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Choose store retention and prefetched-key retention separately; use skip_l1 with noop only for buffer-only operation and account for failed stores remaining in L1.",
+ "Tune the count-based prefetch in-flight limit against L1 allocation failures and in-flight bytes; eager or warm prefetch can raise pressure before useful reuse.",
+ "Record per-adapter capacity/eviction and actual chunk-size uniformity; report token-level L1/L2 hits and sum per-l1_tag bytes rather than ratios."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lazy-evaluation-stream-lifetimes.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lazy-evaluation-stream-lifetimes.json
index 86b259c09..68d2fdcff 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lazy-evaluation-stream-lifetimes.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lazy-evaluation-stream-lifetimes.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/SKILL.md",
"src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/metadata.json",
@@ -42,7 +42,7 @@
"checkout_ref": "checkout:lazy-evaluation-stream-lifetimes",
"source_files": {
"src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/SKILL.md": "a95569bcc0da6e715fd74783a6a1e0489e92b680c582f15dafbf98df589d505a",
- "src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/metadata.json": "9c2293c8956508de987c886c7224d673d22912e7e915b026b05abdd287e97401",
+ "src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/metadata.json": "cfa57ac908025135113d04c3d64b9967f8d9a3af4a00707673d94b8227c85808",
"src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/references/execution-lifetime.md": "c4bc41e08aca68f661325f61e6f0db1760a079dc362f295c847bdc374fee3bf6",
"src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/references/source-map.md": "798b9cb89ed6b885eb47464ca9f137d2802a54571b9327e631d7742d3e5abd5d",
"src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/references/validation.md": "ebfc00bf7d6ab0a1dc415953ae84d56ccdc713d5614aff85e433c4a843793bff",
@@ -94,7 +94,12 @@
}
],
"applicability": {
- "scope": "MLX evaluation/stream/buffer diagnosis at the pinned implementation; qualify the actual backend separately."
+ "scope": "MLX evaluation/stream/buffer diagnosis at the pinned implementation; qualify the actual backend separately.",
+ "prerequisites": "Access to the pinned MLX execution/lifetime contract and a bounded reproduction with an identified evaluation/read boundary; device-dependent conclusions need the actual selected backend.",
+ "input_contract": "MLX revision/build, backend/device, Python/C++ entry, array shapes/dtypes, primitive streams, evaluation/read boundary and transformation context; timing needs warmup/work boundaries, while memory needs live references, in-flight buffers, allocator cache and synchronized observations.",
+ "produced_artifact": "A producer/evaluation/dependency-wait/backend/availability-cleanup trace, ownership table, observed symptom and one discriminating check with build-bound actual evidence status.",
+ "failure_modes": "Timing graph construction or submission as execution, assuming evaluated means completed, synchronizing only the default stream, or treating graph detachment/Python deletion as reusable storage can misdiagnose time or memory.",
+ "refusal_point": "Stop when the trace answers the question; if the required backend is absent, retain the source-supported diagnosis and exact runtime gap rather than substituting CPU import success."
},
"limitations": [
"Source-derived candidate; no local native-backend, model, service or efficacy validation.",
@@ -103,14 +108,25 @@
"provenance_readsets": [
"references/source-map.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Separate graph construction, async_eval submission, eval and host reads, and record which stream the measurement actually synchronizes.",
+ "Preserve the dependency-wait path and selected backend buffer-retention/donation contract before changing streams or assuming storage can be reused.",
+ "Distinguish live, in-flight and allocator-cached memory; inspect the real error/cleanup path before proposing a leak fix or hiding the failure."
+ ]
},
"applicability": {
- "scope": "MLX evaluation/stream/buffer diagnosis at the pinned implementation; qualify the actual backend separately."
+ "scope": "MLX evaluation/stream/buffer diagnosis at the pinned implementation; qualify the actual backend separately.",
+ "prerequisites": "Access to the pinned MLX execution/lifetime contract and a bounded reproduction with an identified evaluation/read boundary; device-dependent conclusions need the actual selected backend.",
+ "input_contract": "MLX revision/build, backend/device, Python/C++ entry, array shapes/dtypes, primitive streams, evaluation/read boundary and transformation context; timing needs warmup/work boundaries, while memory needs live references, in-flight buffers, allocator cache and synchronized observations.",
+ "produced_artifact": "A producer/evaluation/dependency-wait/backend/availability-cleanup trace, ownership table, observed symptom and one discriminating check with build-bound actual evidence status.",
+ "failure_modes": "Timing graph construction or submission as execution, assuming evaluated means completed, synchronizing only the default stream, or treating graph detachment/Python deletion as reusable storage can misdiagnose time or memory.",
+ "refusal_point": "Stop when the trace answers the question; if the required backend is absent, retain the source-supported diagnosis and exact runtime gap rather than substituting CPU import success."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Separate graph construction, async_eval submission, eval and host reads, and record which stream the measurement actually synchronizes.",
+ "Preserve the dependency-wait path and selected backend buffer-retention/donation contract before changing streams or assuming storage can be reused.",
+ "Distinguish live, in-flight and allocator-cached memory; inspect the real error/cleanup path before proposing a leak fix or hiding the failure."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lily.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lily.json
index 49cd72067..00e92ede8 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lily.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lily.json
@@ -25,8 +25,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/repositories/lily/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/repositories/lily/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/repositories/lily/LICENSE",
"src/inferrsi/data/skill_library/repositories/lily/NOTICE.md",
@@ -61,7 +61,7 @@
"src/inferrsi/data/skill_library/repositories/lily/LICENSE": "b9e2b50a48a09fb39dafa71ab1a11a6e87389addda17faa99d55f5f74a0001ac",
"src/inferrsi/data/skill_library/repositories/lily/NOTICE.md": "7a01e666044fb8aca1fd88c636bca43bcc95b8b1fecbe101dce4c7c80fbb5759",
"src/inferrsi/data/skill_library/repositories/lily/SKILL.md": "db815674d5d36964902041d310d3df844d00ab2e6bef0ce4b72ed2696ab2b9aa",
- "src/inferrsi/data/skill_library/repositories/lily/metadata.json": "6d9d54e84f957881cc26894eb82ae666f106cfc6c505924f9a6fb6b35eab457f",
+ "src/inferrsi/data/skill_library/repositories/lily/metadata.json": "52f01d58613e1a93e9cfd958ceca9ab5061baa07c5f94572393520434892b420",
"src/inferrsi/data/skill_library/repositories/lily/references/repo-provenance.md": "96182f6cb3666f89adf1aed6e216c0a84ce3a04a0685a4257e286100ba8ee367",
"src/inferrsi/data/skill_library/repositories/lily/references/repo-routing-metadata.json": "f2f3becb51056e03081c7c4140e19572a6318c3d63015884df56eb49d38a00fc",
"src/inferrsi/data/skill_library/repositories/lily/references/source-identity.json": "4595559c11611c855be150919fca3eeb89e7f96c7a8dd133b9f79a5cd8e05e62",
@@ -177,7 +177,12 @@
"method": "User-approved native Rust/Metal and partial drafting adaptation; stable-ID lanes, integrated source graph and assertion/helper verification. Missing compiler/backend/checkpoint remains explicit."
},
"applicability": {
- "scope": "Exact Lily source/checkpoint/layout/API/cache/qualification investigation; common catalog classification is not CPU inference or cross-device support."
+ "scope": "Exact Lily source/checkpoint/layout/API/cache/qualification investigation; common catalog classification is not CPU inference or cross-device support.",
+ "prerequisites": "Access to the exact Lily source/model-layout/API contracts; native checks require the matching Rust/Metal/checkpoint environment and frozen comparison requirements.",
+ "input_contract": "Source revision, Cargo lock/toolchain, macOS/chip/GPU family, checkpoint and tensor payload identities, affine configuration/layout, tokenizer/template/stops, sequence/output budgets, greedy policy, input history, prefix-cache state and quality/measurement requirements.",
+ "produced_artifact": "A source/model/device-bound representation, serving-cache or qualification diagnosis with the observation, invariant, alternate cause, smallest falsifier, actual result and remaining native gap.",
+ "failure_modes": "Equal bit width, file size, API names or prompt_cache_key do not establish tensor/token/state equivalence; ignoring the extra lookahead slot can understate request capacity, and green CPU references do not replace matched-model Metal goldens.",
+ "refusal_point": "Stop at unsupported formats/APIs, missing compiler/backend/checkpoint, unfrozen gates or the authorized budget; do not weaken ignored/golden tests to obtain a green result."
},
"limitations": [
"Cargo/Rust1.92, supported M5/macOS26/Metal and model resources are unavailable; native evidence remains BLOCKED_REQUIRED_BACKEND.",
@@ -190,14 +195,25 @@
"references/repo-provenance.md",
"references/source-identity.json"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Resolve exact affine representation and strict API semantics, then trace token frontier, cache key and restore/release ownership.",
+ "Retain the source's extra request lookahead slot in capacity arithmetic rather than counting only prompt and visible output.",
+ "Separate reported benchmark version from the checked-out runner and distinguish offline screening, native compilation, Metal operators and matched-model goldens."
+ ]
},
"applicability": {
- "scope": "Exact Lily source/checkpoint/layout/API/cache/qualification investigation; common catalog classification is not CPU inference or cross-device support."
+ "scope": "Exact Lily source/checkpoint/layout/API/cache/qualification investigation; common catalog classification is not CPU inference or cross-device support.",
+ "prerequisites": "Access to the exact Lily source/model-layout/API contracts; native checks require the matching Rust/Metal/checkpoint environment and frozen comparison requirements.",
+ "input_contract": "Source revision, Cargo lock/toolchain, macOS/chip/GPU family, checkpoint and tensor payload identities, affine configuration/layout, tokenizer/template/stops, sequence/output budgets, greedy policy, input history, prefix-cache state and quality/measurement requirements.",
+ "produced_artifact": "A source/model/device-bound representation, serving-cache or qualification diagnosis with the observation, invariant, alternate cause, smallest falsifier, actual result and remaining native gap.",
+ "failure_modes": "Equal bit width, file size, API names or prompt_cache_key do not establish tensor/token/state equivalence; ignoring the extra lookahead slot can understate request capacity, and green CPU references do not replace matched-model Metal goldens.",
+ "refusal_point": "Stop at unsupported formats/APIs, missing compiler/backend/checkpoint, unfrozen gates or the authorized budget; do not weaken ignored/golden tests to obtain a green result."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Resolve exact affine representation and strict API semantics, then trace token frontier, cache key and restore/release ownership.",
+ "Retain the source's extra request lookahead slot in capacity arithmetic rather than counting only prompt and visible output.",
+ "Separate reported benchmark version from the checked-out runner and distinguish offline screening, native compilation, Metal operators and matched-model goldens."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lithos-metal.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lithos-metal.json
index 8d6b9814f..b51b9b146 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lithos-metal.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lithos-metal.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/repositories/lithos-metal/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/repositories/lithos-metal/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/repositories/lithos-metal/LICENSE",
"src/inferrsi/data/skill_library/repositories/lithos-metal/SKILL.md",
@@ -60,7 +60,7 @@
"source_files": {
"src/inferrsi/data/skill_library/repositories/lithos-metal/LICENSE": "cfc7749b96f63bd31c3c42b5c471bf756814053e847c10f3eb003417bc523d30",
"src/inferrsi/data/skill_library/repositories/lithos-metal/SKILL.md": "4c71ba2f26c275829466bba2412738aa27799842e642c75cdfd36d9ed8713f7f",
- "src/inferrsi/data/skill_library/repositories/lithos-metal/metadata.json": "ea2d4e73e33a49433a2834d19489653f828e56f69c4975cc32cf508176eb24ae",
+ "src/inferrsi/data/skill_library/repositories/lithos-metal/metadata.json": "c5fa717d6f79d54e6aa2b0c5b672e80ac7412fbe5901ae24fda78a24c584ba67",
"src/inferrsi/data/skill_library/repositories/lithos-metal/references/mlx-license.md": "ccfab7ccb2ea306f71531c8ca77bb55507606cd90768b1e32b8b52ab5b48cf01",
"src/inferrsi/data/skill_library/repositories/lithos-metal/references/notices.md": "b2c9d2481ef2c9d16b3240c951ae65db417e5d309cd06b7917db9d84fe175886",
"src/inferrsi/data/skill_library/repositories/lithos-metal/references/package-api.md": "4f3ecfabafb2a15eddf21ccf964b8c37635bbb7862418a2c404e7049c0fa2898",
@@ -132,7 +132,11 @@
}
],
"applicability": {
- "scope": "Pinned Lithos Metal 0.1.2 source/API contracts; Apple silicon Metal runtime; model and recipe qualification required separately."
+ "scope": "Pinned Lithos Metal 0.1.2 source/API contracts; Apple silicon Metal runtime; model and recipe qualification required separately.",
+ "prerequisites": "Access to the pinned package/API and selected compiler, generation or serving contracts; revision/API drift needs a fresh bounded inspection, and efficacy claims need campaign-owned gates and matching native/model evidence.",
+ "input_contract": "Actual source/package version, backend/configuration, chip/core count and OS, target/draft checkpoint identities, context and prompt-token IDs, sampling/numerical options and the affected recipe/runtime/session/cache boundary.",
+ "produced_artifact": "A routed region/dependency, committed-state or cache-identity/lifecycle ledger, followed by the smallest supported finding, missing checks and next bounded action.",
+ "failure_modes": "Treating session adoption/cache hits as generation correctness, fewer dispatches as faster rounds, another chip recipe as the default, CPU cache checks as GPU rollback or health liveness as loaded/warm model state can yield unsupported conclusions."
},
"limitations": [
"Source-derived candidate; empirical efficacy validation not_run.",
@@ -147,14 +151,24 @@
"disable-model-invocation": true,
"metadata": {
"disco-role": "operating"
- }
+ },
+ "recommendations": [
+ "Follow session replacement across recipe geometry, committed generation state and serving adoption/cache identity before interpreting a cache hit.",
+ "Keep an unfused/native or target-only control with matching decoded weights, numerical mode, row bounds and sampling policy; verify dependencies and persistent state before timing.",
+ "Distinguish package/CPU checks, synthetic Metal execution, model continuation and HTTP workload evidence; chip recipes and target-compute projections do not establish another device or E2E rate."
+ ]
},
"applicability": {
- "scope": "Pinned Lithos Metal 0.1.2 source/API contracts; Apple silicon Metal runtime; model and recipe qualification required separately."
+ "scope": "Pinned Lithos Metal 0.1.2 source/API contracts; Apple silicon Metal runtime; model and recipe qualification required separately.",
+ "prerequisites": "Access to the pinned package/API and selected compiler, generation or serving contracts; revision/API drift needs a fresh bounded inspection, and efficacy claims need campaign-owned gates and matching native/model evidence.",
+ "input_contract": "Actual source/package version, backend/configuration, chip/core count and OS, target/draft checkpoint identities, context and prompt-token IDs, sampling/numerical options and the affected recipe/runtime/session/cache boundary.",
+ "produced_artifact": "A routed region/dependency, committed-state or cache-identity/lifecycle ledger, followed by the smallest supported finding, missing checks and next bounded action.",
+ "failure_modes": "Treating session adoption/cache hits as generation correctness, fewer dispatches as faster rounds, another chip recipe as the default, CPU cache checks as GPU rollback or health liveness as loaded/warm model state can yield unsupported conclusions."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Follow session replacement across recipe geometry, committed generation state and serving adoption/cache identity before interpreting a cache hit.",
+ "Keep an unfused/native or target-only control with matching decoded weights, numerical mode, row bounds and sampling policy; verify dependencies and persistent state before timing.",
+ "Distinguish package/CPU checks, synthetic Metal execution, model continuation and HTTP workload evidence; chip recipes and target-compute projections do not establish another device or E2E rate."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/local-dream.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/local-dream.json
index 13c0f6966..3ace7c39e 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/local-dream.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/local-dream.json
@@ -29,8 +29,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/repositories/local-dream/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/repositories/local-dream/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/repositories/local-dream/SKILL.md",
"src/inferrsi/data/skill_library/repositories/local-dream/metadata.json",
@@ -53,7 +53,7 @@
"checkout_ref": "checkout:local-dream",
"source_files": {
"src/inferrsi/data/skill_library/repositories/local-dream/SKILL.md": "e51fd3c71d693070abeba57ca67dcca82c7347cb94eb98f43e7c455bcacbaa46",
- "src/inferrsi/data/skill_library/repositories/local-dream/metadata.json": "ab33602252d0f5bb0797c3e1f12466b55a82aeaff10cd24e42a99a5224c8e593",
+ "src/inferrsi/data/skill_library/repositories/local-dream/metadata.json": "63b71cc6a237d8c4e00db875d15594df09d6e3cf87a1cb25183a311b4d480804",
"src/inferrsi/data/skill_library/repositories/local-dream/references/android-lifecycle.md": "f85e5fce39f5de5f88548fd8c21978240c810f899e03ff05ed72eb3508ee5e36",
"src/inferrsi/data/skill_library/repositories/local-dream/references/native-contracts.md": "40206892f5e421f9d8edf303870f5c4aefa3e87123537178d45f438f3d04b7ce",
"src/inferrsi/data/skill_library/repositories/local-dream/references/packaging-and-abi.md": "9a19bca59e19c3bfbfc1a15f5fc756c13d954ace367812aeb000551f4484ad06",
@@ -154,7 +154,11 @@
}
],
"applicability": {
- "scope": "Selected Android app/process and native diffusion boundary contracts; common is a discovery category, not CPU/device compatibility or native qualification."
+ "scope": "Selected Android app/process and native diffusion boundary contracts; common is a discovery category, not CPU/device compatibility or native qualification.",
+ "prerequisites": "Access to the pinned Android lifecycle/native/packaging contracts and identified installed artifacts; stronger conclusions require an authorized, frozen workload on the actual selected Android/ARM64 backend.",
+ "input_contract": "Source revision, installed APK/version/flavor/ABI, native executable and loaded DiT library, QNN/Hexagon/MNN builds, model/export/weights/tokenizer identities, requested and serving configuration, and lifecycle/readiness/generation/cache observations.",
+ "produced_artifact": "A bounded failed-boundary report with counter-signals, identities observed versus declared, the smallest missing discriminator and retained INCONCLUSIVE native scope.",
+ "failure_modes": "Reading UI flags or equal declared configurations as actual dispatch/readiness, treating cancellation/SSE progress as native completion, or interpreting model-directory caches as authenticated model-revision namespaces can misidentify stale state."
},
"limitations": [
"Source-derived candidate; required Android ARM64/MNN/QNN/Hexagon/model execution BLOCKED_REQUIRED_BACKEND / not_run; qualification INCONCLUSIVE.",
@@ -169,14 +173,24 @@
"references/repo-provenance.md",
"references/source-identity.json"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Compare requested and serving model/backend/dimensions/bind address through stop/reopen and health observations; never enable host mode as a diagnostic default.",
+ "Trace the actual MNN/QNN/Anima/DiT path and separate native library/ABI identity from UI runOnCpu flags and declared SDK labels.",
+ "Keep progress, cancellation, native completion, release and saved final output separate; check prompt-cache mode/version/dimensions and textual-inversion caller guards before proposing cache changes."
+ ]
},
"applicability": {
- "scope": "Selected Android app/process and native diffusion boundary contracts; common is a discovery category, not CPU/device compatibility or native qualification."
+ "scope": "Selected Android app/process and native diffusion boundary contracts; common is a discovery category, not CPU/device compatibility or native qualification.",
+ "prerequisites": "Access to the pinned Android lifecycle/native/packaging contracts and identified installed artifacts; stronger conclusions require an authorized, frozen workload on the actual selected Android/ARM64 backend.",
+ "input_contract": "Source revision, installed APK/version/flavor/ABI, native executable and loaded DiT library, QNN/Hexagon/MNN builds, model/export/weights/tokenizer identities, requested and serving configuration, and lifecycle/readiness/generation/cache observations.",
+ "produced_artifact": "A bounded failed-boundary report with counter-signals, identities observed versus declared, the smallest missing discriminator and retained INCONCLUSIVE native scope.",
+ "failure_modes": "Reading UI flags or equal declared configurations as actual dispatch/readiness, treating cancellation/SSE progress as native completion, or interpreting model-directory caches as authenticated model-revision namespaces can misidentify stale state."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Compare requested and serving model/backend/dimensions/bind address through stop/reopen and health observations; never enable host mode as a diagnostic default.",
+ "Trace the actual MNN/QNN/Anima/DiT path and separate native library/ABI identity from UI runOnCpu flags and declared SDK labels.",
+ "Keep progress, cancellation, native completion, release and saved final output separate; check prompt-cache mode/version/dimensions and textual-inversion caller guards before proposing cache changes."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mistral-rs.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mistral-rs.json
index 742d7bd56..57bb8bd5f 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mistral-rs.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mistral-rs.json
@@ -26,8 +26,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/repositories/mistral-rs/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/repositories/mistral-rs/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/repositories/mistral-rs/LICENSE",
"src/inferrsi/data/skill_library/repositories/mistral-rs/SKILL.md",
@@ -67,7 +67,7 @@
"source_files": {
"src/inferrsi/data/skill_library/repositories/mistral-rs/LICENSE": "b40bb09c7b3b6f697a8920cf9d2cb3ff8cfcf77a0fcb6d264a92842b43c4dcae",
"src/inferrsi/data/skill_library/repositories/mistral-rs/SKILL.md": "8917b30fb86437dea7508275ccce1992194d6e0c4858d537a410bf1f900953d0",
- "src/inferrsi/data/skill_library/repositories/mistral-rs/metadata.json": "5e88e2ca64b848b8177336351e03d1fbcb5c495fd54217daa83710ef6faccff0",
+ "src/inferrsi/data/skill_library/repositories/mistral-rs/metadata.json": "3ca0ca5f3a3c9e4c3e2c7d4587940b83123cf05f4e3213b305ccdd1ae89cd679",
"src/inferrsi/data/skill_library/repositories/mistral-rs/references/apache-component-license.md": "c71d239df91726fc519c6eb72d318ec65820627232b2f796219e87dcf35d0ab4",
"src/inferrsi/data/skill_library/repositories/mistral-rs/references/paged-component-notice.md": "a1718d8c5a25f15e61bea98a158880e96faa5b7174e82c4e1470fbacb23a3559",
"src/inferrsi/data/skill_library/repositories/mistral-rs/references/repo-provenance.md": "fab87dacbe0a7298e42137cab7409f443ece0dfefba60961040031e36d96adc8",
@@ -175,7 +175,11 @@
"method": "Stable-ID bounded decode/prefill/qualification author lanes, whole-graph reconciliation, hard cases and offline helper/static/scaffold/wheel checks; compiler and actual Apple evidence not_run."
},
"applicability": {
- "scope": "Selected native Metal paged decode/prefill layout/metadata/workspace and preparation/evidence investigation. Common catalog categorization is not CPU inference or hardware certification."
+ "scope": "Selected native Metal paged decode/prefill layout/metadata/workspace and preparation/evidence investigation. Common catalog categorization is not CPU inference or hardware certification.",
+ "prerequisites": "Access to the pinned Rust/Candle/MSL decode, prefill and qualification contracts; native comparisons need the matched Rust/Apple/Candle/shader stack and separately authorized frozen references.",
+ "input_contract": "mistral.rs/Candle/native artifact revisions, Rust toolchain, Apple SDK/OS/device, actual model/cache dtype and representation, shape/strides/offsets, block table, context and execution limits.",
+ "produced_artifact": "A source/build-bound route and invariant report with actual versus declared facts, safe support result, counter-signal and smallest next native check; unavailable Metal/kernel/model evidence remains blocked.",
+ "failure_modes": "Equating element counts with physical alias/layout identity, transferring decode geometry to prefill, or treating prewarm/pipeline preparation as actual dispatch can misclassify eligibility and native correctness."
},
"limitations": [
"BLOCKED_REQUIRED_BACKEND: actual Rust/Candle/Apple Metal compilation, native tensor/cache/dispatch/parity/model/quality/service/performance evidence not_run.",
@@ -188,14 +192,24 @@
"references/repo-provenance.md",
"references/source-identity.json"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Select decode, prefill or qualification by the failed contract, then follow cache/metadata bytes from producer through view to native consumer.",
+ "Keep static predicates, pipeline preparation, kernel dispatch, numerical correctness and model/service outcomes as separate evidence steps; retain skipped or gated tests.",
+ "Use the matched Candle revision and native shader/build inputs; broad workspace tests can require model fixtures, tokens and downloads."
+ ]
},
"applicability": {
- "scope": "Selected native Metal paged decode/prefill layout/metadata/workspace and preparation/evidence investigation. Common catalog categorization is not CPU inference or hardware certification."
+ "scope": "Selected native Metal paged decode/prefill layout/metadata/workspace and preparation/evidence investigation. Common catalog categorization is not CPU inference or hardware certification.",
+ "prerequisites": "Access to the pinned Rust/Candle/MSL decode, prefill and qualification contracts; native comparisons need the matched Rust/Apple/Candle/shader stack and separately authorized frozen references.",
+ "input_contract": "mistral.rs/Candle/native artifact revisions, Rust toolchain, Apple SDK/OS/device, actual model/cache dtype and representation, shape/strides/offsets, block table, context and execution limits.",
+ "produced_artifact": "A source/build-bound route and invariant report with actual versus declared facts, safe support result, counter-signal and smallest next native check; unavailable Metal/kernel/model evidence remains blocked.",
+ "failure_modes": "Equating element counts with physical alias/layout identity, transferring decode geometry to prefill, or treating prewarm/pipeline preparation as actual dispatch can misclassify eligibility and native correctness."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Select decode, prefill or qualification by the failed contract, then follow cache/metadata bytes from producer through view to native consumer.",
+ "Keep static predicates, pipeline preparation, kernel dispatch, numerical correctness and model/service outcomes as separate evidence steps; retain skipped or gated tests.",
+ "Use the matched Candle revision and native shader/build inputs; broad workspace tests can require model fixtures, tokens and downloads."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlc-llm.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlc-llm.json
index 0a5e0563c..d59cc4aa8 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlc-llm.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlc-llm.json
@@ -27,8 +27,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/repositories/mlc-llm/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/repositories/mlc-llm/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/repositories/mlc-llm/LICENSE",
"src/inferrsi/data/skill_library/repositories/mlc-llm/SKILL.md",
@@ -71,7 +71,7 @@
"source_files": {
"src/inferrsi/data/skill_library/repositories/mlc-llm/LICENSE": "c71d239df91726fc519c6eb72d318ec65820627232b2f796219e87dcf35d0ab4",
"src/inferrsi/data/skill_library/repositories/mlc-llm/SKILL.md": "e0fd1accb4216191c56e7e4a8074633f36d9063727c2546e0d2eed14220d8cf4",
- "src/inferrsi/data/skill_library/repositories/mlc-llm/metadata.json": "c9927e284e44aa09733424c024409ec8efc9593520418bb7d2e53ef301363681",
+ "src/inferrsi/data/skill_library/repositories/mlc-llm/metadata.json": "f20ca326ca0966e8d56b5fa698f971001a4af77370a145631e98476601610fa9",
"src/inferrsi/data/skill_library/repositories/mlc-llm/references/mlc-notice.md": "3a7dfb37bb96ce6f5a1cc0d6d4c8da3b8a35e6ec4692647ba0a37f6155c7e694",
"src/inferrsi/data/skill_library/repositories/mlc-llm/references/repo-provenance.md": "423f21e1de6d7ab2e11c0bdf540357efde025ae4535bdf76ed9770b1a59ba35a",
"src/inferrsi/data/skill_library/repositories/mlc-llm/references/repo-routing-metadata.json": "5309c27e7b60c4fe7aaef28bf6911e7e3c936b58b1a907aa8486bf0e40b4c0d0",
@@ -186,7 +186,11 @@
}
],
"applicability": {
- "scope": "Selected artifact/Apple target/pinned TVM Metal consumer and KV configuration source contracts. Common catalog categorization is not CPU inference or Apple qualification."
+ "scope": "Selected artifact/Apple target/pinned TVM Metal consumer and KV configuration source contracts. Common catalog categorization is not CPU inference or Apple qualification.",
+ "prerequisites": "Access to the pinned MLC and exact TVM fork/gitlink contracts; native checks require separately prepared matching MLC/TVM/FFI and the applicable Apple SDK/device stack.",
+ "input_contract": "MLC revision, TVM gitlink/build, model/tokenizer/quantized-parameter identity, target/host/SDK, compiled artifact/format, runtime device and the actual compiled/runtime configuration sources.",
+ "produced_artifact": "An identity and actual-versus-declared boundary report for artifact, Metal runtime or KV-limit ownership, with a negative signal and smallest missing native discriminator.",
+ "failure_modes": "Treating a schema hash as payload authentication, an opt-in WebLLM artifact contract as a native-engine ABI, serializable Metal fallback as runnable, or automatic capacity estimates as free memory can create unsupported qualification claims."
},
"limitations": [
"Source-derived candidate; original compiled CPU/native tests, compiler/live API/CLI/model/Metal/fit/quality/service/performance execution not_run; required native BLOCKED_REQUIRED_BACKEND.",
@@ -200,14 +204,24 @@
"references/repo-provenance.md",
"references/source-identity.json"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Identify the interface's actual consumer before using a valid artifact manifest as evidence of native model/task support.",
+ "Separate codegen/export from runtime factory/load, pipeline creation, dispatch and stream completion; a serializable kind=metal fallback can still be non-runnable.",
+ "Trace runtime preference, compile-time bounds, nested workspace inputs and explicit overrides separately before interpreting a KV or working-set estimate."
+ ]
},
"applicability": {
- "scope": "Selected artifact/Apple target/pinned TVM Metal consumer and KV configuration source contracts. Common catalog categorization is not CPU inference or Apple qualification."
+ "scope": "Selected artifact/Apple target/pinned TVM Metal consumer and KV configuration source contracts. Common catalog categorization is not CPU inference or Apple qualification.",
+ "prerequisites": "Access to the pinned MLC and exact TVM fork/gitlink contracts; native checks require separately prepared matching MLC/TVM/FFI and the applicable Apple SDK/device stack.",
+ "input_contract": "MLC revision, TVM gitlink/build, model/tokenizer/quantized-parameter identity, target/host/SDK, compiled artifact/format, runtime device and the actual compiled/runtime configuration sources.",
+ "produced_artifact": "An identity and actual-versus-declared boundary report for artifact, Metal runtime or KV-limit ownership, with a negative signal and smallest missing native discriminator.",
+ "failure_modes": "Treating a schema hash as payload authentication, an opt-in WebLLM artifact contract as a native-engine ABI, serializable Metal fallback as runnable, or automatic capacity estimates as free memory can create unsupported qualification claims."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Identify the interface's actual consumer before using a valid artifact manifest as evidence of native model/task support.",
+ "Separate codegen/export from runtime factory/load, pipeline creation, dispatch and stream completion; a serializable kind=metal fallback can still be non-runnable.",
+ "Trace runtime preference, compile-time bounds, nested workspace inputs and explicit overrides separately before interpreting a KV or working-set estimate."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlx-swift-lm.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlx-swift-lm.json
index 9d286e066..e9ee79887 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlx-swift-lm.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlx-swift-lm.json
@@ -27,8 +27,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/repositories/mlx-swift-lm/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/repositories/mlx-swift-lm/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/repositories/mlx-swift-lm/LICENSE",
"src/inferrsi/data/skill_library/repositories/mlx-swift-lm/SKILL.md",
@@ -62,7 +62,7 @@
"source_files": {
"src/inferrsi/data/skill_library/repositories/mlx-swift-lm/LICENSE": "7b1d86acac816ce8192209d7e01141c129b4d48d474c15cc5fb4af62f497c323",
"src/inferrsi/data/skill_library/repositories/mlx-swift-lm/SKILL.md": "501f73e7b327b04b6b80501b42b20dab1ac15c875817e6f4c4b66fe3e69b4fc4",
- "src/inferrsi/data/skill_library/repositories/mlx-swift-lm/metadata.json": "014131f345a28d3367313ed5a4cce858fe4c93d4d47a5f03eb89f609592e032f",
+ "src/inferrsi/data/skill_library/repositories/mlx-swift-lm/metadata.json": "6d5bd7d5f3566b75c8c0b53be95d524db039a517566826c211c978987d78b253",
"src/inferrsi/data/skill_library/repositories/mlx-swift-lm/references/repo-provenance.md": "f1b15edf89c597db0c3fc0bff7db6b2451924ce9273c39623bc0cbdccb568d31",
"src/inferrsi/data/skill_library/repositories/mlx-swift-lm/references/repo-routing-metadata.json": "8fc804347771cd768a7ed027f2041d96c9832fe5f8c5869e6593a3021b1bef8b",
"src/inferrsi/data/skill_library/repositories/mlx-swift-lm/references/scope-and-notices.md": "18e56805a8f275d35af3d8e42f3aa1525871b43400921a5d2dca4897c2047214",
@@ -175,7 +175,11 @@
"method": "Stable-ID lifecycle and recurrent-kernel author lanes, integrated root/local references, hard cases and offline support/static/scaffold/package checks; native Swift/Xcode/MLX compiler/tests not_run."
},
"applicability": {
- "scope": "Selected Swift ownership/cancellation/settling and actual GatedDelta Metal/ops contracts. Common categorization is not CPU inference or Apple qualification."
+ "scope": "Selected Swift ownership/cancellation/settling and actual GatedDelta Metal/ops contracts. Common categorization is not CPU inference or Apple qualification.",
+ "prerequisites": "Access to the Swift lifecycle and selected GatedDelta contracts, with resolved MLX/native identities for a comparison; native checks require a prepared matching Swift/Xcode/Apple stack and separate authorization.",
+ "input_contract": "Exact source revision, resolved Swift/MLX dependencies, Xcode/SDK/OS/device, API family and native artifact identity; the selected question supplies lifecycle ownership/completion observations or recurrent producer shape/dtype/layout and actual consumer/fallback.",
+ "produced_artifact": "An identity and actual-versus-declared report for generation ownership or recurrent kernel invariants, with safe support result, counter-signal and smallest separately authorized native check.",
+ "failure_modes": "Conflating lexical isolation with exclusive access across await or completed task join, treating useKernel as observed dispatch, calling an ops fallback CPU execution, or assuming a manifest range is a resolved dependency can invalidate the conclusion."
},
"limitations": [
"BLOCKED_REQUIRED_BACKEND: native Swift/MLX compiler, host-control tests and required Apple Metal cancellation/backgrounding/kernel/cache/model/fit/quality/service/performance evidence not_run.",
@@ -189,14 +193,24 @@
"references/repo-provenance.md",
"references/source-identity.json"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Resolve mutable-context/cache ownership and task settlement separately from lexical isolation and queued GPU work before cache reuse or producer shutdown.",
+ "Follow recurrent shape/dtype/layout to the actual Metal or ops consumer; predicate eligibility and useKernel alone do not prove dispatch.",
+ "Record the actual dependency resolution and use the documented Xcode macOS test route; the source states swift test does not work for this package."
+ ]
},
"applicability": {
- "scope": "Selected Swift ownership/cancellation/settling and actual GatedDelta Metal/ops contracts. Common categorization is not CPU inference or Apple qualification."
+ "scope": "Selected Swift ownership/cancellation/settling and actual GatedDelta Metal/ops contracts. Common categorization is not CPU inference or Apple qualification.",
+ "prerequisites": "Access to the Swift lifecycle and selected GatedDelta contracts, with resolved MLX/native identities for a comparison; native checks require a prepared matching Swift/Xcode/Apple stack and separate authorization.",
+ "input_contract": "Exact source revision, resolved Swift/MLX dependencies, Xcode/SDK/OS/device, API family and native artifact identity; the selected question supplies lifecycle ownership/completion observations or recurrent producer shape/dtype/layout and actual consumer/fallback.",
+ "produced_artifact": "An identity and actual-versus-declared report for generation ownership or recurrent kernel invariants, with safe support result, counter-signal and smallest separately authorized native check.",
+ "failure_modes": "Conflating lexical isolation with exclusive access across await or completed task join, treating useKernel as observed dispatch, calling an ops fallback CPU execution, or assuming a manifest range is a resolved dependency can invalidate the conclusion."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Resolve mutable-context/cache ownership and task settlement separately from lexical isolation and queued GPU work before cache reuse or producer shutdown.",
+ "Follow recurrent shape/dtype/layout to the actual Metal or ops consumer; predicate eligibility and useKernel alone do not prove dispatch.",
+ "Record the actual dependency resolution and use the documented Xcode macOS test route; the source states swift test does not work for this package."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/npu-skill-factory.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/npu-skill-factory.json
index cafbdc364..50f87f78e 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/npu-skill-factory.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/npu-skill-factory.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/workflows/npu-skill-factory/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/workflows/npu-skill-factory/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/skill-index.json",
"src/inferrsi/data/skill_library/workflows/npu-skill-factory/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/optimization-search-policy.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/optimization-search-policy.json
index 798493ab5..77fd098dd 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/optimization-search-policy.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/optimization-search-policy.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/methods/optimization-search-policy/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/methods/optimization-search-policy/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/methods/optimization-search-policy/SKILL.md",
"src/inferrsi/data/skill_library/methods/optimization-search-policy/cases/sarathi-chunk-budget.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/os-process-placement.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/os-process-placement.json
index 59eecca4c..4dcc882a9 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/os-process-placement.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/os-process-placement.json
@@ -24,8 +24,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/hardware/host/os-process-placement/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/hardware/host/os-process-placement/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/hardware/host/os-process-placement/SKILL.md",
"src/inferrsi/data/skill_library/hardware/host/os-process-placement/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/performance-ceiling-analysis.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/performance-ceiling-analysis.json
index 5fde95df1..4f0d70fce 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/performance-ceiling-analysis.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/performance-ceiling-analysis.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/SKILL.md",
"src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/plan-activation-residency.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/plan-activation-residency.json
index 14d1ac0fb..7545f1ad3 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/plan-activation-residency.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/plan-activation-residency.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/methods/plan-activation-residency/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/methods/plan-activation-residency/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/methods/plan-activation-residency/SKILL.md",
"src/inferrsi/data/skill_library/methods/plan-activation-residency/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/project-maya.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/project-maya.json
index 3327dbace..c5b6b74f1 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/project-maya.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/project-maya.json
@@ -25,8 +25,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/repositories/project-maya/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/repositories/project-maya/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/repositories/project-maya/LICENSE",
"src/inferrsi/data/skill_library/repositories/project-maya/SKILL.md",
@@ -63,7 +63,7 @@
"source_files": {
"src/inferrsi/data/skill_library/repositories/project-maya/LICENSE": "574cffb7c57769a9271043ac0c2b07c5e7e7621cfb81081b901d8981e791d121",
"src/inferrsi/data/skill_library/repositories/project-maya/SKILL.md": "c619a3472be02147a847a24ca97c56afd8b56b6ffd22ce66fcae453155e34bab",
- "src/inferrsi/data/skill_library/repositories/project-maya/metadata.json": "c9f7879ca512c07e9bdf6d1d1a52de055a0fa210b8838c205e27a34436324834",
+ "src/inferrsi/data/skill_library/repositories/project-maya/metadata.json": "c48af69c430f23a8b39a2dbc09495bef54b48147c64068b39cfb3551f4e94c63",
"src/inferrsi/data/skill_library/repositories/project-maya/references/license-boundaries.md": "090839df90d6e96f55794c2ad821d7154dc6a6aa4211cdbb08a1e5f6f2e52a92",
"src/inferrsi/data/skill_library/repositories/project-maya/references/repo-provenance.md": "7bb6b6c14d1a9b77e906804505c723569f952e5d47559e636d2e6621994216bf",
"src/inferrsi/data/skill_library/repositories/project-maya/references/repo-routing-metadata.json": "0acbb111e92316e74d537200cd033f3d7d0f69972bf0953c8cea9dfb2ecb4a34",
@@ -225,7 +225,12 @@
"method": "Stable-ID isolated author lanes, integrated runtime graph, assertion/helper/static and selected direct-compiler CPU verification; required accelerator/model blocks retained."
},
"applicability": {
- "scope": "Model-specific dispatch/residency, API/native state and qualification investigation. Common catalog placement is not CPU inference or cross-device certification."
+ "scope": "Model-specific dispatch/residency, API/native state and qualification investigation. Common catalog placement is not CPU inference or cross-device certification.",
+ "prerequisites": "Access to the selected model-family runtime/state/qualification contract; native changes need frozen correctness/comparison requirements and the matching authorized CUDA/HIP/model resources.",
+ "input_contract": "Full source revision, actual executable/model family, model and asset identities, representation, compiler/runtime/device, rendered workload, sampling policy, single/parallel route, exact tokens/images and relevant completion observations.",
+ "produced_artifact": "A route/identity/evidence-class report with invariant, proposed check, actual result, counter-signal, missing resource and next authorized action; missing native requirements remain INCONCLUSIVE.",
+ "failure_modes": "Transferring coverage through a common facade or quantization name, confusing API history with native KV/private state, double-counting unified RAM/VRAM, or repairing a slot header to appear compatible can invalidate diagnosis.",
+ "refusal_point": "When required CUDA/HIP/model evidence is unavailable, retain BLOCKED_REQUIRED_BACKEND and an unverified/INCONCLUSIVE conclusion; exit-zero skips or CPU simulations cannot clear it."
},
"limitations": [
"BLOCKED_REQUIRED_BACKEND: no actual CUDA/HIP/model/device, quantized payload, vision, parallel/state/restore or cancellation-cleanup evidence.",
@@ -238,14 +243,25 @@
"references/repo-provenance.md",
"references/source-identity.json"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Select the family-specific route and preserve the explicitly shallow Qwen coverage rather than transferring deeper GLM/DeepSeek/MiniMax evidence.",
+ "Separate shared weights, private conversation state, API history and native KV; count unified physical memory once.",
+ "Preserve exact token/image identities and completion boundaries before reuse, then test one falsifiable hypothesis without modifying slot headers to bypass compatibility."
+ ]
},
"applicability": {
- "scope": "Model-specific dispatch/residency, API/native state and qualification investigation. Common catalog placement is not CPU inference or cross-device certification."
+ "scope": "Model-specific dispatch/residency, API/native state and qualification investigation. Common catalog placement is not CPU inference or cross-device certification.",
+ "prerequisites": "Access to the selected model-family runtime/state/qualification contract; native changes need frozen correctness/comparison requirements and the matching authorized CUDA/HIP/model resources.",
+ "input_contract": "Full source revision, actual executable/model family, model and asset identities, representation, compiler/runtime/device, rendered workload, sampling policy, single/parallel route, exact tokens/images and relevant completion observations.",
+ "produced_artifact": "A route/identity/evidence-class report with invariant, proposed check, actual result, counter-signal, missing resource and next authorized action; missing native requirements remain INCONCLUSIVE.",
+ "failure_modes": "Transferring coverage through a common facade or quantization name, confusing API history with native KV/private state, double-counting unified RAM/VRAM, or repairing a slot header to appear compatible can invalidate diagnosis.",
+ "refusal_point": "When required CUDA/HIP/model evidence is unavailable, retain BLOCKED_REQUIRED_BACKEND and an unverified/INCONCLUSIVE conclusion; exit-zero skips or CPU simulations cannot clear it."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Select the family-specific route and preserve the explicitly shallow Qwen coverage rather than transferring deeper GLM/DeepSeek/MiniMax evidence.",
+ "Separate shared weights, private conversation state, API history and native KV; count unified physical memory once.",
+ "Preserve exact token/image identities and completion boundaries before reuse, then test one falsifiable hypothesis without modifying slot headers to bypass compatibility."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/repository-evolution-analysis.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/repository-evolution-analysis.json
index 0113cbb14..95226ff34 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/repository-evolution-analysis.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/repository-evolution-analysis.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/methods/repository-evolution-analysis/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/methods/repository-evolution-analysis/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/methods/repository-evolution-analysis/SKILL.md",
"src/inferrsi/data/skill_library/methods/repository-evolution-analysis/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/rsi-evidence.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/rsi-evidence.json
index e1df63741..84641977b 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/rsi-evidence.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/rsi-evidence.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/workflows/rsi-evidence/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/workflows/rsi-evidence/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/skill-index.json",
"src/inferrsi/data/skill_library/workflows/rsi-evidence/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/skill-evolution-playbook.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/skill-evolution-playbook.json
index f19d61f2f..26570c559 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/skill-evolution-playbook.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/skill-evolution-playbook.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/methods/skill-evolution-playbook/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/methods/skill-evolution-playbook/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/methods/skill-evolution-playbook/SKILL.md",
"src/inferrsi/data/skill_library/methods/skill-evolution-playbook/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/splash.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/splash.json
index 3062af59c..875ef8241 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/splash.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/splash.json
@@ -25,8 +25,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/repositories/splash/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/repositories/splash/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/repositories/splash/LICENSE",
"src/inferrsi/data/skill_library/repositories/splash/SKILL.md",
@@ -61,7 +61,7 @@
"src/inferrsi/data/skill_library/repositories/splash/LICENSE": "cfc7749b96f63bd31c3c42b5c471bf756814053e847c10f3eb003417bc523d30",
"src/inferrsi/data/skill_library/repositories/splash/SKILL.md": "619b87c282dcec4befa319085764fb53fa506a05c6a7112500bcca662f059ece",
"src/inferrsi/data/skill_library/repositories/splash/THIRD_PARTY_NOTICES": "2fedf1277bf732f8435c3e3be2798f52f25f7e2bd4155865cb366da55b88e50b",
- "src/inferrsi/data/skill_library/repositories/splash/metadata.json": "d4c679f474c31549156aad417e23c2802b9f4bc904472e5cb16916f2996b50ba",
+ "src/inferrsi/data/skill_library/repositories/splash/metadata.json": "9d0613387fdaee61de7e79146e2a896f19ac7f5c1d355f74afe767074f64dd2a",
"src/inferrsi/data/skill_library/repositories/splash/references/repo-provenance.md": "20e3462f1b6a8752eb0fa6f36e76867f142a49c1953f498f895e29c4fd4f0204",
"src/inferrsi/data/skill_library/repositories/splash/references/repo-routing-metadata.json": "287ae591f42949b8d33312659ec8303c47d5c73093c64c125edb1cf6ea8de8b9",
"src/inferrsi/data/skill_library/repositories/splash/references/source-identity.json": "2f778882ec440538a46f48332650412b446722246c8dc5690083f19da7dce61d",
@@ -189,7 +189,12 @@
"method": "User-approved native Metal/C++ and isolated source-Python adaptation; stable-ID lanes, integrated graph, assertion cases and create/verify stages. Required-backend block disables automatic admission."
},
"applicability": {
- "scope": "Pinned Splash representation, planning, state/serving and verification investigation; common discovery does not qualify a different Mac/model/backend."
+ "scope": "Pinned Splash representation, planning, state/serving and verification investigation; common discovery does not qualify a different Mac/model/backend.",
+ "prerequisites": "Access to the pinned representation/planning, serving-cache or qualification route; native checks need matching Apple/backend/model resources, an authorized workload and frozen semantic/measurement controls.",
+ "input_contract": "Source/build, macOS/device, target and compatible draft artifacts/formats, tokenizer/template/vision selection, context/cache/memory controls, KV/recurrent precision, ANE and verifier-row policy, concurrency/priority/sampling, quality controls and cache state.",
+ "produced_artifact": "An execution-identity report with observation, source-backed invariant, alternate explanation, smallest discriminating check, actual result and unresolved Metal/ANE/model/disk-state gap.",
+ "failure_modes": "Treating an API alias as new weights, equal bit width as loader/kernel compatibility, or a fixed seed as fixed cache/batch/adaptive-verification scheduling can invalidate the comparison; CPU cache/codec checks do not prove GPU state.",
+ "refusal_point": "Stop at unsupported formats, unavailable backend/model resources, unsafe service exposure or the authorized budget."
},
"limitations": [
"Required Metal/ANE/model/actual device and disk-state execution remains BLOCKED_REQUIRED_BACKEND; Python policy/codec checks do not clear it.",
@@ -201,14 +206,25 @@
"references/repo-provenance.md",
"references/source-identity.json"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Resolve representation, memory/cache owner, validity key and reclaim/rollback boundary before tuning a numeric control.",
+ "Freeze semantic and workload controls beyond the seed, and retain actual-path fallbacks, admission waits, evictions, cancellations and errors.",
+ "Inspect options without starting the normal serve path, which can download target/draft weights; installation and serving require separate task authority."
+ ]
},
"applicability": {
- "scope": "Pinned Splash representation, planning, state/serving and verification investigation; common discovery does not qualify a different Mac/model/backend."
+ "scope": "Pinned Splash representation, planning, state/serving and verification investigation; common discovery does not qualify a different Mac/model/backend.",
+ "prerequisites": "Access to the pinned representation/planning, serving-cache or qualification route; native checks need matching Apple/backend/model resources, an authorized workload and frozen semantic/measurement controls.",
+ "input_contract": "Source/build, macOS/device, target and compatible draft artifacts/formats, tokenizer/template/vision selection, context/cache/memory controls, KV/recurrent precision, ANE and verifier-row policy, concurrency/priority/sampling, quality controls and cache state.",
+ "produced_artifact": "An execution-identity report with observation, source-backed invariant, alternate explanation, smallest discriminating check, actual result and unresolved Metal/ANE/model/disk-state gap.",
+ "failure_modes": "Treating an API alias as new weights, equal bit width as loader/kernel compatibility, or a fixed seed as fixed cache/batch/adaptive-verification scheduling can invalidate the comparison; CPU cache/codec checks do not prove GPU state.",
+ "refusal_point": "Stop at unsupported formats, unavailable backend/model resources, unsafe service exposure or the authorized budget."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Resolve representation, memory/cache owner, validity key and reclaim/rollback boundary before tuning a numeric control.",
+ "Freeze semantic and workload controls beyond the seed, and retain actual-path fallbacks, admission waits, evictions, cancellations and errors.",
+ "Inspect options without starting the normal serve path, which can download target/draft weights; installation and serving require separate task authority."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/storage-io-measurement.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/storage-io-measurement.json
index ac42b98c1..a79653f8a 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/storage-io-measurement.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/storage-io-measurement.json
@@ -24,8 +24,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/hardware/host/storage-io-measurement/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/hardware/host/storage-io-measurement/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/hardware/host/storage-io-measurement/SKILL.md",
"src/inferrsi/data/skill_library/hardware/host/storage-io-measurement/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-input-pipeline.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-input-pipeline.json
index 1c56b3e7d..9c0200a4b 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-input-pipeline.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-input-pipeline.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/methods/training-input-pipeline/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/methods/training-input-pipeline/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/methods/training-input-pipeline/SKILL.md",
"src/inferrsi/data/skill_library/methods/training-input-pipeline/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-resource-budget.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-resource-budget.json
index 9efe6daab..8d50cc80c 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-resource-budget.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-resource-budget.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/methods/training-resource-budget/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/methods/training-resource-budget/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/methods/training-resource-budget/SKILL.md",
"src/inferrsi/data/skill_library/methods/training-resource-budget/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/video-diffusion-optimization.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/video-diffusion-optimization.json
index dcfebda27..2a6e36d8c 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/video-diffusion-optimization.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/video-diffusion-optimization.json
@@ -26,8 +26,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/methods/video-diffusion-optimization/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/methods/video-diffusion-optimization/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/methods/video-diffusion-optimization/SKILL.md",
"src/inferrsi/data/skill_library/methods/video-diffusion-optimization/assets/campaign.template.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/vllm-metal.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/vllm-metal.json
index 37c7ebf9c..6d36773d6 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/vllm-metal.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/vllm-metal.json
@@ -26,8 +26,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/repositories/vllm-metal/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/repositories/vllm-metal/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/repositories/vllm-metal/LICENSE",
"src/inferrsi/data/skill_library/repositories/vllm-metal/SKILL.md",
@@ -65,7 +65,7 @@
"source_files": {
"src/inferrsi/data/skill_library/repositories/vllm-metal/LICENSE": "c71d239df91726fc519c6eb72d318ec65820627232b2f796219e87dcf35d0ab4",
"src/inferrsi/data/skill_library/repositories/vllm-metal/SKILL.md": "5f83b1b989e8dc239a6f34a6b3bd63872533ecc1404483a69dabfb1865648439",
- "src/inferrsi/data/skill_library/repositories/vllm-metal/metadata.json": "63d51d1ace6b0ba503c28751a8ffbf9224038395f4146a94c775c628d8075470",
+ "src/inferrsi/data/skill_library/repositories/vllm-metal/metadata.json": "55652ec6de744d92044951e25ea89aa18c92b14846dd4b597444d773bb8c31e3",
"src/inferrsi/data/skill_library/repositories/vllm-metal/references/repo-provenance.md": "970bda4d1081704aa82c94d6544a98bbf479430c4fc2c82a1a82f206043aa72a",
"src/inferrsi/data/skill_library/repositories/vllm-metal/references/repo-routing-metadata.json": "230a2b383bcf2855af4de50374c41f11b0b0841268ae45e424c1c97f679b477a",
"src/inferrsi/data/skill_library/repositories/vllm-metal/references/scope-and-notices.md": "aaef98eaea4cff56f39852e1d131163ebc01334e79034b1890f67a2660727d3e",
@@ -185,7 +185,11 @@
"method": "Stable-ID manual author sequencing plus bounded cache author lane; integrated graph followed by assertion/helper/static and selected pure Python source tests; Apple native blocks retained."
},
"applicability": {
- "scope": "Selected plugin preparation, KV/offload ownership and speculative/deferred/dispatch evidence investigation. Common catalog scope does not certify CPU inference or cross-backend qualification."
+ "scope": "Selected plugin preparation, KV/offload ownership and speculative/deferred/dispatch evidence investigation. Common catalog scope does not certify CPU inference or cross-backend qualification.",
+ "prerequisites": "Access to the plugin-specific platform/ABI, cache-lifecycle and qualification contracts; actual engine changes require separate authority, resource ownership, frozen gates and cleanup, with a supported artifact pairing.",
+ "input_contract": "Source/package/runtime/native artifact identities, Python/OS/chip/device, backend/executor, checkpoint/tokenizer/draft identities, format/dtype, effective feature flags and workload.",
+ "produced_artifact": "An identity/route/invariant report with actual safe-check result, counter-signal and next authorized native check; partial CPU evidence stays separate from unverified transfers, sleep/wake, model and HTTP outcomes.",
+ "failure_modes": "Using lazy import success as native registration or ABI evidence, treating a feature switch as dispatch, or conflating cache capacity with physical residency and kernel time with HTTP time can overstate plugin behavior."
},
"limitations": [
"BLOCKED_REQUIRED_BACKEND: actual Apple MLX/Metal and MPS kernels/model loading, cache transfers/sleep/wake and real worker dispatch are not verified.",
@@ -198,14 +202,24 @@
"references/repo-provenance.md",
"references/source-identity.json"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Resolve platform registration, backend/runner and native ABI pairing before capacity tuning; a package visibility check establishes only its reported software facts.",
+ "Keep MLX and MPS, full-attention and recurrent state, logical cache capacity and physical residency as separate contracts.",
+ "Preserve failed paths and distinguish parser/API checks from real transfers, sleep/wake, model output and HTTP performance evidence."
+ ]
},
"applicability": {
- "scope": "Selected plugin preparation, KV/offload ownership and speculative/deferred/dispatch evidence investigation. Common catalog scope does not certify CPU inference or cross-backend qualification."
+ "scope": "Selected plugin preparation, KV/offload ownership and speculative/deferred/dispatch evidence investigation. Common catalog scope does not certify CPU inference or cross-backend qualification.",
+ "prerequisites": "Access to the plugin-specific platform/ABI, cache-lifecycle and qualification contracts; actual engine changes require separate authority, resource ownership, frozen gates and cleanup, with a supported artifact pairing.",
+ "input_contract": "Source/package/runtime/native artifact identities, Python/OS/chip/device, backend/executor, checkpoint/tokenizer/draft identities, format/dtype, effective feature flags and workload.",
+ "produced_artifact": "An identity/route/invariant report with actual safe-check result, counter-signal and next authorized native check; partial CPU evidence stays separate from unverified transfers, sleep/wake, model and HTTP outcomes.",
+ "failure_modes": "Using lazy import success as native registration or ABI evidence, treating a feature switch as dispatch, or conflating cache capacity with physical residency and kernel time with HTTP time can overstate plugin behavior."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Resolve platform registration, backend/runner and native ABI pairing before capacity tuning; a package visibility check establishes only its reported software facts.",
+ "Keep MLX and MPS, full-attention and recurrent state, logical cache capacity and physical residency as separate contracts.",
+ "Preserve failed paths and distinguish parser/API checks from real transfers, sleep/wake, model output and HTTP performance evidence."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/weight-loading-optimization.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/weight-loading-optimization.json
index f894ccf93..1df5e1bfa 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/weight-loading-optimization.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/weight-loading-optimization.json
@@ -29,8 +29,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/methods/weight-loading-optimization/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/methods/weight-loading-optimization/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/methods/weight-loading-optimization/SKILL.md",
"src/inferrsi/data/skill_library/methods/weight-loading-optimization/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-afd-p2p-exchanges.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-afd-p2p-exchanges.json
index 6f18573b8..1f0d714c8 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-afd-p2p-exchanges.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-afd-p2p-exchanges.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/SKILL.md",
"src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/metadata.json",
@@ -42,7 +42,7 @@
"checkout_ref": "checkout:audit-afd-p2p-exchanges",
"source_files": {
"src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/SKILL.md": "a46110891eca860915e314d6921a22856d631453b22b17837fe30262d6780535",
- "src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/metadata.json": "4e6100dfb9fc7bc00b19915f4fd38ad509704f2a609c658adf4d632d4e05b6c1",
+ "src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/metadata.json": "ae8c7582265dd6902cc4172293fc709244cc915df9390672fde25687b9e79b4b",
"src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/references/afd-plugin-LICENSE.md": "0e8566ea362276878e1f38f97da49964dbea301c15d4aa9d727f97a4c39fd943",
"src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/references/source-notes.md": "5073d87450cb60d1b245308dff8807e73749c07de976d50f2fa26694fad81a72",
"src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
@@ -92,7 +92,12 @@
}
],
"applicability": {
- "scope": "Source-based CUDA P2pNccl rank/payload audit; selected eager dense-FFN path with compute_gate_on_attention=false. Repository identity afd_plugin; runtime/backend compatibility unverified."
+ "scope": "Source-based CUDA P2pNccl rank/payload audit; selected eager dense-FFN path with compute_gate_on_attention=false. Repository identity afd_plugin; runtime/backend compatibility unverified.",
+ "prerequisites": "Access to the pinned source ledger and caller-level evidence for the selected eager dense FFN path with Attention-side gating disabled; any proposed cross-stream reuse needs an established communicator completion contract.",
+ "input_contract": "Plugin and dependency revisions, connector and role, A/F worker counts, DP/PCP/TP coordinates, worker/device ranks, subgroup roster, stage, per-peer token counts, payload shape/dtype/device, transfer context and returned slices.",
+ "produced_artifact": "A bounded rank-namespace and per-peer payload ownership ledger, the earliest evidenced mismatch or missing edge, and a proposed verification plan.",
+ "failure_modes": "Inferring data peers from control senders, accepting equal total row counts despite a peer permutation, or treating an even-split fallback or Python return as proof of ownership/completion can produce a wrong exchange diagnosis.",
+ "refusal_point": "Stop on A < F, invalid role ranks, duplicate/missing assignments or an unresolved rank namespace; return the conflicting callers and ledger entries without inventing ranks or attributing the fault to networking."
},
"limitations": [
"Source-derived candidate; native exchange, model quality, backend compatibility, performance and agent efficacy not_run.",
@@ -107,14 +112,25 @@
"../../../methods/inference-benchmark-design/SKILL.md",
"../../../methods/dense-performance-evidence/SKILL.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Record role, world and subgroup-local ranks separately; reconstruct the actual subgroup roster before interpreting send/receive peers.",
+ "Track distinct row identities and stage-specific peer lengths through concatenation and splitting; use same-total swapped lengths as a countercheck.",
+ "Establish the last communicator/stream completion obligation before reusing or releasing a receive buffer."
+ ]
},
"applicability": {
- "scope": "Source-based CUDA P2pNccl rank/payload audit; selected eager dense-FFN path with compute_gate_on_attention=false. Repository identity afd_plugin; runtime/backend compatibility unverified."
+ "scope": "Source-based CUDA P2pNccl rank/payload audit; selected eager dense-FFN path with compute_gate_on_attention=false. Repository identity afd_plugin; runtime/backend compatibility unverified.",
+ "prerequisites": "Access to the pinned source ledger and caller-level evidence for the selected eager dense FFN path with Attention-side gating disabled; any proposed cross-stream reuse needs an established communicator completion contract.",
+ "input_contract": "Plugin and dependency revisions, connector and role, A/F worker counts, DP/PCP/TP coordinates, worker/device ranks, subgroup roster, stage, per-peer token counts, payload shape/dtype/device, transfer context and returned slices.",
+ "produced_artifact": "A bounded rank-namespace and per-peer payload ownership ledger, the earliest evidenced mismatch or missing edge, and a proposed verification plan.",
+ "failure_modes": "Inferring data peers from control senders, accepting equal total row counts despite a peer permutation, or treating an even-split fallback or Python return as proof of ownership/completion can produce a wrong exchange diagnosis.",
+ "refusal_point": "Stop on A < F, invalid role ranks, duplicate/missing assignments or an unresolved rank namespace; return the conflicting callers and ledger entries without inventing ranks or attributing the fault to networking."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Record role, world and subgroup-local ranks separately; reconstruct the actual subgroup roster before interpreting send/receive peers.",
+ "Track distinct row identities and stage-specific peer lengths through concatenation and splitting; use same-total swapped lengths as a countercheck.",
+ "Establish the last communicator/stream completion obligation before reusing or releasing a receive buffer."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-flare-session-lifecycle.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-flare-session-lifecycle.json
index f86ac899a..d5b231f17 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-flare-session-lifecycle.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-flare-session-lifecycle.json
@@ -29,8 +29,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/LICENSE",
"src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/SKILL.md",
@@ -52,7 +52,7 @@
"source_files": {
"src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/LICENSE": "cfc7749b96f63bd31c3c42b5c471bf756814053e847c10f3eb003417bc523d30",
"src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/SKILL.md": "02a5f7c18e44201215ecec17c6ac96b27be9148bcc189df287f39d1b372b7721",
- "src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/metadata.json": "58716737446f46bd1de77aa323c5652f97277d44dd952c82b42399b2cb769b87",
+ "src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/metadata.json": "9c667c7735ebe4fb07247e098218f6b03a1c1fafa776f1372f8d948ea8ea756b",
"src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/references/diagnostics.md": "0908a0eae53a2e10775a8b6ee3e3944b2fa410174402b5a1a8c98eb674dd4b4d",
"src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/references/lifecycle.md": "107ebe3c22e5dba8007c3b8ea61149f10bc55c51b557a4230f3e5de859bb0f7c",
"src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/references/source-ledger.json": "61e51a97ffd7c1fa7866bdd440b024d0c5e0065635ac8d0ba53d2c69c7f9e470",
@@ -126,7 +126,12 @@
"method": "AREX scope/ground/construct/verify and repository evidence/usability methods adapted to InferRSI native owners and packaging; no DisCo runtime or managed import."
},
"applicability": {
- "scope": "Pinned Flare complete-session restore and admission audit for its NVIDIA SM120/SM121 research prototype. CUDA catalog scope is content classification, not device or backend qualification."
+ "scope": "Pinned Flare complete-session restore and admission audit for its NVIDIA SM120/SM121 research prototype. CUDA catalog scope is content classification, not device or backend qualification.",
+ "prerequisites": "Access to package/source identity and bounded restore traces, with preparation compatibility established before cache-state diagnosis. Running, freeing resources or changing configuration requires separate authority.",
+ "input_contract": "Execution ABI and prepared-contract version, ordinary/speculative configuration, model/layout identity, request IDs and consumed-token endpoints, GPU/host prefixes, continuation row, ticket/waiter identities, transfer events, fences and resource counts.",
+ "produced_artifact": "A checkpoint/ticket ownership table, five-outcome admission ledger and diagnosis identifying the earliest unsupported transition and smallest missing observation.",
+ "failure_modes": "A KV-only lookup hit can omit continuation state; admission can be mistaken for actual reuse; client cancellation can be mistaken for transfer completion; mixing absolute timestamps from client and native clocks can corrupt the diagnosis.",
+ "refusal_point": "Stop with an unresolved edge when package identity, layout, transfer completion or clock provenance is unknown."
},
"limitations": [
"Source-only candidate; native Flare/model/GPU/service and skill efficacy were not evaluated.",
@@ -139,14 +144,25 @@
"references/source-notes.md",
"references/source-ledger.json"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Pair the immutable KV parent chain with the complete continuation row at the same consumed-token endpoint before calling a checkpoint restorable.",
+ "Distinguish lookup outcomes, destination reservation, ticket creation, transfer completion, publication and actual reused-token evidence.",
+ "Follow scheduler-owned tickets through waiter cancellation and check write/use fences before allowing storage reuse."
+ ]
},
"applicability": {
- "scope": "Pinned Flare complete-session restore and admission audit for its NVIDIA SM120/SM121 research prototype. CUDA catalog scope is content classification, not device or backend qualification."
+ "scope": "Pinned Flare complete-session restore and admission audit for its NVIDIA SM120/SM121 research prototype. CUDA catalog scope is content classification, not device or backend qualification.",
+ "prerequisites": "Access to package/source identity and bounded restore traces, with preparation compatibility established before cache-state diagnosis. Running, freeing resources or changing configuration requires separate authority.",
+ "input_contract": "Execution ABI and prepared-contract version, ordinary/speculative configuration, model/layout identity, request IDs and consumed-token endpoints, GPU/host prefixes, continuation row, ticket/waiter identities, transfer events, fences and resource counts.",
+ "produced_artifact": "A checkpoint/ticket ownership table, five-outcome admission ledger and diagnosis identifying the earliest unsupported transition and smallest missing observation.",
+ "failure_modes": "A KV-only lookup hit can omit continuation state; admission can be mistaken for actual reuse; client cancellation can be mistaken for transfer completion; mixing absolute timestamps from client and native clocks can corrupt the diagnosis.",
+ "refusal_point": "Stop with an unresolved edge when package identity, layout, transfer completion or clock provenance is unknown."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Pair the immutable KV parent chain with the complete continuation row at the same consumed-token endpoint before calling a checkpoint restorable.",
+ "Distinguish lookup outcomes, destination reservation, ticket creation, transfer completion, publication and actual reused-token evidence.",
+ "Follow scheduler-owned tickets through waiter cancellation and check write/use fences before allowing storage reuse."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/diagnose-attention-bottlenecks.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/diagnose-attention-bottlenecks.json
index 13bd48286..35f3cbb6e 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/diagnose-attention-bottlenecks.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/diagnose-attention-bottlenecks.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/hardware/cuda/diagnose-attention-bottlenecks/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/hardware/cuda/diagnose-attention-bottlenecks/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/hardware/cuda/diagnose-attention-bottlenecks/SKILL.md",
"src/inferrsi/data/skill_library/hardware/cuda/diagnose-attention-bottlenecks/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/host-overhead-profiling.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/host-overhead-profiling.json
index aa21e20fb..201d673c6 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/host-overhead-profiling.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/host-overhead-profiling.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/hardware/cuda/host-overhead-profiling/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/hardware/cuda/host-overhead-profiling/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/hardware/cuda/host-overhead-profiling/SKILL.md",
"src/inferrsi/data/skill_library/hardware/cuda/host-overhead-profiling/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/reason-hopper-kernels.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/reason-hopper-kernels.json
index c2aa42eec..d435de2f7 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/reason-hopper-kernels.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/reason-hopper-kernels.json
@@ -36,8 +36,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/c5e13944c06a8a03ab7015a8933981a983e42186/src/inferrsi/data/skill_library/hardware/cuda/reason-hopper-kernels/SKILL.md",
- "commit": "c5e13944c06a8a03ab7015a8933981a983e42186",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb/src/inferrsi/data/skill_library/hardware/cuda/reason-hopper-kernels/SKILL.md",
+ "commit": "e30ccae2aedfb190036ce1eaaef6f7fbb1cd46bb",
"paths": [
"src/inferrsi/data/skill_library/hardware/cuda/reason-hopper-kernels/SKILL.md",
"src/inferrsi/data/skill_library/hardware/cuda/reason-hopper-kernels/metadata.json",
From ca893f2eae2bf6d5a14e91f60953bcf2825c998b Mon Sep 17 00:00:00 2001
From: Dong1017 <55723162+Dong1017@users.noreply.github.com>
Date: Sat, 31 Oct 2026 12:55:03 +0800
Subject: [PATCH 05/11] docs: regenerate skill projections from published
source history
Keep the canonical source snapshot reachable as a preceding commit.
Agent-assisted-by: Codex / gpt-6-astra [effort=xhigh] (author)
Signed-off-by: Dong1017 <55723162+Dong1017@users.noreply.github.com>
---
docs/skill-coverage.md | 62 +++++++++----------
.../repo_packs/inferrsi/knowledge/README.md | 4 +-
.../ascend-aclgraph-capture-replay.json | 4 +-
.../ascend/skills/ascend-device-health.json | 4 +-
.../skills/ascend-distributed-parallel.json | 4 +-
.../ascend/skills/ascend-graph-mode.json | 4 +-
.../skills/ascend-model-adaptation.json | 4 +-
.../skills/ascend-operator-profiling.json | 4 +-
.../ascend/skills/ascend-operator-tuning.json | 4 +-
.../skills/ascend-optimization-casebook.json | 4 +-
.../skills/ascend-performance-analysis.json | 4 +-
.../ascend-video-diffusion-optimization.json | 4 +-
.../ascend/skills/ascendc-tiling-design.json | 4 +-
.../skills/cann-attribution-overlap.json | 4 +-
.../torchnpu-device-memory-lifecycle.json | 4 +-
...torchnpu-pinned-host-memory-lifecycle.json | 4 +-
.../skills/assess-serving-preemption.json | 4 +-
.../assess-split-backward-schedule.json | 4 +-
.../skills/audit-fixed-workload-claim.json | 4 +-
.../audit-h3-metal-fasth3-semantics.json | 32 +++++++---
.../skills/audit-h3-runtime-residency.json | 32 +++++++---
.../skills/audit-llama-runtime-state.json | 32 +++++++---
.../audit-recurrent-kv-finalization.json | 32 +++++++---
.../common/skills/audit-sd-cpp-residency.json | 32 +++++++---
.../audit-sglang-diffusion-lifetimes.json | 32 +++++++---
.../skills/audit-vllm-kv-lifetimes.json | 35 ++++++++---
.../bound-windowed-attention-replay.json | 30 ++++++---
.../budget-speculative-verification.json | 30 ++++++---
.../skills/calibrate-serving-simulation.json | 4 +-
.../skills/cgroup-resource-controls.json | 4 +-
.../checkpoint-recovery-consistency.json | 4 +-
.../skills/choose-parallel-training-plan.json | 4 +-
.../skills/cite-check-downstream-pr.json | 4 +-
.../common/skills/claim-contract.json | 4 +-
.../knowledge/common/skills/comfy-ui.json | 32 +++++++---
.../common/skills/competitor-scan.json | 4 +-
.../skills/dcp-async-save-performance.json | 35 ++++++++---
.../skills/dense-performance-evidence.json | 4 +-
.../diagnose-vllm-admission-pressure.json | 35 ++++++++---
.../skills/diffusers-invocation-state.json | 32 +++++++---
.../common/skills/draft-acceptance-issue.json | 4 +-
.../knowledge/common/skills/dwarfstar.json | 32 +++++++---
.../evaluate-elastic-serving-placement.json | 4 +-
.../skills/generation-cache-lifecycle.json | 32 +++++++---
.../common/skills/host-memory-management.json | 4 +-
.../skills/host-overhead-optimization.json | 4 +-
.../knowledge/common/skills/husky.json | 32 +++++++---
.../skills/inference-benchmark-design.json | 4 +-
.../inference-optimization-casebook.json | 4 +-
.../common/skills/kernel-writing-guide.json | 4 +-
.../common/skills/kv-tier-management.json | 33 +++++++---
.../lazy-evaluation-stream-lifetimes.json | 32 +++++++---
.../knowledge/common/skills/lily.json | 32 +++++++---
.../knowledge/common/skills/lithos-metal.json | 30 ++++++---
.../knowledge/common/skills/local-dream.json | 30 ++++++---
.../knowledge/common/skills/mistral-rs.json | 30 ++++++---
.../knowledge/common/skills/mlc-llm.json | 30 ++++++---
.../knowledge/common/skills/mlx-swift-lm.json | 30 ++++++---
.../common/skills/npu-skill-factory.json | 4 +-
.../skills/optimization-search-policy.json | 4 +-
.../common/skills/os-process-placement.json | 4 +-
.../skills/performance-ceiling-analysis.json | 4 +-
.../skills/plan-activation-residency.json | 4 +-
.../knowledge/common/skills/project-maya.json | 32 +++++++---
.../skills/repository-evolution-analysis.json | 4 +-
.../knowledge/common/skills/rsi-evidence.json | 4 +-
.../skills/skill-evolution-playbook.json | 4 +-
.../knowledge/common/skills/splash.json | 32 +++++++---
.../common/skills/storage-io-measurement.json | 4 +-
.../skills/training-input-pipeline.json | 4 +-
.../skills/training-resource-budget.json | 4 +-
.../skills/video-diffusion-optimization.json | 4 +-
.../knowledge/common/skills/vllm-metal.json | 30 ++++++---
.../skills/weight-loading-optimization.json | 4 +-
.../cuda/skills/audit-afd-p2p-exchanges.json | 32 +++++++---
.../skills/audit-flare-session-lifecycle.json | 32 +++++++---
.../diagnose-attention-bottlenecks.json | 4 +-
.../cuda/skills/host-overhead-profiling.json | 4 +-
.../cuda/skills/reason-hopper-kernels.json | 4 +-
79 files changed, 823 insertions(+), 357 deletions(-)
diff --git a/docs/skill-coverage.md b/docs/skill-coverage.md
index 659f69008..247c8d7eb 100644
--- a/docs/skill-coverage.md
+++ b/docs/skill-coverage.md
@@ -4,7 +4,7 @@ Generated from the registry, the skill sidecars and the two maintained
matrices. **Do not edit by hand**; regenerate after any registry, skill,
matrix or frozen-plan change.
-- Input digest: `sha256:43760f8a7e388c4c55bf61fd9680da4ecabe56264eafe5ebe1beacc3ae0b85d8`
+- Input digest: `sha256:aac486c04d4b627a74f2078c266811cefa42aeef8ea920806e2a889fb19851b7`
- Regenerate: `python tools/build_skill_coverage.py`
- Verify without writing: `python tools/build_skill_coverage.py --check`
@@ -79,40 +79,40 @@ optimization is fully covered or that any procedure has measured benefit.
| Skill | Backend | Declared task | Applicability | Conditions | SearchSpace layer / stage | Reference Matrix | State | Gaps |
|---|---|---|---|---|---|---|---|---|
-| [dcp-async-save-performance](../src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/SKILL.md) | common | Diagnose training-visible checkpoint pauses using PyTorch Distributed Checkpoint asynchronous saving and staging ... | **unknown** | — | 1 Accelerator · Diagnose / Configure / Optimize
2 RAM/Host · Observe / Profile / Configure / Optimize / Verify
3 SSD · Observe / Profile / Configure / Optimize / Verify | 1 · Systems sources (C1 / DCP async-save-only approval — pytorch/pytorch @ ...) | candidate · not_run · 1 readset(s) | no-applicability, no-eval |
-| [diffusers-invocation-state](../src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/SKILL.md) | common | Trace Diffusers pipeline-wrapper and scheduler invocation state when repeated calls, from_pipe reuse or interruption ... | FluxPipeline and FlowMatchEulerDiscreteScheduler invocation ownership at this source pin ... | — | 2 RAM/Host †guidance: Diffusers invocation guidance (source-derived) | Diffusers invocation source extraction (huggingface/diffusers @ `aea7b70be0f594bcdc15a6f2677ee2ca85e49e8e` ...) | candidate · not_run · 1 readset(s) | no-eval |
-| [lazy-evaluation-stream-lifetimes](../src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/SKILL.md) | common | Diagnose MLX lazy evaluation, cross-stream ordering and array-buffer lifetimes when timing, memory use or asynchronous ... | MLX evaluation/stream/buffer diagnosis at the pinned implementation; qualify the actual ... | — | 1 Accelerator †guidance: MLX framework guidance (source-derived) | MLX framework source extraction (ml-explore/mlx @ `bff10220c742907f0c68e55ff174aaea56f74e3d` (MIT)) | candidate · not_run · 1 readset(s) | no-eval |
+| [dcp-async-save-performance](../src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/SKILL.md) | common | Diagnose training-visible checkpoint pauses using PyTorch Distributed Checkpoint asynchronous saving and staging ... | Source-supported diagnosis of training-visible PyTorch DCP save pauses ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator · Diagnose / Configure / Optimize
2 RAM/Host · Observe / Profile / Configure / Optimize / Verify
3 SSD · Observe / Profile / Configure / Optimize / Verify | 1 · Systems sources (C1 / DCP async-save-only approval — pytorch/pytorch @ ...) | candidate · not_run · 1 readset(s) | no-eval |
+| [diffusers-invocation-state](../src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/SKILL.md) | common | Trace Diffusers pipeline-wrapper and scheduler invocation state when repeated calls, from_pipe reuse or interruption ... | FluxPipeline and FlowMatchEulerDiscreteScheduler invocation ownership at this source pin ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 2 RAM/Host †guidance: Diffusers invocation guidance (source-derived) | Diffusers invocation source extraction (huggingface/diffusers @ `aea7b70be0f594bcdc15a6f2677ee2ca85e49e8e` ...) | candidate · not_run · 1 readset(s) | no-eval |
+| [lazy-evaluation-stream-lifetimes](../src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/SKILL.md) | common | Diagnose MLX lazy evaluation, cross-stream ordering and array-buffer lifetimes when timing, memory use or asynchronous ... | MLX evaluation/stream/buffer diagnosis at the pinned implementation; qualify the actual ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator †guidance: MLX framework guidance (source-derived) | MLX framework source extraction (ml-explore/mlx @ `bff10220c742907f0c68e55ff174aaea56f74e3d` (MIT)) | candidate · not_run · 1 readset(s) | no-eval |
## Registered: Repositories (26)
| Skill | Backend | Declared task | Applicability | Conditions | SearchSpace layer / stage | Reference Matrix | State | Gaps |
|---|---|---|---|---|---|---|---|---|
-| [audit-afd-p2p-exchanges](../src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/SKILL.md) | cuda | Audit CUDA P2pNccl Attention/FFN rank namespaces and per-stage, per-peer hidden-state ownership when uneven worker ... | Source-based CUDA P2pNccl rank/payload audit; selected eager dense-FFN path with ... | — | 1 Accelerator · Diagnose / Configure / Verify | 1 · Systems sources (vllm-project/afd-plugin @ `df8d6a0c403c3ca1b4ed9cf1687e7204908bebec` ...) | candidate · not_run · 1 readset(s) | no-eval |
-| [audit-flare-session-lifecycle](../src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/SKILL.md) | cuda | Audit Flare complete prefix/session restoration, inbound transfer tickets, publication/use/retirement fences and ... | Pinned Flare complete-session restore and admission audit for its NVIDIA SM120/SM121 ... | — | 1 Accelerator †guidance: Flare preparation and complete-session source contracts
2 RAM/Host †guidance: Flare preparation and complete-session source contracts
3 SSD †guidance: Flare preparation and complete-session source contracts | Flare preparation and complete-session source knowledge (NVIDIA/flare-local-inference → [NVIDIA/local-inference](https://github ...) | candidate · not_run · 2 readset(s) | no-eval |
-| [audit-h3-metal-fasth3-semantics](../src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/SKILL.md) | common | Audit h3-metal FastH3 effective configuration, checkpoint tensor and schedule semantics, partial VSA tiles, sparse ... | h3-metal FastH3 requested/effective configuration, checkpoint transformation and ... | — | 1 Accelerator · Observe / Diagnose / Verify | 1 · Systems sources (zuiho-kai/h3-metal @ `4ba47a3eabdbebf801ac6664b31f8ec168bc57a8` ...) | candidate · not_run · 3 readset(s) | no-eval |
-| [audit-h3-runtime-residency](../src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/SKILL.md) | common | Audit MiniMax-H3 generation in h3.c on Apple Silicon by tracing conditioning, AdaLN precompute, DiT residency, Metal ... | MiniMax-H3 joint audio/video C/Metal runtime audit: raw checkpoint compatibility, input ... | — | 1 Accelerator · Observe / Diagnose / Verify
2 RAM/Host · Observe / Diagnose / Verify
3 SSD · Diagnose / Configure / Verify | 1 · Systems sources (antirez/h3.c @ `8974cc055ea9c02fcd14cc27dfda3e1027c05153` (MIT ...) | candidate · not_run · 3 readset(s) | no-eval |
-| [audit-llama-runtime-state](../src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/SKILL.md) | common | Audit llama.cpp context memory, partial decode progress, graph reuse and server cancellation when a retry, stale ... | Pinned llama.cpp context-memory, graph-reuse and server task-lifecycle diagnosis ... | — | 1 Accelerator · Diagnose / Verify
2 RAM/Host · Diagnose / Verify | 1 · Systems sources (ggml-org/llama.cpp @ `707616feb5f379e99ff7442680a55b4a1704c7c9` (MIT ...) | candidate · not_run · 1 readset(s) | no-eval |
-| [audit-recurrent-kv-finalization](../src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/SKILL.md) | common | Audit vLLM-RLT synchronous early-exit KV propagation, later depth-specific attention reads, and request storage ... | Pinned vLLM-RLT native Ouro synchronous scheduling with host KVMetadata: LAST_EXITED ... | — | — | 1 · Systems sources (InferMatrixLab/vllm-rlt (legacy ThinkFlowLab/vllm-rlt) @ ...) | candidate · not_run · 2 readset(s) | no-SearchSpace-layer, no-eval |
-| [audit-sd-cpp-residency](../src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/SKILL.md) | common | Audit stable-diffusion.cpp image-generation memory pressure by separating catalog freshness and weight-byte delivery ... | Source-grounded audit of native image-generation weight/runtime residency, optional ... | — | 1 Accelerator · Observe / Diagnose / Verify
2 RAM/Host · Observe / Diagnose / Verify
3 SSD · Diagnose / Verify | 1 · Systems sources (leejet/stable-diffusion.cpp @ `b13a48219d5cdf408edcb93da6cd69b612ed03d ...) | candidate · not_run · 2 readset(s) | no-eval |
-| [audit-sglang-diffusion-lifetimes](../src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/SKILL.md) | common | Diagnose SGLang multimodal_gen stage state, component residency and request timeout/deletion when grouped outputs ... | Bounded monolithic multimodal_gen composed-pipeline state and lifecycle diagnosis; actual ... | — | 1 Accelerator · Diagnose / Verify
2 RAM/Host · Diagnose / Verify | 1 · Systems sources (sgl-project/sglang @ `df13d5dce31abfeff0e701a62e8c39a136758e6c` ...) | candidate · not_run · 1 readset(s) | no-eval |
-| [audit-vllm-kv-lifetimes](../src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/SKILL.md) | common | Audit vLLM V1 asynchronous KV load/save completion, block retention, preemption, and receive-failure contracts when ... | **unknown** | — | 1 Accelerator · Observe / Diagnose / Verify | 1 · Systems sources (vllm-project/vllm @ `7f730b4cea9a66520a9200265928ebe016b48a39` ...) | candidate · not_run · 1 readset(s) | no-applicability, no-eval |
-| [bound-windowed-attention-replay](../src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/SKILL.md) | common | Bound replay or prefill query rows across a producer/consumer boundary whose retained inputs and separately preserved ... | Decoders with an explicit producer/consumer boundary and a demonstrated dependency ... | — | 1 Accelerator †guidance: DeepInfer bounded source guidance
2 RAM/Host †guidance: DeepInfer bounded source guidance | DeepInfer source extraction (deepseek-ai/DeepInfer @ `8b16149741a4935e1dec8cea27853f273b374c5b` ...) | candidate · not_run · 2 readset(s) | no-eval |
-| [budget-speculative-verification](../src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/SKILL.md) | common | Decide how many draft tokens to verify per decode step when acceptance varies by request and position. Use when an ... | Engines with multi-token speculative decoding whose verification forward cost grows with ... | — | 1 Accelerator †guidance: DeepInfer bounded source guidance | DeepInfer source extraction (deepseek-ai/DeepInfer @ `8b16149741a4935e1dec8cea27853f273b374c5b` ...) | candidate · not_run · 2 readset(s) | no-eval |
-| [comfy-ui](../src/inferrsi/data/skill_library/repositories/comfy-ui/SKILL.md) | common | Audit ComfyUI workflow-output cache validity, lazy dependencies and model-weight residency as separate state contracts ... | Source-grounded ComfyUI workflow-output cache and model-weight ownership audit; confirm ... | — | 1 Accelerator · Diagnose / Configure / Verify
2 RAM/Host · Diagnose / Configure / Verify | 1 · Systems sources (Comfy-Org/ComfyUI @ `b1a247c9cf9dbfda4b2c0a0c1d50950667eeaa0c` (GNU ...) | candidate · not_run · 1 readset(s) | no-eval |
-| [diagnose-vllm-admission-pressure](../src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/SKILL.md) | common | Diagnose vLLM V1 admission stalls and KV-pressure preemption by separating token, input, active-request ... | **unknown** | — | 1 Accelerator · Observe / Diagnose / Configure / Optimize / Verify | 1 · Systems sources (vllm-project/vllm @ `7f730b4cea9a66520a9200265928ebe016b48a39` ...) | candidate · not_run · 14 readset(s) | no-applicability, no-eval |
-| [dwarfstar](../src/inferrsi/data/skill_library/repositories/dwarfstar/SKILL.md) | common | Investigate DwarfStar model fit, expert SSD streaming, session and vision cache correctness, speculative sampling and ... | Pinned DwarfStar source investigation and experiment planning; not arbitrary GGUF support ... | — | — | DwarfStar portable AREX extraction (antirez/ds4 @ `ee7e8405232c8971000cc7ba2b945b00bf6d20d3`; AREX-Skill ...) | candidate · partial_cpu_only · 2 readset(s) | no-SearchSpace-layer, no-eval |
-| [generation-cache-lifecycle](../src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/SKILL.md) | common | Trace MLX-LM prefill, generation batching, prompt-cache reuse and cooperative stopping when request state, cache rows ... | MLX-LM generation, batch-row and prefix-cache diagnosis; model/backend/runtime ... | — | 2 RAM/Host †guidance: MLX-LM application guidance (source-derived) | MLX-LM application source extraction (ml-explore/mlx-lm @ `ab7f7048dc341ced2d84b1887eeb17d60b0d1b02` (MIT)) | candidate · not_run · 1 readset(s) | no-eval |
-| [husky](../src/inferrsi/data/skill_library/repositories/husky/SKILL.md) | common | Review Husky's official-document claims as research hypotheses, separating author reports, implementation evidence ... | Review Husky document claims and specify missing evidence; common catalog scope is not ... | — | — | Husky document-only research adaptation (Official https://husky.underdog.ai/; displayed2026-09-20 ...) | candidate · not_run · 2 readset(s) | no-SearchSpace-layer, no-eval |
-| [kv-tier-management](../src/inferrsi/data/skill_library/repositories/lmcache/kv-tier-management/SKILL.md) | common | Manage runtime KV cache tiers (GPU → CPU RAM → local disk) for LLM serving: decide when KV is stored, retained ... | **unknown** | — | 2 RAM/Host · Configure / Optimize
3 SSD · Optimize | 1 · Systems sources (LMCache/LMCache @ c3b875a) | candidate · not_run · 0 readset(s) | no-applicability, no-eval |
-| [lily](../src/inferrsi/data/skill_library/repositories/lily/SKILL.md) | common | Investigate Lily's exact Qwen3.6-35B-A3B MLX-affine-Q4 contract, Metal state and kernels, strict greedy text API ... | Exact Lily source/checkpoint/layout/API/cache/qualification investigation; common catalog ... | — | — | Lily portable AREX extraction (perplexityai/pplx-garden `lily/` @ `1ed972ed3f0bd5616c997c9507c25616c6 ...) | candidate · not_run · 2 readset(s) | no-SearchSpace-layer, no-eval |
-| [lithos-metal](../src/inferrsi/data/skill_library/repositories/lithos-metal/SKILL.md) | common | Inspect Lithos Metal compilation and fusion, GPU-resident generation and DSpark state, or local serving and ... | Pinned Lithos Metal 0.1.2 source/API contracts; Apple silicon Metal runtime; model and ... | — | 1 Accelerator · Diagnose / Configure / Optimize / Verify | 1 · Systems sources (lithos-ai/lithos-metal @ `6653b7df5f2c4392ec7b038960f2796b1499923f` ...) | candidate · not_run · 1 readset(s) | no-eval |
-| [local-dream](../src/inferrsi/data/skill_library/repositories/local-dream/SKILL.md) | common | Audit Local Dream Android desired-versus-serving process state, remote exposure, native backend and DiT packaging/ABI ... | Selected Android app/process and native diffusion boundary contracts; common is a ... | — | 2 RAM/Host †guidance: Local Dream Android diffusion source contracts | Local Dream Android diffusion boundary source (xororz/local-dream @ `db162df4d9bb6efc6595250423d8df26241f4c27` ...) | candidate · not_run · 2 readset(s) | no-eval |
-| [mistral-rs](../src/inferrsi/data/skill_library/repositories/mistral-rs/SKILL.md) | common | Investigate mistral.rs Metal paged-attention decode layouts, prefill coordinates and workspace, actual-model prewarm ... | Selected native Metal paged decode/prefill layout/metadata/workspace and ... | — | 1 Accelerator †guidance: mistral.rs Metal paged-attention source contracts
2 RAM/Host †guidance: mistral.rs Metal paged-attention source contracts | mistral.rs Metal paged-attention extraction (EricLBuehler/mistral.rs @ `9459891e7697238225885518119682ac8209a612` ...) | candidate · not_run · 2 readset(s) | no-eval |
-| [mlc-llm](../src/inferrsi/data/skill_library/repositories/mlc-llm/SKILL.md) | common | Investigate MLC LLM compiled Metal artifact/schema, Apple target and pinned TVM module/storage/stream contracts, and ... | Selected artifact/Apple target/pinned TVM Metal consumer and KV configuration source ... | — | 1 Accelerator †guidance: MLC LLM artifact, target and runtime source contracts
2 RAM/Host †guidance: MLC LLM artifact, target and runtime source contracts | MLC LLM compiled Metal and runtime contracts (mlc-ai/mlc-llm @ `8978ea9aec626d64096aaa92fbc89bda2e3ea802`; TVM ...) | candidate · not_run · 2 readset(s) | no-eval |
-| [mlx-swift-lm](../src/inferrsi/data/skill_library/repositories/mlx-swift-lm/SKILL.md) | common | Investigate MLX Swift LM generation ownership, cancellation and settled task completion, and recurrent GatedDelta Metal ... | Selected Swift ownership/cancellation/settling and actual GatedDelta Metal/ops contracts ... | — | 1 Accelerator †guidance: MLX Swift LM ownership and recurrent Metal source contracts
2 RAM/Host †guidance: MLX Swift LM ownership and recurrent Metal source contracts | MLX Swift LM lifecycle and recurrent Metal extraction (ml-explore/mlx-swift-lm @ `3d066d3a0ffdc7e70c476a3d31488ea62dfdd0ac` ...) | candidate · not_run · 2 readset(s) | no-eval |
-| [project-maya](../src/inferrsi/data/skill_library/repositories/project-maya/SKILL.md) | common | Investigate Project Maya model-family dispatch, representation and memory tiers, API versus native state, speculative ... | Model-specific dispatch/residency, API/native state and qualification investigation ... | — | 1 Accelerator †guidance: Project Maya model-specific source contracts
2 RAM/Host †guidance: Project Maya model-specific source contracts
3 SSD †guidance: Project Maya model-specific source contracts | Project Maya source and CPU-slice extraction (mw00/project-maya @ `249f9a87d1a9d72bec2b684d704b0aa9296b1c9c` ...) | candidate · partial_cpu_only · 2 readset(s) | no-eval |
-| [splash](../src/inferrsi/data/skill_library/repositories/splash/SKILL.md) | common | Investigate Splash model-format compatibility, automatic memory planning, Metal and ANE execution, DFlash2 ... | Pinned Splash representation, planning, state/serving and verification investigation ... | — | — | Splash portable AREX extraction (incoai/splash @ `39a97385f8560cf8a93972cec9b0a963c9195250`; AREX ...) | candidate · partial_cpu_only · 2 readset(s) | no-SearchSpace-layer, no-eval |
-| [vllm-metal](../src/inferrsi/data/skill_library/repositories/vllm-metal/SKILL.md) | common | Investigate vllm-metal Apple inference through platform and native-ABI selection, paged KV/offload ownership ... | Selected plugin preparation, KV/offload ownership and speculative/deferred/dispatch ... | — | 1 Accelerator †guidance: vLLM Metal selected source contracts
2 RAM/Host †guidance: vLLM Metal selected source contracts
3 SSD †guidance: vLLM Metal selected source contracts | vLLM Metal portable AREX source and CPU extraction (vllm-project/vllm-metal @ `e7df7a6ff215f99c7842552d230fdbc709835934` ...) | candidate · partial_cpu_only · 2 readset(s) | no-eval |
+| [audit-afd-p2p-exchanges](../src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/SKILL.md) | cuda | Audit CUDA P2pNccl Attention/FFN rank namespaces and per-stage, per-peer hidden-state ownership when uneven worker ... | Source-based CUDA P2pNccl rank/payload audit; selected eager dense-FFN path with ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator · Diagnose / Configure / Verify | 1 · Systems sources (vllm-project/afd-plugin @ `df8d6a0c403c3ca1b4ed9cf1687e7204908bebec` ...) | candidate · not_run · 1 readset(s) | no-eval |
+| [audit-flare-session-lifecycle](../src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/SKILL.md) | cuda | Audit Flare complete prefix/session restoration, inbound transfer tickets, publication/use/retirement fences and ... | Pinned Flare complete-session restore and admission audit for its NVIDIA SM120/SM121 ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator †guidance: Flare preparation and complete-session source contracts
2 RAM/Host †guidance: Flare preparation and complete-session source contracts
3 SSD †guidance: Flare preparation and complete-session source contracts | Flare preparation and complete-session source knowledge (NVIDIA/flare-local-inference → [NVIDIA/local-inference](https://github ...) | candidate · not_run · 2 readset(s) | no-eval |
+| [audit-h3-metal-fasth3-semantics](../src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/SKILL.md) | common | Audit h3-metal FastH3 effective configuration, checkpoint tensor and schedule semantics, partial VSA tiles, sparse ... | h3-metal FastH3 requested/effective configuration, checkpoint transformation and ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator · Observe / Diagnose / Verify | 1 · Systems sources (zuiho-kai/h3-metal @ `4ba47a3eabdbebf801ac6664b31f8ec168bc57a8` ...) | candidate · not_run · 3 readset(s) | no-eval |
+| [audit-h3-runtime-residency](../src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/SKILL.md) | common | Audit MiniMax-H3 generation in h3.c on Apple Silicon by tracing conditioning, AdaLN precompute, DiT residency, Metal ... | MiniMax-H3 joint audio/video C/Metal runtime audit: raw checkpoint compatibility, input ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator · Observe / Diagnose / Verify
2 RAM/Host · Observe / Diagnose / Verify
3 SSD · Diagnose / Configure / Verify | 1 · Systems sources (antirez/h3.c @ `8974cc055ea9c02fcd14cc27dfda3e1027c05153` (MIT ...) | candidate · not_run · 3 readset(s) | no-eval |
+| [audit-llama-runtime-state](../src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/SKILL.md) | common | Audit llama.cpp context memory, partial decode progress, graph reuse and server cancellation when a retry, stale ... | Pinned llama.cpp context-memory, graph-reuse and server task-lifecycle diagnosis ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator · Diagnose / Verify
2 RAM/Host · Diagnose / Verify | 1 · Systems sources (ggml-org/llama.cpp @ `707616feb5f379e99ff7442680a55b4a1704c7c9` (MIT ...) | candidate · not_run · 1 readset(s) | no-eval |
+| [audit-recurrent-kv-finalization](../src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/SKILL.md) | common | Audit vLLM-RLT synchronous early-exit KV propagation, later depth-specific attention reads, and request storage ... | Pinned vLLM-RLT native Ouro synchronous scheduling with host KVMetadata: LAST_EXITED ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | — | 1 · Systems sources (InferMatrixLab/vllm-rlt (legacy ThinkFlowLab/vllm-rlt) @ ...) | candidate · not_run · 2 readset(s) | no-SearchSpace-layer, no-eval |
+| [audit-sd-cpp-residency](../src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/SKILL.md) | common | Audit stable-diffusion.cpp image-generation memory pressure by separating catalog freshness and weight-byte delivery ... | Source-grounded audit of native image-generation weight/runtime residency, optional ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator · Observe / Diagnose / Verify
2 RAM/Host · Observe / Diagnose / Verify
3 SSD · Diagnose / Verify | 1 · Systems sources (leejet/stable-diffusion.cpp @ `b13a48219d5cdf408edcb93da6cd69b612ed03d ...) | candidate · not_run · 2 readset(s) | no-eval |
+| [audit-sglang-diffusion-lifetimes](../src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/SKILL.md) | common | Diagnose SGLang multimodal_gen stage state, component residency and request timeout/deletion when grouped outputs ... | Bounded monolithic multimodal_gen composed-pipeline state and lifecycle diagnosis; actual ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator · Diagnose / Verify
2 RAM/Host · Diagnose / Verify | 1 · Systems sources (sgl-project/sglang @ `df13d5dce31abfeff0e701a62e8c39a136758e6c` ...) | candidate · not_run · 1 readset(s) | no-eval |
+| [audit-vllm-kv-lifetimes](../src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/SKILL.md) | common | Audit vLLM V1 asynchronous KV load/save completion, block retention, preemption, and receive-failure contracts when ... | Pinned vLLM V1 scheduler/connector ownership audit for asynchronous KV load/save ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator · Observe / Diagnose / Verify | 1 · Systems sources (vllm-project/vllm @ `7f730b4cea9a66520a9200265928ebe016b48a39` ...) | candidate · not_run · 1 readset(s) | no-eval |
+| [bound-windowed-attention-replay](../src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/SKILL.md) | common | Bound replay or prefill query rows across a producer/consumer boundary whose retained inputs and separately preserved ... | Decoders with an explicit producer/consumer boundary and a demonstrated dependency ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator †guidance: DeepInfer bounded source guidance
2 RAM/Host †guidance: DeepInfer bounded source guidance | DeepInfer source extraction (deepseek-ai/DeepInfer @ `8b16149741a4935e1dec8cea27853f273b374c5b` ...) | candidate · not_run · 2 readset(s) | no-eval |
+| [budget-speculative-verification](../src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/SKILL.md) | common | Decide how many draft tokens to verify per decode step when acceptance varies by request and position. Use when an ... | Engines with multi-token speculative decoding whose verification forward cost grows with ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator †guidance: DeepInfer bounded source guidance | DeepInfer source extraction (deepseek-ai/DeepInfer @ `8b16149741a4935e1dec8cea27853f273b374c5b` ...) | candidate · not_run · 2 readset(s) | no-eval |
+| [comfy-ui](../src/inferrsi/data/skill_library/repositories/comfy-ui/SKILL.md) | common | Audit ComfyUI workflow-output cache validity, lazy dependencies and model-weight residency as separate state contracts ... | Source-grounded ComfyUI workflow-output cache and model-weight ownership audit; confirm ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator · Diagnose / Configure / Verify
2 RAM/Host · Diagnose / Configure / Verify | 1 · Systems sources (Comfy-Org/ComfyUI @ `b1a247c9cf9dbfda4b2c0a0c1d50950667eeaa0c` (GNU ...) | candidate · not_run · 1 readset(s) | no-eval |
+| [diagnose-vllm-admission-pressure](../src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/SKILL.md) | common | Diagnose vLLM V1 admission stalls and KV-pressure preemption by separating token, input, active-request ... | Pinned vLLM V1 branch-based diagnosis of queued requests, low scheduled-token counts and ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator · Observe / Diagnose / Configure / Optimize / Verify | 1 · Systems sources (vllm-project/vllm @ `7f730b4cea9a66520a9200265928ebe016b48a39` ...) | candidate · not_run · 14 readset(s) | no-eval |
+| [dwarfstar](../src/inferrsi/data/skill_library/repositories/dwarfstar/SKILL.md) | common | Investigate DwarfStar model fit, expert SSD streaming, session and vision cache correctness, speculative sampling and ... | Pinned DwarfStar source investigation and experiment planning; not arbitrary GGUF support ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | — | DwarfStar portable AREX extraction (antirez/ds4 @ `ee7e8405232c8971000cc7ba2b945b00bf6d20d3`; AREX-Skill ...) | candidate · partial_cpu_only · 2 readset(s) | no-SearchSpace-layer, no-eval |
+| [generation-cache-lifecycle](../src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/SKILL.md) | common | Trace MLX-LM prefill, generation batching, prompt-cache reuse and cooperative stopping when request state, cache rows ... | MLX-LM generation, batch-row and prefix-cache diagnosis; model/backend/runtime ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 2 RAM/Host †guidance: MLX-LM application guidance (source-derived) | MLX-LM application source extraction (ml-explore/mlx-lm @ `ab7f7048dc341ced2d84b1887eeb17d60b0d1b02` (MIT)) | candidate · not_run · 1 readset(s) | no-eval |
+| [husky](../src/inferrsi/data/skill_library/repositories/husky/SKILL.md) | common | Review Husky's official-document claims as research hypotheses, separating author reports, implementation evidence ... | Review Husky document claims and specify missing evidence; common catalog scope is not ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | — | Husky document-only research adaptation (Official https://husky.underdog.ai/; displayed2026-09-20 ...) | candidate · not_run · 2 readset(s) | no-SearchSpace-layer, no-eval |
+| [kv-tier-management](../src/inferrsi/data/skill_library/repositories/lmcache/kv-tier-management/SKILL.md) | common | Manage runtime KV cache tiers (GPU → CPU RAM → local disk) for LLM serving: decide when KV is stored, retained ... | LMCache multiprocess runtime KV tier policy across engine GPU cache, host L1 and ... | failure_modes, input_contract, produced_artifact
3 advice | 2 RAM/Host · Configure / Optimize
3 SSD · Optimize | 1 · Systems sources (LMCache/LMCache @ c3b875a) | candidate · not_run · 0 readset(s) | no-eval |
+| [lily](../src/inferrsi/data/skill_library/repositories/lily/SKILL.md) | common | Investigate Lily's exact Qwen3.6-35B-A3B MLX-affine-Q4 contract, Metal state and kernels, strict greedy text API ... | Exact Lily source/checkpoint/layout/API/cache/qualification investigation; common catalog ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | — | Lily portable AREX extraction (perplexityai/pplx-garden `lily/` @ `1ed972ed3f0bd5616c997c9507c25616c6 ...) | candidate · not_run · 2 readset(s) | no-SearchSpace-layer, no-eval |
+| [lithos-metal](../src/inferrsi/data/skill_library/repositories/lithos-metal/SKILL.md) | common | Inspect Lithos Metal compilation and fusion, GPU-resident generation and DSpark state, or local serving and ... | Pinned Lithos Metal 0.1.2 source/API contracts; Apple silicon Metal runtime; model and ... | failure_modes, input_contract, produced_artifact
3 advice | 1 Accelerator · Diagnose / Configure / Optimize / Verify | 1 · Systems sources (lithos-ai/lithos-metal @ `6653b7df5f2c4392ec7b038960f2796b1499923f` ...) | candidate · not_run · 1 readset(s) | no-eval |
+| [local-dream](../src/inferrsi/data/skill_library/repositories/local-dream/SKILL.md) | common | Audit Local Dream Android desired-versus-serving process state, remote exposure, native backend and DiT packaging/ABI ... | Selected Android app/process and native diffusion boundary contracts; common is a ... | failure_modes, input_contract, produced_artifact
3 advice | 2 RAM/Host †guidance: Local Dream Android diffusion source contracts | Local Dream Android diffusion boundary source (xororz/local-dream @ `db162df4d9bb6efc6595250423d8df26241f4c27` ...) | candidate · not_run · 2 readset(s) | no-eval |
+| [mistral-rs](../src/inferrsi/data/skill_library/repositories/mistral-rs/SKILL.md) | common | Investigate mistral.rs Metal paged-attention decode layouts, prefill coordinates and workspace, actual-model prewarm ... | Selected native Metal paged decode/prefill layout/metadata/workspace and ... | failure_modes, input_contract, produced_artifact
3 advice | 1 Accelerator †guidance: mistral.rs Metal paged-attention source contracts
2 RAM/Host †guidance: mistral.rs Metal paged-attention source contracts | mistral.rs Metal paged-attention extraction (EricLBuehler/mistral.rs @ `9459891e7697238225885518119682ac8209a612` ...) | candidate · not_run · 2 readset(s) | no-eval |
+| [mlc-llm](../src/inferrsi/data/skill_library/repositories/mlc-llm/SKILL.md) | common | Investigate MLC LLM compiled Metal artifact/schema, Apple target and pinned TVM module/storage/stream contracts, and ... | Selected artifact/Apple target/pinned TVM Metal consumer and KV configuration source ... | failure_modes, input_contract, produced_artifact
3 advice | 1 Accelerator †guidance: MLC LLM artifact, target and runtime source contracts
2 RAM/Host †guidance: MLC LLM artifact, target and runtime source contracts | MLC LLM compiled Metal and runtime contracts (mlc-ai/mlc-llm @ `8978ea9aec626d64096aaa92fbc89bda2e3ea802`; TVM ...) | candidate · not_run · 2 readset(s) | no-eval |
+| [mlx-swift-lm](../src/inferrsi/data/skill_library/repositories/mlx-swift-lm/SKILL.md) | common | Investigate MLX Swift LM generation ownership, cancellation and settled task completion, and recurrent GatedDelta Metal ... | Selected Swift ownership/cancellation/settling and actual GatedDelta Metal/ops contracts ... | failure_modes, input_contract, produced_artifact
3 advice | 1 Accelerator †guidance: MLX Swift LM ownership and recurrent Metal source contracts
2 RAM/Host †guidance: MLX Swift LM ownership and recurrent Metal source contracts | MLX Swift LM lifecycle and recurrent Metal extraction (ml-explore/mlx-swift-lm @ `3d066d3a0ffdc7e70c476a3d31488ea62dfdd0ac` ...) | candidate · not_run · 2 readset(s) | no-eval |
+| [project-maya](../src/inferrsi/data/skill_library/repositories/project-maya/SKILL.md) | common | Investigate Project Maya model-family dispatch, representation and memory tiers, API versus native state, speculative ... | Model-specific dispatch/residency, API/native state and qualification investigation ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator †guidance: Project Maya model-specific source contracts
2 RAM/Host †guidance: Project Maya model-specific source contracts
3 SSD †guidance: Project Maya model-specific source contracts | Project Maya source and CPU-slice extraction (mw00/project-maya @ `249f9a87d1a9d72bec2b684d704b0aa9296b1c9c` ...) | candidate · partial_cpu_only · 2 readset(s) | no-eval |
+| [splash](../src/inferrsi/data/skill_library/repositories/splash/SKILL.md) | common | Investigate Splash model-format compatibility, automatic memory planning, Metal and ANE execution, DFlash2 ... | Pinned Splash representation, planning, state/serving and verification investigation ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | — | Splash portable AREX extraction (incoai/splash @ `39a97385f8560cf8a93972cec9b0a963c9195250`; AREX ...) | candidate · partial_cpu_only · 2 readset(s) | no-SearchSpace-layer, no-eval |
+| [vllm-metal](../src/inferrsi/data/skill_library/repositories/vllm-metal/SKILL.md) | common | Investigate vllm-metal Apple inference through platform and native-ABI selection, paged KV/offload ownership ... | Selected plugin preparation, KV/offload ownership and speculative/deferred/dispatch ... | failure_modes, input_contract, produced_artifact
3 advice | 1 Accelerator †guidance: vLLM Metal selected source contracts
2 RAM/Host †guidance: vLLM Metal selected source contracts
3 SSD †guidance: vLLM Metal selected source contracts | vLLM Metal portable AREX source and CPU extraction (vllm-project/vllm-metal @ `e7df7a6ff215f99c7842552d230fdbc709835934` ...) | candidate · partial_cpu_only · 2 readset(s) | no-eval |
## Registered: Methods (20)
@@ -221,7 +221,7 @@ an entrypoint routes to procedures below it rather than adding one.
| `no-SearchSpace-layer` | 15: ascend-video-diffusion-optimization, audit-fixed-workload-claim, audit-recurrent-kv-finalization, cite-check-downstream-pr, competitor-scan, draft-acceptance-issue, dwarfstar, husky, inference-benchmark-design, lily, npu-skill-factory, repository-evolution-analysis, skill-evolution-playbook, splash, video-diffusion-optimization |
| `no-Reference-Matrix-row` | 8: ascend-video-diffusion-optimization, audit-fixed-workload-claim, cite-check-downstream-pr, claim-contract, competitor-scan, draft-acceptance-issue, rsi-evidence, video-diffusion-optimization |
| `no-sidecar` | 0: — |
-| `no-applicability` | 41: as listed in the Gaps column above |
+| `no-applicability` | 37: as listed in the Gaps column above |
| `no-eval` | 75: as listed in the Gaps column above |
These are inventory gaps, not defects. A workflow skill may legitimately
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/README.md b/src/inferrsi/data/repo_packs/inferrsi/knowledge/README.md
index 1aba78e0f..745c2a8d3 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/README.md
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/README.md
@@ -15,7 +15,7 @@ These are knowledge abstracts (`--kind knowledge`), with namespaced package
IDs distinct from explicitly selected `checkout_ref` skills. Ascend/CUDA
procedures use their corresponding backend; shared methods use `common`.
-Source snapshot: [95b1e4bac05edd101392a70920fc2295074f0848](https://github.com/InferMatrixLab/InferRSI/tree/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library).
+Source snapshot: [224f6534dc5596ff78fd2f8329d73658c6bc78da](https://github.com/InferMatrixLab/InferRSI/tree/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library).
```bash
inferrsi knowledge search 'profiling' --repo-type inferrsi --limit 50
@@ -29,7 +29,7 @@ contain the current input paths and bytes (Git checkout filters are respected).
```bash
python tools/build_skill_index.py --source-revision
-python tools/build_skill_index.py --source-revision 95b1e4bac05edd101392a70920fc2295074f0848 --check
+python tools/build_skill_index.py --source-revision 224f6534dc5596ff78fd2f8329d73658c6bc78da --check
```
`--check` detects missing, stale and unexpected generated records without writing.
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-aclgraph-capture-replay.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-aclgraph-capture-replay.json
index 0610a0059..467c280e6 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-aclgraph-capture-replay.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-aclgraph-capture-replay.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/hardware/ascend/ascend-aclgraph-capture-replay/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/hardware/ascend/ascend-aclgraph-capture-replay/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-aclgraph-capture-replay/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-aclgraph-capture-replay/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-device-health.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-device-health.json
index 7357a200e..c772ccb0f 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-device-health.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-device-health.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/hardware/ascend/ascend-device-health/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/hardware/ascend/ascend-device-health/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-device-health/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-device-health/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-distributed-parallel.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-distributed-parallel.json
index 026359c20..41bc38666 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-distributed-parallel.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-distributed-parallel.json
@@ -27,8 +27,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/hardware/ascend/ascend-distributed-parallel/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/hardware/ascend/ascend-distributed-parallel/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-distributed-parallel/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-distributed-parallel/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-graph-mode.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-graph-mode.json
index 63f56357a..256e50fd9 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-graph-mode.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-graph-mode.json
@@ -26,8 +26,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/hardware/ascend/ascend-graph-mode/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/hardware/ascend/ascend-graph-mode/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-graph-mode/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-graph-mode/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-model-adaptation.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-model-adaptation.json
index 0c5eaea45..9e7b5a5f0 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-model-adaptation.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-model-adaptation.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/hardware/ascend/ascend-model-adaptation/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/hardware/ascend/ascend-model-adaptation/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-model-adaptation/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-model-adaptation/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-profiling.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-profiling.json
index 5bfd264e9..d3cd21640 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-profiling.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-profiling.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-profiling/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-profiling/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-profiling/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-profiling/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-tuning.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-tuning.json
index 3444a5836..d95afd452 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-tuning.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-tuning.json
@@ -24,8 +24,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-tuning/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-tuning/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-tuning/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-tuning/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-optimization-casebook.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-optimization-casebook.json
index ebe1fef6e..c0943f86d 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-optimization-casebook.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-optimization-casebook.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/hardware/ascend/ascend-optimization-casebook/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/hardware/ascend/ascend-optimization-casebook/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-optimization-casebook/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-optimization-casebook/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-performance-analysis.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-performance-analysis.json
index 8dbac6ea2..fb47f4d6f 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-performance-analysis.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-performance-analysis.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/hardware/ascend/ascend-performance-analysis/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/hardware/ascend/ascend-performance-analysis/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-performance-analysis/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-performance-analysis/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-video-diffusion-optimization.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-video-diffusion-optimization.json
index 6d41c745e..ae629ffa8 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-video-diffusion-optimization.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-video-diffusion-optimization.json
@@ -26,8 +26,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/hardware/ascend/ascend-video-diffusion-optimization/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/hardware/ascend/ascend-video-diffusion-optimization/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-video-diffusion-optimization/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-video-diffusion-optimization/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascendc-tiling-design.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascendc-tiling-design.json
index 05bcbb1f8..ada79f388 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascendc-tiling-design.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascendc-tiling-design.json
@@ -24,8 +24,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/hardware/ascend/ascendc-tiling-design/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/hardware/ascend/ascendc-tiling-design/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascendc-tiling-design/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascendc-tiling-design/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/cann-attribution-overlap.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/cann-attribution-overlap.json
index a39a0572b..acb5add78 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/cann-attribution-overlap.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/cann-attribution-overlap.json
@@ -24,8 +24,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/hardware/ascend/cann-attribution-overlap/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/hardware/ascend/cann-attribution-overlap/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/cann-attribution-overlap/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/cann-attribution-overlap/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-device-memory-lifecycle.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-device-memory-lifecycle.json
index 5e31dee93..34da136ca 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-device-memory-lifecycle.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-device-memory-lifecycle.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/hardware/ascend/torchnpu-device-memory-lifecycle/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/hardware/ascend/torchnpu-device-memory-lifecycle/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/torchnpu-device-memory-lifecycle/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/torchnpu-device-memory-lifecycle/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-pinned-host-memory-lifecycle.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-pinned-host-memory-lifecycle.json
index f9852f7ae..573bf36a4 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-pinned-host-memory-lifecycle.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-pinned-host-memory-lifecycle.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/hardware/ascend/torchnpu-pinned-host-memory-lifecycle/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/hardware/ascend/torchnpu-pinned-host-memory-lifecycle/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/torchnpu-pinned-host-memory-lifecycle/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/torchnpu-pinned-host-memory-lifecycle/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-serving-preemption.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-serving-preemption.json
index 04d778b86..493140b27 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-serving-preemption.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-serving-preemption.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/methods/assess-serving-preemption/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/methods/assess-serving-preemption/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/methods/assess-serving-preemption/SKILL.md",
"src/inferrsi/data/skill_library/methods/assess-serving-preemption/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-split-backward-schedule.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-split-backward-schedule.json
index 9a06b4b11..cf9137b3b 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-split-backward-schedule.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-split-backward-schedule.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/methods/assess-split-backward-schedule/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/methods/assess-split-backward-schedule/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/methods/assess-split-backward-schedule/SKILL.md",
"src/inferrsi/data/skill_library/methods/assess-split-backward-schedule/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-fixed-workload-claim.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-fixed-workload-claim.json
index c70f2ed85..524ec576b 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-fixed-workload-claim.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-fixed-workload-claim.json
@@ -20,8 +20,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/methods/audit-fixed-workload-claim/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/methods/audit-fixed-workload-claim/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/methods/audit-fixed-workload-claim/SKILL.md",
"src/inferrsi/data/skill_library/methods/audit-fixed-workload-claim/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-metal-fasth3-semantics.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-metal-fasth3-semantics.json
index 3d744a251..8c330b417 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-metal-fasth3-semantics.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-metal-fasth3-semantics.json
@@ -29,8 +29,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/LICENSE",
"src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/SKILL.md",
@@ -56,7 +56,7 @@
"source_files": {
"src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/LICENSE": "663e51edcd9a3a8f858018ce6924ad2895a24c269c26c39ad75325bcd0f7e5bc",
"src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/SKILL.md": "e5205de6eba52b3d1693666af755640697717e52ee7834e1ae30e4071b208d54",
- "src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/metadata.json": "86a67bb56110c3a9e191e0ff576194162e960cbddca0a99c593ea7aef8df8a4e",
+ "src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/metadata.json": "f1681981db4f9077ee27ff71571dceecd69daf25136bd374b7d9ad3b59762a6a",
"src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/references/THIRD_PARTY_NOTICES.md": "a0d77db7d2aca0eda811ab3c386cc32b3143f2c2609a042cf4642954f0f6fb09",
"src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/references/preflight.md": "5cf745005914c74b18a09f127d1d02ce0cb64f6cd7899501ce3cb40938c42a4c",
"src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/references/repo-provenance.md": "6a4c14207223dcc2584f7133bd14ff4590c2258fc2bd44f3721b6bf63e3e35cf",
@@ -148,7 +148,12 @@
"method": "Bounded fork-specific source audit, bundled selected-declaration helper and assertion-backed cases. Native Apple/Metal/model evidence remains blocked; no numerical/performance/skill-efficacy claim."
},
"applicability": {
- "scope": "h3-metal FastH3 requested/effective configuration, checkpoint transformation and schedule/RoPE semantics, VSA prefix/partial tiles and dispatch/fallback audit. Common catalog placement is discovery, not CPU inference or Apple qualification."
+ "scope": "h3-metal FastH3 requested/effective configuration, checkpoint transformation and schedule/RoPE semantics, VSA prefix/partial tiles and dispatch/fallback audit. Common catalog placement is discovery, not CPU inference or Apple qualification.",
+ "prerequisites": "Access to the fork-specific source and semantic contracts; native checks additionally need their matching Metal/model resources and separate execution authority.",
+ "input_contract": "Source revision/patches, requested parameters and explicitly captured environment controls, checkpoint path/key-family/config identity, requested/aligned frames, latent dimensions, packed prefix segments and sparse-pipeline availability or unknown.",
+ "produced_artifact": "A ledger of requested and normalized parameters, ambient overrides, source invariants, counterexamples, observed/unknown facts and the smallest discriminating next check.",
+ "failure_modes": "Ignoring ambient overrides, mixing checkpoint identity switches, using global rather than per-head Q/K/V concatenation, pooling padded boundary rows, or double-adding coarse attention can invalidate the semantic audit.",
+ "refusal_point": "Stop on conflicting configuration, mixed checkpoint identity, missing resources or the authorized boundary; name the missing native test/fixture and leave it blocked/not_run."
},
"limitations": [
"BLOCKED_REQUIRED_BACKEND: native Apple/Metal, actual FastH3 checkpoint/schedule/sparse parity, audio/video quality, target memory fit, performance and service evidence not_run.",
@@ -163,14 +168,25 @@
"references/source-identity.json",
"references/repo-provenance.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Record requested parameters, normalized values and later environment overrides separately, including requested versus aligned frame counts.",
+ "Audit checkpoint transformations and schedule consumers beyond tensor names/shapes; tiny permutation fixtures can falsify mappings but do not establish model parity.",
+ "Resolve sparse eligibility and the dense-fine-plus-gated-coarse fallback explicitly; preserve unknown pipeline availability."
+ ]
},
"applicability": {
- "scope": "h3-metal FastH3 requested/effective configuration, checkpoint transformation and schedule/RoPE semantics, VSA prefix/partial tiles and dispatch/fallback audit. Common catalog placement is discovery, not CPU inference or Apple qualification."
+ "scope": "h3-metal FastH3 requested/effective configuration, checkpoint transformation and schedule/RoPE semantics, VSA prefix/partial tiles and dispatch/fallback audit. Common catalog placement is discovery, not CPU inference or Apple qualification.",
+ "prerequisites": "Access to the fork-specific source and semantic contracts; native checks additionally need their matching Metal/model resources and separate execution authority.",
+ "input_contract": "Source revision/patches, requested parameters and explicitly captured environment controls, checkpoint path/key-family/config identity, requested/aligned frames, latent dimensions, packed prefix segments and sparse-pipeline availability or unknown.",
+ "produced_artifact": "A ledger of requested and normalized parameters, ambient overrides, source invariants, counterexamples, observed/unknown facts and the smallest discriminating next check.",
+ "failure_modes": "Ignoring ambient overrides, mixing checkpoint identity switches, using global rather than per-head Q/K/V concatenation, pooling padded boundary rows, or double-adding coarse attention can invalidate the semantic audit.",
+ "refusal_point": "Stop on conflicting configuration, mixed checkpoint identity, missing resources or the authorized boundary; name the missing native test/fixture and leave it blocked/not_run."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Record requested parameters, normalized values and later environment overrides separately, including requested versus aligned frame counts.",
+ "Audit checkpoint transformations and schedule consumers beyond tensor names/shapes; tiny permutation fixtures can falsify mappings but do not establish model parity.",
+ "Resolve sparse eligibility and the dense-fine-plus-gated-coarse fallback explicitly; preserve unknown pipeline availability."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-runtime-residency.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-runtime-residency.json
index c6c800c98..3f4f8ccdd 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-runtime-residency.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-runtime-residency.json
@@ -30,8 +30,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/LICENSE",
"src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/SKILL.md",
@@ -57,7 +57,7 @@
"source_files": {
"src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/LICENSE": "663e51edcd9a3a8f858018ce6924ad2895a24c269c26c39ad75325bcd0f7e5bc",
"src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/SKILL.md": "fdb4d9483c6efa836f9e59617da605fa9c2186e11c2041b05d5c81850ba0475d",
- "src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/metadata.json": "fe626a77af3f084f36fd73ffdb6f67976e7dceeafb94ba22078ef03c3a96cfae",
+ "src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/metadata.json": "952bd2947b68e56d249f10798e028d191a8ffbdb2f123775be66946d3e2ef77e",
"src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/references/THIRD_PARTY_NOTICES.md": "a0d77db7d2aca0eda811ab3c386cc32b3143f2c2609a042cf4642954f0f6fb09",
"src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/references/minimax-h3-apple.md": "57ce01754ba96c378f13a91305ef3d418b926f99b2753f0af99315572bce063a",
"src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/references/minimax-h3-weight-inventory.json": "fbaacaeec340d3ce893556ee864405a0746af4d13530a38cb3d5aa9d60929271",
@@ -177,7 +177,12 @@
"method": "Manual pinned-source distillation informed by the existing Repo-To-Skill ledger; no AREX/DisCo execution or behavioral efficacy claim."
},
"applicability": {
- "scope": "MiniMax-H3 joint audio/video C/Metal runtime audit: raw checkpoint compatibility, input encoders, schedule precompute, phased/cached residency, numerical controls and SSD streaming on Apple Silicon. Common catalog placement is a discovery classification, not cross-backend qualification."
+ "scope": "MiniMax-H3 joint audio/video C/Metal runtime audit: raw checkpoint compatibility, input encoders, schedule precompute, phased/cached residency, numerical controls and SSD streaming on Apple Silicon. Common catalog placement is a discovery classification, not cross-backend qualification.",
+ "prerequisites": "Access to the pinned source architecture and selected component/format inventory; native checks need matching model, fixture and backend resources within an authorized experiment budget.",
+ "input_contract": "Runtime revision/patches, macOS/SDK/GPU and memory budget, FL2VA/Ref2VA selection, component formats, prompt and ordered media, requested/aligned frames, canvases, steps/seed, layer/reuse/token/precision policies and preview/interactive-cache state.",
+ "produced_artifact": "A phase/component/owner ledger with resident weights, runtime/host buffers, overlap, release/reuse boundaries and observations/unknowns, followed by a discriminating check, actual result and remaining gap.",
+ "failure_modes": "Estimating fit from model-file bytes, confusing runtime int8 with incompatible MLX bundles, ignoring preview/cache overlap, or reducing DiT residency to address a VAE/host-output peak can produce a wrong capacity diagnosis.",
+ "refusal_point": "Stop at unavailable model/fixture/backend resources or the accepted experiment budget; missing target evidence remains unverified/INCONCLUSIVE."
},
"limitations": [
"Native h3.c/Metal execution, target peak memory, audio/video quality, performance, service behavior and skill efficacy were not run.",
@@ -191,14 +196,25 @@
"references/source-identity.json",
"references/minimax-h3-weight-inventory.json"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Resolve the actual conditioning and output components before counting residency; do not add alternative checkpoints or count shared files twice.",
+ "Trace one-shot, preview and interactive ownership separately, including AdaLN schedule dependencies and prepared-cache reset boundaries.",
+ "Classify approximation and precision controls before comparison, then measure the relevant phase with tensor, physical/compressed/swap and storage observations."
+ ]
},
"applicability": {
- "scope": "MiniMax-H3 joint audio/video C/Metal runtime audit: raw checkpoint compatibility, input encoders, schedule precompute, phased/cached residency, numerical controls and SSD streaming on Apple Silicon. Common catalog placement is a discovery classification, not cross-backend qualification."
+ "scope": "MiniMax-H3 joint audio/video C/Metal runtime audit: raw checkpoint compatibility, input encoders, schedule precompute, phased/cached residency, numerical controls and SSD streaming on Apple Silicon. Common catalog placement is a discovery classification, not cross-backend qualification.",
+ "prerequisites": "Access to the pinned source architecture and selected component/format inventory; native checks need matching model, fixture and backend resources within an authorized experiment budget.",
+ "input_contract": "Runtime revision/patches, macOS/SDK/GPU and memory budget, FL2VA/Ref2VA selection, component formats, prompt and ordered media, requested/aligned frames, canvases, steps/seed, layer/reuse/token/precision policies and preview/interactive-cache state.",
+ "produced_artifact": "A phase/component/owner ledger with resident weights, runtime/host buffers, overlap, release/reuse boundaries and observations/unknowns, followed by a discriminating check, actual result and remaining gap.",
+ "failure_modes": "Estimating fit from model-file bytes, confusing runtime int8 with incompatible MLX bundles, ignoring preview/cache overlap, or reducing DiT residency to address a VAE/host-output peak can produce a wrong capacity diagnosis.",
+ "refusal_point": "Stop at unavailable model/fixture/backend resources or the accepted experiment budget; missing target evidence remains unverified/INCONCLUSIVE."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Resolve the actual conditioning and output components before counting residency; do not add alternative checkpoints or count shared files twice.",
+ "Trace one-shot, preview and interactive ownership separately, including AdaLN schedule dependencies and prepared-cache reset boundaries.",
+ "Classify approximation and precision controls before comparison, then measure the relevant phase with tensor, physical/compressed/swap and storage observations."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-llama-runtime-state.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-llama-runtime-state.json
index 21fde87ea..b5910acf6 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-llama-runtime-state.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-llama-runtime-state.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/SKILL.md",
"src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/metadata.json",
@@ -41,7 +41,7 @@
"checkout_ref": "checkout:audit-llama-runtime-state",
"source_files": {
"src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/SKILL.md": "95e81b6bc096a954fa73139ea5bc001d1efa6c99747161da62f012b494e5c013",
- "src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/metadata.json": "1d30b31dc8352fc6a8d7b8abbef8118c840e47e5b90795dae0c758bf9717278c",
+ "src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/metadata.json": "22f1d01b48ea6e2b2ef2deeb20c1404ce8b586eb18442b5adb1390310b4b7503",
"src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/references/source-notes.md": "3d12a401e9decb55a2f4f3b2bfa0d360ef5110f63a0469dfe285654e78054190",
"src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
},
@@ -94,7 +94,12 @@
}
],
"applicability": {
- "scope": "Pinned llama.cpp context-memory, graph-reuse and server task-lifecycle diagnosis; concrete model/backend behavior requires native validation."
+ "scope": "Pinned llama.cpp context-memory, graph-reuse and server task-lifecycle diagnosis; concrete model/backend behavior requires native validation.",
+ "prerequisites": "Pinned llama.cpp source or a concrete runtime trace, access to the relevant memory/graph/server ownership evidence and an identified caller recovery policy that remains in force.",
+ "input_contract": "Source/build and model revisions, memory type, actual backend/device and execution mode, context/batch/ubatch parameters, sequence IDs/positions/output flags and return status; graph cases need dispatch evidence, server cases need correlated HTTP/task/slot/cache logs.",
+ "produced_artifact": "An owner/event ledger of ubatch progress, retained memory, graph-gate decisions, input readiness and slot release/retention, with the first supported broken obligation or smallest missing observation.",
+ "failure_modes": "Assuming whole-batch rollback after partial decode failure, treating token count as a complete graph-reuse key, or equating posted cancellation with released slots can hide retained state or misdiagnose correct deferred work.",
+ "refusal_point": "Stop before reset or repair if revision, memory type, actual backend, sequence ownership or partial progress is unknown; escalate unsupported rollback or unresolved in-flight work."
},
"limitations": [
"Source-derived candidate; native runtime and efficacy validation not run.",
@@ -104,14 +109,25 @@
"provenance_readsets": [
"references/source-notes.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Reconcile per-sequence retained positions with completed ubatches before retrying a failed logical batch.",
+ "Check both graph allow_reuse and input can_reuse gates, then establish copy readiness independently of graph identity.",
+ "Trace cancellation posting, normal task processing, slot release and prompt retention as separate events."
+ ]
},
"applicability": {
- "scope": "Pinned llama.cpp context-memory, graph-reuse and server task-lifecycle diagnosis; concrete model/backend behavior requires native validation."
+ "scope": "Pinned llama.cpp context-memory, graph-reuse and server task-lifecycle diagnosis; concrete model/backend behavior requires native validation.",
+ "prerequisites": "Pinned llama.cpp source or a concrete runtime trace, access to the relevant memory/graph/server ownership evidence and an identified caller recovery policy that remains in force.",
+ "input_contract": "Source/build and model revisions, memory type, actual backend/device and execution mode, context/batch/ubatch parameters, sequence IDs/positions/output flags and return status; graph cases need dispatch evidence, server cases need correlated HTTP/task/slot/cache logs.",
+ "produced_artifact": "An owner/event ledger of ubatch progress, retained memory, graph-gate decisions, input readiness and slot release/retention, with the first supported broken obligation or smallest missing observation.",
+ "failure_modes": "Assuming whole-batch rollback after partial decode failure, treating token count as a complete graph-reuse key, or equating posted cancellation with released slots can hide retained state or misdiagnose correct deferred work.",
+ "refusal_point": "Stop before reset or repair if revision, memory type, actual backend, sequence ownership or partial progress is unknown; escalate unsupported rollback or unresolved in-flight work."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Reconcile per-sequence retained positions with completed ubatches before retrying a failed logical batch.",
+ "Check both graph allow_reuse and input can_reuse gates, then establish copy readiness independently of graph identity.",
+ "Trace cancellation posting, normal task processing, slot release and prompt retention as separate events."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-recurrent-kv-finalization.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-recurrent-kv-finalization.json
index d2a5bba0e..ed4ff460f 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-recurrent-kv-finalization.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-recurrent-kv-finalization.json
@@ -26,8 +26,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/LICENSE",
"src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/SKILL.md",
@@ -48,7 +48,7 @@
"source_files": {
"src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/LICENSE": "cfc7749b96f63bd31c3c42b5c471bf756814053e847c10f3eb003417bc523d30",
"src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/SKILL.md": "a468d75f25e77b96e6fb5376692f7a89c0464cc00d4df035f1dc6cc923f713cc",
- "src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/metadata.json": "db0b4593d7a6fbb898fdf11a1710b529abbade847ea57319843669b375b68fc8",
+ "src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/metadata.json": "dfd605f8f31740565063fad0785dea3e7751b8deea4d43e1a66ca30d92ae901f",
"src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/references/NOTICE.md": "cca794656897ecdc37f983f7f23c4cd12306dbe3c9cd2f54fece6340753ae4cf",
"src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/references/source-identity.json": "5e52db0a0bda4673b9aac46a3fc035b824780def72921fee6796edb2ed9fca65",
"src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/references/source-notes.md": "0956f59cd9a1fac27da53228dd60c51a07b87639a43c09bfab296c4aeeceaab9",
@@ -112,7 +112,12 @@
}
],
"applicability": {
- "scope": "Pinned vLLM-RLT native Ouro synchronous scheduling with host KVMetadata: LAST_EXITED propagation versus SHARED, later depth-specific Torch-reference attention and normal release/wait/free order. No speculation, connector, preemption, prefix sharing, split/packed prefill or graph/static-buffer path qualification."
+ "scope": "Pinned vLLM-RLT native Ouro synchronous scheduling with host KVMetadata: LAST_EXITED propagation versus SHARED, later depth-specific Torch-reference attention and normal release/wait/free order. No speculation, connector, preemption, prefix sharing, split/packed prefill or graph/static-buffer path qualification.",
+ "prerequisites": "The selected native Ouro synchronous path uses host KVMetadata, no speculation/connector/preemption, atomic prefill and disabled graph/static-buffer/packed-prefill paths; changed execution conditions need a new bounded check.",
+ "input_contract": "Repository revision, model/configuration, recurrent-layer set, loop count, cache layout and block size, dtype/device, attention backend and effective flags, plus the traced token/depth/layer and its next consumer.",
+ "produced_artifact": "A bounded token/finalization/consumer/retirement trace separating source call order, observed data and written validity, completion evidence and remaining native model/backend/performance gaps.",
+ "failure_modes": "Substituting mark_finalized for K/V propagation, accepting allocated pages as a complete written prefix, applying LAST_EXITED copy expectations to SHARED, or treating event recording as completion can invalidate the ownership diagnosis.",
+ "refusal_point": "On a failed prerequisite or complete-prefix check, retain the first failing token/depth/layer and trace the missing operation. Stop the conclusion when flags or metadata ownership leave the selected path."
},
"limitations": [
"Source-derived candidate; native CPU/CUDA, model correctness/quality, performance and agent efficacy are not validated.",
@@ -123,14 +128,25 @@
"references/source-notes.md",
"references/source-identity.json"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Distinguish copied K/V values, logical written positions and allocated capacity; follow the later depth-specific attention consumer.",
+ "Preserve executed shallower values during LAST_EXITED propagation and use SHARED as a separate one-plane control.",
+ "Follow the latest request event through runner release before scheduler cache release; page reuse is not evidence of erased old bytes."
+ ]
},
"applicability": {
- "scope": "Pinned vLLM-RLT native Ouro synchronous scheduling with host KVMetadata: LAST_EXITED propagation versus SHARED, later depth-specific Torch-reference attention and normal release/wait/free order. No speculation, connector, preemption, prefix sharing, split/packed prefill or graph/static-buffer path qualification."
+ "scope": "Pinned vLLM-RLT native Ouro synchronous scheduling with host KVMetadata: LAST_EXITED propagation versus SHARED, later depth-specific Torch-reference attention and normal release/wait/free order. No speculation, connector, preemption, prefix sharing, split/packed prefill or graph/static-buffer path qualification.",
+ "prerequisites": "The selected native Ouro synchronous path uses host KVMetadata, no speculation/connector/preemption, atomic prefill and disabled graph/static-buffer/packed-prefill paths; changed execution conditions need a new bounded check.",
+ "input_contract": "Repository revision, model/configuration, recurrent-layer set, loop count, cache layout and block size, dtype/device, attention backend and effective flags, plus the traced token/depth/layer and its next consumer.",
+ "produced_artifact": "A bounded token/finalization/consumer/retirement trace separating source call order, observed data and written validity, completion evidence and remaining native model/backend/performance gaps.",
+ "failure_modes": "Substituting mark_finalized for K/V propagation, accepting allocated pages as a complete written prefix, applying LAST_EXITED copy expectations to SHARED, or treating event recording as completion can invalidate the ownership diagnosis.",
+ "refusal_point": "On a failed prerequisite or complete-prefix check, retain the first failing token/depth/layer and trace the missing operation. Stop the conclusion when flags or metadata ownership leave the selected path."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Distinguish copied K/V values, logical written positions and allocated capacity; follow the later depth-specific attention consumer.",
+ "Preserve executed shallower values during LAST_EXITED propagation and use SHARED as a separate one-plane control.",
+ "Follow the latest request event through runner release before scheduler cache release; page reuse is not evidence of erased old bytes."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sd-cpp-residency.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sd-cpp-residency.json
index d08ff234c..6bc677079 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sd-cpp-residency.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sd-cpp-residency.json
@@ -25,8 +25,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/LICENSE",
"src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/SKILL.md",
@@ -49,7 +49,7 @@
"source_files": {
"src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/LICENSE": "b53fa08f515cb5a6fff7b9fd8fcd0961a4b80df29d74372d25e1f9171aa042ee",
"src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/SKILL.md": "4c04d39812fb566e3ecb5807c170d5c32848eb2b93a1f0fedcf66ba9a9c92860",
- "src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/metadata.json": "13ab6d874c4e3135be80ef7c3cb6bfd28fb9121012e0536831ec3705ccb8c5eb",
+ "src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/metadata.json": "412c649b16526038ee0299f8b2a610bb7f120eaab25cd83575809fe03b604b2e",
"src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/references/loader-storage-boundaries.md": "c1ec047aa37fca6b374f9bfd750e8f30d45af1d48fe728db23873c0b84ad94e8",
"src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/references/phase-residency.md": "7b537cb972347dd68e303960f142892346d8b19b884e4d08016ea36a23bdaa7a",
"src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/references/source-identity.json": "af4830370b6eddc2d176a9ae14da4ab56a90b0fe048a01a89ee051df4365d8e9",
@@ -117,7 +117,12 @@
}
],
"applicability": {
- "scope": "Source-grounded audit of native image-generation weight/runtime residency, optional segment prefetch and bounded non-ZIP safetensors catalog/read/staging ownership; confirm the actual source, GGML, backend, model and executing path independently."
+ "scope": "Source-grounded audit of native image-generation weight/runtime residency, optional segment prefetch and bounded non-ZIP safetensors catalog/read/staging ownership; confirm the actual source, GGML, backend, model and executing path independently.",
+ "prerequisites": "Access to the pinned source/storage boundaries and evidence of the actual image runner/backend dispatch; native capacity or prefetch checks require matching GGML, model and backend resources.",
+ "input_contract": "Runtime/GGML revisions and selection, compiled/loaded backends and devices, model/components and numerical image settings, effective memory/prefetch/placement settings and actual dispatch; retain available phase allocations, cache/workspace, transfer events and completed-image counts.",
+ "produced_artifact": "A phase-by-runner/backend/device ownership table for stored and resident weights, retained runtime, pending allocations, pin/prefetch ownership and release/eviction, with one discriminating check and its actual status.",
+ "failure_modes": "Treating max_vram as proof of segmentation, unpin as physical freeing, or configured prefetch as completed useful overlap can misidentify the capacity cause; missing observations cannot be read as zero.",
+ "refusal_point": "Stop when the cause is resolved or matching GGML/model/backend resources are unavailable; invoking this audit does not authorize dependency installation, model downloads, policy changes or device provisioning."
},
"limitations": [
"Source-derived candidate; native inference, allocator behavior, image quality and performance were not run.",
@@ -132,14 +137,25 @@
"references/source-notes.md",
"references/loader-storage-boundaries.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Separate parameter storage, compute-resident weights, retained workspace/output caches and pending allocations for each runner and phase.",
+ "Verify the actual monolithic, segmented or multi-device path before interpreting a memory budget or prefetch setting.",
+ "For a policy comparison, hold model, inputs, numerical settings and device fixed; check completed images and quality before latency, and separate startup, first and warm requests."
+ ]
},
"applicability": {
- "scope": "Source-grounded audit of native image-generation weight/runtime residency, optional segment prefetch and bounded non-ZIP safetensors catalog/read/staging ownership; confirm the actual source, GGML, backend, model and executing path independently."
+ "scope": "Source-grounded audit of native image-generation weight/runtime residency, optional segment prefetch and bounded non-ZIP safetensors catalog/read/staging ownership; confirm the actual source, GGML, backend, model and executing path independently.",
+ "prerequisites": "Access to the pinned source/storage boundaries and evidence of the actual image runner/backend dispatch; native capacity or prefetch checks require matching GGML, model and backend resources.",
+ "input_contract": "Runtime/GGML revisions and selection, compiled/loaded backends and devices, model/components and numerical image settings, effective memory/prefetch/placement settings and actual dispatch; retain available phase allocations, cache/workspace, transfer events and completed-image counts.",
+ "produced_artifact": "A phase-by-runner/backend/device ownership table for stored and resident weights, retained runtime, pending allocations, pin/prefetch ownership and release/eviction, with one discriminating check and its actual status.",
+ "failure_modes": "Treating max_vram as proof of segmentation, unpin as physical freeing, or configured prefetch as completed useful overlap can misidentify the capacity cause; missing observations cannot be read as zero.",
+ "refusal_point": "Stop when the cause is resolved or matching GGML/model/backend resources are unavailable; invoking this audit does not authorize dependency installation, model downloads, policy changes or device provisioning."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Separate parameter storage, compute-resident weights, retained workspace/output caches and pending allocations for each runner and phase.",
+ "Verify the actual monolithic, segmented or multi-device path before interpreting a memory budget or prefetch setting.",
+ "For a policy comparison, hold model, inputs, numerical settings and device fixed; check completed images and quality before latency, and separate startup, first and warm requests."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sglang-diffusion-lifetimes.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sglang-diffusion-lifetimes.json
index fcea3403c..6a2493a65 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sglang-diffusion-lifetimes.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sglang-diffusion-lifetimes.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/LICENSE",
"src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/SKILL.md",
@@ -43,7 +43,7 @@
"source_files": {
"src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/LICENSE": "1495e1e757ef4d0925a2350563cf5754bb23c51701a8ec4fb3c5cdcbedae6747",
"src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/SKILL.md": "b0a4ba4751087126143ef86983798deb0ed00e24b6ff06bf6ef76a9df6e274f2",
- "src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/metadata.json": "6ed57dfb4f25ecb34b81f2188b055e05d172ba2581e6a3b605b814b2f733a4d6",
+ "src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/metadata.json": "e2cb137978960d29b3644967a29dab930c768d3453bce8aff2aad96fd90a56d9",
"src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/references/source-notes.md": "d0fcfd1177ae2f351041c5b2abba7f5121cec133006d9e4dca8f74ddb567332d",
"src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
},
@@ -101,7 +101,12 @@
}
],
"applicability": {
- "scope": "Bounded monolithic multimodal_gen composed-pipeline state and lifecycle diagnosis; actual backend/model behavior requires native validation."
+ "scope": "Bounded monolithic multimodal_gen composed-pipeline state and lifecycle diagnosis; actual backend/model behavior requires native validation.",
+ "prerequisites": "Pinned source or a concrete trace of the selected composed-pipeline executor and installed residency manager, with the applicable monolithic route identified.",
+ "input_contract": "Source/build, backend and executor/pipeline class, model revision, modality/shape, steps/scheduler/seed/output count, grouped/serial route, ranks, placement, warmup/save settings and symptom; timeout cases need correlated request/job and client/scheduler/worker events.",
+ "produced_artifact": "A version-bound owner/event ledger, first unresolved boundary, causal hypothesis, smallest real discriminating check, exact results and missing dependencies.",
+ "failure_modes": "Assuming all executors share the SyncExecutor contract, treating closed component use as eviction, assuming grouped schedulers are isolated, or reading socket close/job deletion as stopped device compute can misdiagnose ownership.",
+ "refusal_point": "Stop performance claims when the executing path, cleanup or output quality is unresolved; retain the failure and a known baseline rather than inferring a blanket leak or safety conclusion."
},
"limitations": [
"Source-derived candidate; upstream runtime and efficacy validation not run.",
@@ -111,14 +116,25 @@
"provenance_readsets": [
"references/source-notes.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Trace the actual executor and separate request, stage, component-use, module, scheduler, finalizer and allocator-cache lifetimes.",
+ "Check scheduler aliases and declared output-copy policies before introducing overlap; compare real serial/grouped outputs with fixed timesteps and seeds.",
+ "Correlate worker, finalizer and device completion after client timeout or deletion; preserve an unknown execution outcome until that evidence exists."
+ ]
},
"applicability": {
- "scope": "Bounded monolithic multimodal_gen composed-pipeline state and lifecycle diagnosis; actual backend/model behavior requires native validation."
+ "scope": "Bounded monolithic multimodal_gen composed-pipeline state and lifecycle diagnosis; actual backend/model behavior requires native validation.",
+ "prerequisites": "Pinned source or a concrete trace of the selected composed-pipeline executor and installed residency manager, with the applicable monolithic route identified.",
+ "input_contract": "Source/build, backend and executor/pipeline class, model revision, modality/shape, steps/scheduler/seed/output count, grouped/serial route, ranks, placement, warmup/save settings and symptom; timeout cases need correlated request/job and client/scheduler/worker events.",
+ "produced_artifact": "A version-bound owner/event ledger, first unresolved boundary, causal hypothesis, smallest real discriminating check, exact results and missing dependencies.",
+ "failure_modes": "Assuming all executors share the SyncExecutor contract, treating closed component use as eviction, assuming grouped schedulers are isolated, or reading socket close/job deletion as stopped device compute can misdiagnose ownership.",
+ "refusal_point": "Stop performance claims when the executing path, cleanup or output quality is unresolved; retain the failure and a known baseline rather than inferring a blanket leak or safety conclusion."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Trace the actual executor and separate request, stage, component-use, module, scheduler, finalizer and allocator-cache lifetimes.",
+ "Check scheduler aliases and declared output-copy policies before introducing overlap; compare real serial/grouped outputs with fixed timesteps and seeds.",
+ "Correlate worker, finalizer and device completion after client timeout or deletion; preserve an unknown execution outcome until that evidence exists."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-vllm-kv-lifetimes.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-vllm-kv-lifetimes.json
index ec6f11635..9e9196a20 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-vllm-kv-lifetimes.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-vllm-kv-lifetimes.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/SKILL.md",
"src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/metadata.json",
@@ -39,7 +39,7 @@
"checkout_ref": "checkout:audit-vllm-kv-lifetimes",
"source_files": {
"src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/SKILL.md": "4c3f27459df239f9f4fc971c559d4c9ddfbfebd70507a8a93be8eae59a30b69d",
- "src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/metadata.json": "b43fc3a7ccae59c664fcdf9ff77962edcc07c2a3df8b59324db551642bd23e23",
+ "src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/metadata.json": "8ded6e60c562131616dc8cdcf0d7efc04aef1884885da0b30fb16f42d7433b7c",
"src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/references/source-notes.md": "3665b034594fc8cc767c408002f4e16f8803369242405027c7f37ce6c9ecebe9",
"src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
},
@@ -78,12 +78,33 @@
"provenance_readsets": [
"references/source-notes.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "applicability": {
+ "scope": "Pinned vLLM V1 scheduler/connector ownership audit for asynchronous KV load/save completion, retained blocks, preemption and failed receives; concrete connector conformance is not established by the interface.",
+ "prerequisites": "A concrete connector and worker/model-runner call trace at identified revisions, with request/block events sufficient to compare actual callers against the pinned scheduler/interface contracts.",
+ "input_contract": "Engine/connector revision and class, producer/consumer roles, KV-group count/block namespace, failure policy, concurrent-batch and graph modes; correlate bounded request IDs with allocation, matched tokens, load/save events, receive failures, preemption, finish and pool return.",
+ "produced_artifact": "An ownership/event ledger and bounded contract-conformance report naming the earliest evidenced broken obligation, owner, affected requests/blocks and smallest missing observation, with a proposed targeted regression check.",
+ "failure_modes": "Equating reserved tokens with usable KV, finished responses with reusable blocks, or send completion with scheduler write-fence completion can hide unsafe reuse; missing failed-receive completion can leave waits unreleased.",
+ "refusal_point": "Escalate absent concrete caller evidence, ambiguous block namespaces, unsupported layouts or revision drift; stop before inferring a backend fix, freeing retained blocks or claiming a reproduced bug."
+ },
+ "recommendations": [
+ "Trace lookup, allocation, worker execution and completion propagation before changing request state; repeated or undecided lookup and ordinary admission waits can be valid.",
+ "Check delayed-free ownership and scheduler in-flight-write fences separately, including empty-step save/metadata ordering in the real worker.",
+ "Keep receive failure and finished_recving correlated, and use request-level failure evidence for multi-group layouts while preserving the existing recovery policy."
+ ]
+ },
+ "applicability": {
+ "scope": "Pinned vLLM V1 scheduler/connector ownership audit for asynchronous KV load/save completion, retained blocks, preemption and failed receives; concrete connector conformance is not established by the interface.",
+ "prerequisites": "A concrete connector and worker/model-runner call trace at identified revisions, with request/block events sufficient to compare actual callers against the pinned scheduler/interface contracts.",
+ "input_contract": "Engine/connector revision and class, producer/consumer roles, KV-group count/block namespace, failure policy, concurrent-batch and graph modes; correlate bounded request IDs with allocation, matched tokens, load/save events, receive failures, preemption, finish and pool return.",
+ "produced_artifact": "An ownership/event ledger and bounded contract-conformance report naming the earliest evidenced broken obligation, owner, affected requests/blocks and smallest missing observation, with a proposed targeted regression check.",
+ "failure_modes": "Equating reserved tokens with usable KV, finished responses with reusable blocks, or send completion with scheduler write-fence completion can hide unsafe reuse; missing failed-receive completion can leave waits unreleased.",
+ "refusal_point": "Escalate absent concrete caller evidence, ambiguous block namespaces, unsupported layouts or revision drift; stop before inferring a backend fix, freeing retained blocks or claiming a reproduced bug."
},
- "applicability": {},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Trace lookup, allocation, worker execution and completion propagation before changing request state; repeated or undecided lookup and ordinary admission waits can be valid.",
+ "Check delayed-free ownership and scheduler in-flight-write fences separately, including empty-step save/metadata ordering in the real worker.",
+ "Keep receive failure and finished_recving correlated, and use request-level failure evidence for multi-group layouts while preserving the existing recovery policy."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/bound-windowed-attention-replay.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/bound-windowed-attention-replay.json
index 16eb129f7..138ada970 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/bound-windowed-attention-replay.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/bound-windowed-attention-replay.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/SKILL.md",
"src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/metadata.json",
@@ -42,7 +42,7 @@
"checkout_ref": "checkout:bound-windowed-attention-replay",
"source_files": {
"src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/SKILL.md": "4703beb6e5194ef2ff2c0b5ec69f64e35417a4bc0dfb50a10c80a920b77ef3b5",
- "src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/metadata.json": "014be24d0e669f4acbc0717dc323eec784026cf185186619d5a19d6d84496131",
+ "src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/metadata.json": "c68254b1a96a8834ae46c7a1d9035856fa4783315b219aa95e31461bd8ac58c9",
"src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/references/source-notes.md": "fd917bd5ac80bee9c9f4525f8e0300f6d67dc144e2a16632af42a67f3d53d098",
"src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/references/validation.md": "60e71e70c5519bcdf6c171eec2be2f342e8801bedd746d6005f11adc646dfaff",
"src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
@@ -90,7 +90,11 @@
"applicability": {
"scope": "Decoders with an explicit producer/consumer boundary and a demonstrated dependency closure for compact query rows, including separately preserved historical state",
"prerequisites": "A derived dependency closure; a full-length KV write at the boundary; known rollback depth; control over metadata construction",
- "hardware": "Backend-independent retain arithmetic; window sizes, boundary index and ring lengths come from the model config and do not transfer"
+ "hardware": "Backend-independent retain arithmetic; window sizes, boundary index and ring lengths come from the model config and do not transfer",
+ "input_contract": "The actual producer/consumer split and transitive dependency closure, boundary index, window and rollback depth, consumer KV sharing/compression layout and separately preserved historical keys.",
+ "produced_artifact": "A dependency listing, enforced retain length, compact query/position/store-slot/visibility metadata, a separate historical-key retention statement and a diagnostic equivalence-reference mode.",
+ "failure_modes": "Applying window-plus-rollback without dependency closure, discarding full-layout state, clipping away the sampling row, or comparing across preemption can invalidate replay; compact queries do not imply historical keys were dropped.",
+ "refusal_point": "Stop if required consumer state is unavailable without a preserved full-length copy, the boundary representation cannot be captured, or comparison preemption moves the anchor. Do not use for clipped-region outputs or required bit-identical preemption replay."
},
"limitations": [
"Source-derived candidate; no equivalence run was performed.",
@@ -104,16 +108,26 @@
"references/source-notes.md",
"references/validation.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Derive every consumer dependency before computing keep; preserve the required full-layout representation separately from the compact query rows.",
+ "Use the visibility-only diagnostic mode to isolate equivalence from row gathering, and retain one sampling row for a fully clipped chunk.",
+ "Check metadata at chunk/store boundaries, actual prompt-logprob handling and zero preemption during equivalence comparison."
+ ]
},
"applicability": {
"scope": "Decoders with an explicit producer/consumer boundary and a demonstrated dependency closure for compact query rows, including separately preserved historical state",
"prerequisites": "A derived dependency closure; a full-length KV write at the boundary; known rollback depth; control over metadata construction",
- "hardware": "Backend-independent retain arithmetic; window sizes, boundary index and ring lengths come from the model config and do not transfer"
+ "hardware": "Backend-independent retain arithmetic; window sizes, boundary index and ring lengths come from the model config and do not transfer",
+ "input_contract": "The actual producer/consumer split and transitive dependency closure, boundary index, window and rollback depth, consumer KV sharing/compression layout and separately preserved historical keys.",
+ "produced_artifact": "A dependency listing, enforced retain length, compact query/position/store-slot/visibility metadata, a separate historical-key retention statement and a diagnostic equivalence-reference mode.",
+ "failure_modes": "Applying window-plus-rollback without dependency closure, discarding full-layout state, clipping away the sampling row, or comparing across preemption can invalidate replay; compact queries do not imply historical keys were dropped.",
+ "refusal_point": "Stop if required consumer state is unavailable without a preserved full-length copy, the boundary representation cannot be captured, or comparison preemption moves the anchor. Do not use for clipped-region outputs or required bit-identical preemption replay."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Derive every consumer dependency before computing keep; preserve the required full-layout representation separately from the compact query rows.",
+ "Use the visibility-only diagnostic mode to isolate equivalence from row gathering, and retain one sampling row for a fully clipped chunk.",
+ "Check metadata at chunk/store boundaries, actual prompt-logprob handling and zero preemption during equivalence comparison."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/budget-speculative-verification.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/budget-speculative-verification.json
index 63ea52616..c88908d7a 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/budget-speculative-verification.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/budget-speculative-verification.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/SKILL.md",
"src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/metadata.json",
@@ -42,7 +42,7 @@
"checkout_ref": "checkout:budget-speculative-verification",
"source_files": {
"src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/SKILL.md": "f157562d1edc905cb4fc07246567e3c5b844efb9fbf5fe39f0d40553a1b1a307",
- "src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/metadata.json": "4d68bf4c1767eaee12ade963bbd551d0e61d1bcd81a984bb02fd3aae46965919",
+ "src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/metadata.json": "1aa80a3109072c1ec211f38dbeb9e9cb9edb56aeac187946b43e65dd7502e46f",
"src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/references/source-notes.md": "6fa955022f16cd598f44fd6521901d1cb868b06cec545cabe2b82769a641b5d6",
"src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/references/validation.md": "673d93f6700052974b340fab25b3438e0e6344960569ac9abec7d7720f662288",
"src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
@@ -88,7 +88,11 @@
"applicability": {
"scope": "Engines with multi-token speculative decoding whose verification forward cost grows with query rows",
"prerequisites": "A calibrated per-position acceptance signal; a measured (rows, step_time) curve; a handle on metadata construction",
- "hardware": "Backend-independent procedure; the reference implementation targets NVIDIA SM100 and Ascend 950 only"
+ "hardware": "Backend-independent procedure; the reference implementation targets NVIDIA SM100 and Ascend 950 only",
+ "input_contract": "Draft width and row representation, the defined individual/conditional versus cumulative acceptance semantics, verify rows/time curve, sampler cost and acceptance round actually available at schedule time.",
+ "produced_artifact": "A per-step verification row budget, per-request query lengths from one through draft_width plus one, cumulative acceptance vectors and predicted-versus-realized acceptance error logs.",
+ "failure_modes": "Confusing verification savings with reduced drafting, misreading confidence semantics or available rounds, allowing later-position confidence to select an earlier token, or generating non-prefix grants can invalidate the budget or sampling claim.",
+ "refusal_point": "Stop if unthrottled drafting or sampler cost dominates, the signal cannot be made cumulative, or no round supplies acceptance estimates; the procedure also does not apply without a per-step cost model."
},
"limitations": [
"Source-derived candidate; no local efficacy validation and no reproduced gain.",
@@ -103,16 +107,26 @@
"references/source-notes.md",
"references/validation.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Measure draft, verify and sampler costs separately and fit the verify curve for graph sizes that will actually replay.",
+ "Preserve the one-row floor and prefix grants, and audit CPU/GPU allocation timing against the per-position decision-independence contract.",
+ "Report verification rows, accepted tokens, total step time and throughput separately; use an independently specified sampling oracle rather than requiring unchanged accepted length."
+ ]
},
"applicability": {
"scope": "Engines with multi-token speculative decoding whose verification forward cost grows with query rows",
"prerequisites": "A calibrated per-position acceptance signal; a measured (rows, step_time) curve; a handle on metadata construction",
- "hardware": "Backend-independent procedure; the reference implementation targets NVIDIA SM100 and Ascend 950 only"
+ "hardware": "Backend-independent procedure; the reference implementation targets NVIDIA SM100 and Ascend 950 only",
+ "input_contract": "Draft width and row representation, the defined individual/conditional versus cumulative acceptance semantics, verify rows/time curve, sampler cost and acceptance round actually available at schedule time.",
+ "produced_artifact": "A per-step verification row budget, per-request query lengths from one through draft_width plus one, cumulative acceptance vectors and predicted-versus-realized acceptance error logs.",
+ "failure_modes": "Confusing verification savings with reduced drafting, misreading confidence semantics or available rounds, allowing later-position confidence to select an earlier token, or generating non-prefix grants can invalidate the budget or sampling claim.",
+ "refusal_point": "Stop if unthrottled drafting or sampler cost dominates, the signal cannot be made cumulative, or no round supplies acceptance estimates; the procedure also does not apply without a per-step cost model."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Measure draft, verify and sampler costs separately and fit the verify curve for graph sizes that will actually replay.",
+ "Preserve the one-row floor and prefix grants, and audit CPU/GPU allocation timing against the per-position decision-independence contract.",
+ "Report verification rows, accepted tokens, total step time and throughput separately; use an independently specified sampling oracle rather than requiring unchanged accepted length."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/calibrate-serving-simulation.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/calibrate-serving-simulation.json
index 1fb80053f..dfa546a59 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/calibrate-serving-simulation.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/calibrate-serving-simulation.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/methods/calibrate-serving-simulation/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/methods/calibrate-serving-simulation/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/methods/calibrate-serving-simulation/SKILL.md",
"src/inferrsi/data/skill_library/methods/calibrate-serving-simulation/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cgroup-resource-controls.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cgroup-resource-controls.json
index 0361b093d..291752a82 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cgroup-resource-controls.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cgroup-resource-controls.json
@@ -25,8 +25,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/hardware/host/cgroup-resource-controls/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/hardware/host/cgroup-resource-controls/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/hardware/host/cgroup-resource-controls/SKILL.md",
"src/inferrsi/data/skill_library/hardware/host/cgroup-resource-controls/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/checkpoint-recovery-consistency.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/checkpoint-recovery-consistency.json
index 5a7d2cc47..2cb453fde 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/checkpoint-recovery-consistency.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/checkpoint-recovery-consistency.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/methods/checkpoint-recovery-consistency/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/methods/checkpoint-recovery-consistency/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/methods/checkpoint-recovery-consistency/SKILL.md",
"src/inferrsi/data/skill_library/methods/checkpoint-recovery-consistency/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/choose-parallel-training-plan.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/choose-parallel-training-plan.json
index 15dda5be9..cfc1ead65 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/choose-parallel-training-plan.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/choose-parallel-training-plan.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/methods/choose-parallel-training-plan/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/methods/choose-parallel-training-plan/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/methods/choose-parallel-training-plan/SKILL.md",
"src/inferrsi/data/skill_library/methods/choose-parallel-training-plan/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cite-check-downstream-pr.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cite-check-downstream-pr.json
index d61046382..ef2f13585 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cite-check-downstream-pr.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cite-check-downstream-pr.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/workflows/cite-check-downstream-pr/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/workflows/cite-check-downstream-pr/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/skill-index.json",
"src/inferrsi/data/skill_library/workflows/cite-check-downstream-pr/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/claim-contract.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/claim-contract.json
index e1ef97ce5..94270e56f 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/claim-contract.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/claim-contract.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/workflows/claim-contract/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/workflows/claim-contract/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/skill-index.json",
"src/inferrsi/data/skill_library/workflows/claim-contract/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/comfy-ui.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/comfy-ui.json
index e323beb0a..f2574de51 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/comfy-ui.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/comfy-ui.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/repositories/comfy-ui/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/repositories/comfy-ui/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/repositories/comfy-ui/LICENSE",
"src/inferrsi/data/skill_library/repositories/comfy-ui/SKILL.md",
@@ -45,7 +45,7 @@
"source_files": {
"src/inferrsi/data/skill_library/repositories/comfy-ui/LICENSE": "3972dc9744f6499f0f9b2dbf76696f2ae7ad8af9b23dde66d6af86c9dfb36986",
"src/inferrsi/data/skill_library/repositories/comfy-ui/SKILL.md": "c993bd6ff8e9b5f82f39f175ec8242341136e4ee0808af414e1d94ad3e196c5e",
- "src/inferrsi/data/skill_library/repositories/comfy-ui/metadata.json": "74742851a002a51c208362cdcd2a59fdab5d1734d8492b1f93fbec3cf21e0841",
+ "src/inferrsi/data/skill_library/repositories/comfy-ui/metadata.json": "81ad1c73b12426e4bf7961889f4b9a1c92739c6be18380402f7866e021cac67b",
"src/inferrsi/data/skill_library/repositories/comfy-ui/references/graph-cache-invalidation.md": "614c9a4e6943d24d7d7be7d632682ce8b1dcc239124a37d9dc1d487113772eac",
"src/inferrsi/data/skill_library/repositories/comfy-ui/references/model-residency.md": "014c68386e7cdfdb7b4f566836e8b2ff61bd5a5a506e7b4723f83fe80c5d4de7",
"src/inferrsi/data/skill_library/repositories/comfy-ui/references/repo-provenance.md": "ca996ef515af42a3ce304747fa3f12a31c00533deaace5ba2ce21aefa0a62e2e",
@@ -104,7 +104,12 @@
}
],
"applicability": {
- "scope": "Source-grounded ComfyUI workflow-output cache and model-weight ownership audit; confirm the actual revision, node registry, cache caller and static/dynamic patcher before transfer."
+ "scope": "Source-grounded ComfyUI workflow-output cache and model-weight ownership audit; confirm the actual revision, node registry, cache caller and static/dynamic patcher before transfer.",
+ "prerequisites": "Access to the pinned provenance and the actual node registry/cache caller; residency questions additionally require the concrete patcher and effective dynamic/static mode. Native memory conclusions need the actual backend.",
+ "input_contract": "Exact revision, API-format prompt/output targets, relevant node classes/constants/links/hidden inputs and external state, selected cache caller/mode and patcher settings, plus available execution/load/memory/timing observations.",
+ "produced_artifact": "A state/owner/key/lifetime/consumer ledger with invalidation or release events, evidence and unresolved boundaries, followed by a discriminating check and its actual result.",
+ "failure_modes": "Treating identical JSON as proof of unchanged external inputs, confusing eviction with key validity, or equating workflow cache hits with device-resident model weights can explain the wrong state contract.",
+ "refusal_point": "Stop when the cause/action is supported or required environment, backend or model resources are unavailable; do not manufacture a hit, hardware result or performance threshold."
},
"limitations": [
"Source-derived candidate; no local executor-cache, device-residency or performance qualification.",
@@ -114,14 +119,25 @@
"provenance_readsets": [
"references/repo-provenance.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Draw the affected dependency cone with declared, lazy and dynamically expanded links, then mark unfingerprinted external state unresolved.",
+ "Separate output/object cache validity and retention from patch identity, partial unload, full detach and allocator-visible release.",
+ "Change one test variable while fixing prompt, output target, model, backend, precision, seed and cache mode; check values and executed nodes before timing."
+ ]
},
"applicability": {
- "scope": "Source-grounded ComfyUI workflow-output cache and model-weight ownership audit; confirm the actual revision, node registry, cache caller and static/dynamic patcher before transfer."
+ "scope": "Source-grounded ComfyUI workflow-output cache and model-weight ownership audit; confirm the actual revision, node registry, cache caller and static/dynamic patcher before transfer.",
+ "prerequisites": "Access to the pinned provenance and the actual node registry/cache caller; residency questions additionally require the concrete patcher and effective dynamic/static mode. Native memory conclusions need the actual backend.",
+ "input_contract": "Exact revision, API-format prompt/output targets, relevant node classes/constants/links/hidden inputs and external state, selected cache caller/mode and patcher settings, plus available execution/load/memory/timing observations.",
+ "produced_artifact": "A state/owner/key/lifetime/consumer ledger with invalidation or release events, evidence and unresolved boundaries, followed by a discriminating check and its actual result.",
+ "failure_modes": "Treating identical JSON as proof of unchanged external inputs, confusing eviction with key validity, or equating workflow cache hits with device-resident model weights can explain the wrong state contract.",
+ "refusal_point": "Stop when the cause/action is supported or required environment, backend or model resources are unavailable; do not manufacture a hit, hardware result or performance threshold."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Draw the affected dependency cone with declared, lazy and dynamically expanded links, then mark unfingerprinted external state unresolved.",
+ "Separate output/object cache validity and retention from patch identity, partial unload, full detach and allocator-visible release.",
+ "Change one test variable while fixing prompt, output target, model, backend, precision, seed and cache mode; check values and executed nodes before timing."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/competitor-scan.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/competitor-scan.json
index a4a5de8a4..06d75ab0e 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/competitor-scan.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/competitor-scan.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/workflows/competitor-scan/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/workflows/competitor-scan/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/skill-index.json",
"src/inferrsi/data/skill_library/workflows/competitor-scan/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dcp-async-save-performance.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dcp-async-save-performance.json
index 5feddf468..357849cc7 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dcp-async-save-performance.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dcp-async-save-performance.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/SKILL.md",
"src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/metadata.json",
@@ -41,7 +41,7 @@
"checkout_ref": "checkout:dcp-async-save-performance",
"source_files": {
"src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/SKILL.md": "645139ee0c88ec865cad952e24cda35f4648d4b777b732c9b7d274b6c7830845",
- "src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/metadata.json": "db7ad368394ccbfdeff6a16e3423c300d51305b1f95ff5c3936df91738a8cc35",
+ "src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/metadata.json": "49764fa454403fb38891d82ee63de1e07a90d3ff6a3cd28550aee7ea34cbdcb0",
"src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/references/sources.md": "22afe4f4d77f441096cc657d55d763473bfcbea8fb531c8e493fe49ab84955f7",
"src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
},
@@ -71,12 +71,33 @@
"provenance_readsets": [
"references/sources.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "applicability": {
+ "scope": "Source-supported diagnosis of training-visible PyTorch DCP save pauses, snapshot/staging/upload lifetimes and bounded CPU staging-buffer reuse; distributed recovery validity and backend durability need separate evidence.",
+ "prerequisites": "Known correspondence between the installed PyTorch API and source/version notes, identified checkpointed-state mutation boundaries and an initial one-outstanding-save policy before overlap is considered.",
+ "input_contract": "Installed PyTorch revision, stager, storage writer and thread/process executor, checkpoint contents, ranks per host and state-mutation boundaries; performance assessment needs caller/staging/upload waits and host/resource observations.",
+ "produced_artifact": "A completion/lifetime diagnosis and smallest justified staging/cache/backpressure lever, with separate caller-pause, staging, upload and step-time measurements proposed for authorized validation.",
+ "failure_modes": "Mutating source state before staging completes, overwriting cached snapshots before prior upload completes, ignoring caller-side state_dict work or treating async return as durability can corrupt the snapshot or misstate the pause.",
+ "refusal_point": "Reject staging overlap that races checkpointed-state mutation; do not infer backend durability from asynchronous return or writer completion."
+ },
+ "recommendations": [
+ "Place the staging wait before the earliest mutation of any checkpointed state, including forward-mutated buffers rather than only optimizer.step.",
+ "Wait for the previous upload before reusing a stager's CPU storage; observe the final upload before closing caller-owned staging resources.",
+ "Measure checkpoint and intervening steps with the same state/cadence, retaining first-use and cached-buffer costs plus per-host memory and I/O pressure."
+ ]
+ },
+ "applicability": {
+ "scope": "Source-supported diagnosis of training-visible PyTorch DCP save pauses, snapshot/staging/upload lifetimes and bounded CPU staging-buffer reuse; distributed recovery validity and backend durability need separate evidence.",
+ "prerequisites": "Known correspondence between the installed PyTorch API and source/version notes, identified checkpointed-state mutation boundaries and an initial one-outstanding-save policy before overlap is considered.",
+ "input_contract": "Installed PyTorch revision, stager, storage writer and thread/process executor, checkpoint contents, ranks per host and state-mutation boundaries; performance assessment needs caller/staging/upload waits and host/resource observations.",
+ "produced_artifact": "A completion/lifetime diagnosis and smallest justified staging/cache/backpressure lever, with separate caller-pause, staging, upload and step-time measurements proposed for authorized validation.",
+ "failure_modes": "Mutating source state before staging completes, overwriting cached snapshots before prior upload completes, ignoring caller-side state_dict work or treating async return as durability can corrupt the snapshot or misstate the pause.",
+ "refusal_point": "Reject staging overlap that races checkpointed-state mutation; do not infer backend durability from asynchronous return or writer completion."
},
- "applicability": {},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Place the staging wait before the earliest mutation of any checkpointed state, including forward-mutated buffers rather than only optimizer.step.",
+ "Wait for the previous upload before reusing a stager's CPU storage; observe the final upload before closing caller-owned staging resources.",
+ "Measure checkpoint and intervening steps with the same state/cadence, retaining first-use and cached-buffer costs plus per-host memory and I/O pressure."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dense-performance-evidence.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dense-performance-evidence.json
index 3a549ce9a..4faa1834e 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dense-performance-evidence.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dense-performance-evidence.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/methods/dense-performance-evidence/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/methods/dense-performance-evidence/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/methods/dense-performance-evidence/SKILL.md",
"src/inferrsi/data/skill_library/methods/dense-performance-evidence/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diagnose-vllm-admission-pressure.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diagnose-vllm-admission-pressure.json
index fdd5f275f..9cfacf268 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diagnose-vllm-admission-pressure.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diagnose-vllm-admission-pressure.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/SKILL.md",
"src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/metadata.json",
@@ -52,7 +52,7 @@
"checkout_ref": "checkout:diagnose-vllm-admission-pressure",
"source_files": {
"src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/SKILL.md": "d99fc28d48127d3bc627f2a0cab6f3f16158a77da49428838a08865adaea9905",
- "src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/metadata.json": "78459416734f61a8faa819d9a1a3ad7b9a985e8bf8e149813c604d15be6bb92f",
+ "src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/metadata.json": "a4a4c4b7645852bce4abafde8e4128baacbdf1bc2a69362f7c3120db597fbe40",
"src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/references/papers-20261015/INTEGRATION_NOTES.md": "82b24be82451ee121892f61575d0cf4ecb9920813a25ccb26cce2b45dda6d8b3",
"src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/references/papers-20261015/cases/migration-is-not-free-memory.md": "2ac168f2725cba7fc92dfcc67c01bf2258dff6a2dbe1ddb063bc60c175d82d0a",
"src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/references/papers-20261015/cases/parallelism-change-with-state-cost.md": "5accc81086e4500d66f894ef6af9951e1cbb76941f6c631aab5c5561a786d8d8",
@@ -106,12 +106,33 @@
"references/papers-20261015/references/orca-readset.md",
"references/source-notes.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "applicability": {
+ "scope": "Pinned vLLM V1 branch-based diagnosis of queued requests, low scheduled-token counts and KV-pressure preemption, separating admission/token/input/allocation/connector-wait constraints.",
+ "prerequisites": "The deployed scheduler must be mapped to the pinned branches, with effective configuration and a bounded request/step trace; another revision requires re-establishing that correspondence.",
+ "input_contract": "Deployed revision, scheduler class/configuration, connector role, KV layout, policy and speculative/encoder/concurrent-batch features; per-step queues, tokens, both budgets, allocation results, victims, transfer/free state and timing.",
+ "produced_artifact": "A per-step accounting ledger, observed limiting branch and competing explanation, missing instrumentation and one conditional configuration/progress hypothesis with predicted effects and trade-offs.",
+ "failure_modes": "Inferring memory exhaustion from a queue or high KV usage, conflating token/input budgets, ignoring pending frees or reading async token accounting as load completion can attribute admission stalls to the wrong branch.",
+ "refusal_point": "When required state is missing, leave the diagnosis unresolved and request the smallest trace; escalate allocation or connector behavior outside the readset and stop before live mutation or performance claims."
+ },
+ "recommendations": [
+ "Trace the first actual blocking or skip branch and record running versus admission caps and token versus input budgets separately.",
+ "Inspect deferred frees and request-free eligibility before blaming preemption victim selection, then reconcile restored budgets for an already-scheduled victim.",
+ "Reconstruct waiting admission independently, including queue precedence and async prefix states; change no setting without a branch-specific hypothesis."
+ ]
+ },
+ "applicability": {
+ "scope": "Pinned vLLM V1 branch-based diagnosis of queued requests, low scheduled-token counts and KV-pressure preemption, separating admission/token/input/allocation/connector-wait constraints.",
+ "prerequisites": "The deployed scheduler must be mapped to the pinned branches, with effective configuration and a bounded request/step trace; another revision requires re-establishing that correspondence.",
+ "input_contract": "Deployed revision, scheduler class/configuration, connector role, KV layout, policy and speculative/encoder/concurrent-batch features; per-step queues, tokens, both budgets, allocation results, victims, transfer/free state and timing.",
+ "produced_artifact": "A per-step accounting ledger, observed limiting branch and competing explanation, missing instrumentation and one conditional configuration/progress hypothesis with predicted effects and trade-offs.",
+ "failure_modes": "Inferring memory exhaustion from a queue or high KV usage, conflating token/input budgets, ignoring pending frees or reading async token accounting as load completion can attribute admission stalls to the wrong branch.",
+ "refusal_point": "When required state is missing, leave the diagnosis unresolved and request the smallest trace; escalate allocation or connector behavior outside the readset and stop before live mutation or performance claims."
},
- "applicability": {},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Trace the first actual blocking or skip branch and record running versus admission caps and token versus input budgets separately.",
+ "Inspect deferred frees and request-free eligibility before blaming preemption victim selection, then reconcile restored budgets for an already-scheduled victim.",
+ "Reconstruct waiting admission independently, including queue precedence and async prefix states; change no setting without a branch-specific hypothesis."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diffusers-invocation-state.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diffusers-invocation-state.json
index 1109b3b5a..4edb012d7 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diffusers-invocation-state.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diffusers-invocation-state.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/SKILL.md",
"src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/metadata.json",
@@ -42,7 +42,7 @@
"checkout_ref": "checkout:diffusers-invocation-state",
"source_files": {
"src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/SKILL.md": "7581b56336d07213a87d01477793ae1a8acf0b7cf876fce36c9a761ee32c1ff4",
- "src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/metadata.json": "68ac7f64b7ab947fc2db664bc3990f102c1cc5539b42350aa5818d0bbf17d86c",
+ "src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/metadata.json": "66f8d20c1644aba964adfff3b439bfd3ecd3875885191e6ce10721bc50877b1a",
"src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/references/ownership.md": "40688cc44183027efbf2fca8d3548c299cdf2e33fef54f76ab3549353eb9bd1b",
"src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/references/source-map.md": "10562a75683bd9bd07ec2e37a1cbd0f93f2534d6f85b55a566adb2d97485215f",
"src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/references/validation.md": "5a0b854b87d091d7d505b0bdd5c03330052507a851ab6edf35b93a6675f2d4f7",
@@ -84,7 +84,12 @@
}
],
"applicability": {
- "scope": "FluxPipeline and FlowMatchEulerDiscreteScheduler invocation ownership at this source pin; inspect the actual caller contract."
+ "scope": "FluxPipeline and FlowMatchEulerDiscreteScheduler invocation ownership at this source pin; inspect the actual caller contract.",
+ "prerequisites": "Access to the pinned ownership path and an identified caller admission/serialization contract; other pipeline or scheduler classes need their own contracts.",
+ "input_contract": "Diffusers/PyTorch revisions, pipeline/scheduler classes and object identities, construction/from_pipe path, actual components, scheduler config and resolved timesteps/sigmas, shape/dtype, callback/interrupt timing, output/cleanup path and overlapping intervals.",
+ "produced_artifact": "An object-identity/ownership map and call-to-schedule-reset/step/interrupt/output/cleanup trace, the caller admission contract, one next check and actual evidence status.",
+ "failure_modes": "Assuming from_pipe deep-copies scheduler state, passing an enumeration index instead of a scheduler timestep, or treating cooperative iteration skipping as kernel cancellation/exception cleanup can misidentify invocation ownership.",
+ "refusal_point": "Stop when the ownership question is resolved or a required fixture/backend is absent; do not infer service-wide race or serialization behavior from this readset."
},
"limitations": [
"Source-derived candidate; no installed API, native scheduler, model, concurrency or service validation.",
@@ -94,14 +99,25 @@
"provenance_readsets": [
"references/source-map.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Check wrapper and scheduler identities explicitly and follow the effective schedule reset and step-index progression.",
+ "Trace interruption through its loop-boundary consumer and normal decode/cleanup tail while preserving the caller's existing shared-state serialization.",
+ "Use a real scheduler CPU check only for its scheduler cycle; keep pipeline reuse/interruption and serving claims tied to matching fixtures and execution evidence."
+ ]
},
"applicability": {
- "scope": "FluxPipeline and FlowMatchEulerDiscreteScheduler invocation ownership at this source pin; inspect the actual caller contract."
+ "scope": "FluxPipeline and FlowMatchEulerDiscreteScheduler invocation ownership at this source pin; inspect the actual caller contract.",
+ "prerequisites": "Access to the pinned ownership path and an identified caller admission/serialization contract; other pipeline or scheduler classes need their own contracts.",
+ "input_contract": "Diffusers/PyTorch revisions, pipeline/scheduler classes and object identities, construction/from_pipe path, actual components, scheduler config and resolved timesteps/sigmas, shape/dtype, callback/interrupt timing, output/cleanup path and overlapping intervals.",
+ "produced_artifact": "An object-identity/ownership map and call-to-schedule-reset/step/interrupt/output/cleanup trace, the caller admission contract, one next check and actual evidence status.",
+ "failure_modes": "Assuming from_pipe deep-copies scheduler state, passing an enumeration index instead of a scheduler timestep, or treating cooperative iteration skipping as kernel cancellation/exception cleanup can misidentify invocation ownership.",
+ "refusal_point": "Stop when the ownership question is resolved or a required fixture/backend is absent; do not infer service-wide race or serialization behavior from this readset."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Check wrapper and scheduler identities explicitly and follow the effective schedule reset and step-index progression.",
+ "Trace interruption through its loop-boundary consumer and normal decode/cleanup tail while preserving the caller's existing shared-state serialization.",
+ "Use a real scheduler CPU check only for its scheduler cycle; keep pipeline reuse/interruption and serving claims tied to matching fixtures and execution evidence."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/draft-acceptance-issue.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/draft-acceptance-issue.json
index cd000a389..7bdaf64fc 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/draft-acceptance-issue.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/draft-acceptance-issue.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/workflows/draft-acceptance-issue/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/workflows/draft-acceptance-issue/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/skill-index.json",
"src/inferrsi/data/skill_library/workflows/draft-acceptance-issue/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dwarfstar.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dwarfstar.json
index bc848dd3a..d814af8a8 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dwarfstar.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dwarfstar.json
@@ -25,8 +25,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/repositories/dwarfstar/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/repositories/dwarfstar/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/repositories/dwarfstar/LICENSE",
"src/inferrsi/data/skill_library/repositories/dwarfstar/SKILL.md",
@@ -57,7 +57,7 @@
"source_files": {
"src/inferrsi/data/skill_library/repositories/dwarfstar/LICENSE": "4a8a1358f5b048627d71e4988de0bafd386f4988991929629e6cb9e29b7b9b2f",
"src/inferrsi/data/skill_library/repositories/dwarfstar/SKILL.md": "a04a252ecdcd4710a3273595f4bd1f3eb8e24d5bb1bfe005f058a78388254dc9",
- "src/inferrsi/data/skill_library/repositories/dwarfstar/metadata.json": "e1f0c2dc3d7ea9278039c2e8c3e8ca4f8e3aba33504b7784d628e903f0b5fc8c",
+ "src/inferrsi/data/skill_library/repositories/dwarfstar/metadata.json": "c883f6bc05e585f072cf03f734b7a296a65842c7921d3afcfe89e1977405fbd2",
"src/inferrsi/data/skill_library/repositories/dwarfstar/references/repo-provenance.md": "7d72f9c85ee60fe003f5f5a585b52e9989ebcedca625b584fa39a976d4433998",
"src/inferrsi/data/skill_library/repositories/dwarfstar/references/repo-routing-metadata.json": "8e195b420b67151932af5f435c284f49677cc7863e650c83cfa9e762fd00db0b",
"src/inferrsi/data/skill_library/repositories/dwarfstar/references/source-identity.json": "a38fe07d003accd83b7626d7e189f7b49b4b73e7a223e4d441b0e33add5d9c91",
@@ -154,7 +154,12 @@
"method": "User-approved native C adaptation and partial drafting; manual stable-ID subskill orchestration and integrated create/verify stages. Required-backend blocks disable automatic admission."
},
"applicability": {
- "scope": "Pinned DwarfStar source investigation and experiment planning; not arbitrary GGUF support or cross-backend qualification."
+ "scope": "Pinned DwarfStar source investigation and experiment planning; not arbitrary GGUF support or cross-backend qualification.",
+ "prerequisites": "Access to the pinned native-source contract and relevant memory/state/qualification route; native verification needs a compatible build, matching hardware/weights and task-specific execution authority.",
+ "input_contract": "Engine revision/patches, actual model artifact/representation, selected backend/build and topology, memory/storage budget, context/chunk/session count, image inputs, sampling/speculation policy, cache state and quality/service requirements.",
+ "produced_artifact": "A bounded executing-tuple report with observed signal, source-backed mechanism, alternate explanation, proposed falsifier, actual check outcome and unresolved backend/model/service gap.",
+ "failure_modes": "Inferring model support from flags, aliases, file mapping or quantization names can select the wrong graph; treating CPU bookkeeping or parser checks as native backend/service qualification can overstate the result.",
+ "refusal_point": "Stop at unavailable hardware/weights, unsafe exposure or the authorized budget; retain memory-guard refusals rather than disabling the guard to claim fit."
},
"limitations": [
"Metal/CUDA/ROCm inference, actual model output, SSD runtime I/O, GPU recurrent rollback, physical distributed behavior, service quality/performance and skill efficacy remain unverified.",
@@ -166,14 +171,25 @@
"references/repo-provenance.md",
"references/source-identity.json"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Choose the memory, serving-state or qualification route by the observed symptom and identify the owner, validity key and release/rollback boundary.",
+ "Freeze comparison controls, test the smallest invariant first and retain fallback messages, cancellations, refusals and unavailable fixtures.",
+ "Base model support on the actual loaded artifact and graph; use native backend evidence for model, tensor-parallel or service conclusions."
+ ]
},
"applicability": {
- "scope": "Pinned DwarfStar source investigation and experiment planning; not arbitrary GGUF support or cross-backend qualification."
+ "scope": "Pinned DwarfStar source investigation and experiment planning; not arbitrary GGUF support or cross-backend qualification.",
+ "prerequisites": "Access to the pinned native-source contract and relevant memory/state/qualification route; native verification needs a compatible build, matching hardware/weights and task-specific execution authority.",
+ "input_contract": "Engine revision/patches, actual model artifact/representation, selected backend/build and topology, memory/storage budget, context/chunk/session count, image inputs, sampling/speculation policy, cache state and quality/service requirements.",
+ "produced_artifact": "A bounded executing-tuple report with observed signal, source-backed mechanism, alternate explanation, proposed falsifier, actual check outcome and unresolved backend/model/service gap.",
+ "failure_modes": "Inferring model support from flags, aliases, file mapping or quantization names can select the wrong graph; treating CPU bookkeeping or parser checks as native backend/service qualification can overstate the result.",
+ "refusal_point": "Stop at unavailable hardware/weights, unsafe exposure or the authorized budget; retain memory-guard refusals rather than disabling the guard to claim fit."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Choose the memory, serving-state or qualification route by the observed symptom and identify the owner, validity key and release/rollback boundary.",
+ "Freeze comparison controls, test the smallest invariant first and retain fallback messages, cancellations, refusals and unavailable fixtures.",
+ "Base model support on the actual loaded artifact and graph; use native backend evidence for model, tensor-parallel or service conclusions."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/evaluate-elastic-serving-placement.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/evaluate-elastic-serving-placement.json
index 2464b7b7d..ae3161a5d 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/evaluate-elastic-serving-placement.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/evaluate-elastic-serving-placement.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/methods/evaluate-elastic-serving-placement/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/methods/evaluate-elastic-serving-placement/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/methods/evaluate-elastic-serving-placement/SKILL.md",
"src/inferrsi/data/skill_library/methods/evaluate-elastic-serving-placement/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/generation-cache-lifecycle.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/generation-cache-lifecycle.json
index 34c39a8fc..7018b6ebd 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/generation-cache-lifecycle.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/generation-cache-lifecycle.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/SKILL.md",
"src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/metadata.json",
@@ -43,7 +43,7 @@
"checkout_ref": "checkout:generation-cache-lifecycle",
"source_files": {
"src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/SKILL.md": "bf1177cb9985c12a9bf11edb64771018e1e314c4713920de8fd4a989ff70115f",
- "src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/metadata.json": "d86cecb615e93ba0f6be09d1ead102cd5a17068c07e1be2549ac5cd8f7519279",
+ "src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/metadata.json": "c2e30a2a3365dd27cac7916dc26aa356f1277d07ebee891c173ee37b77111fa1",
"src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/references/cache-stop.md": "bba6b5c73261065309cbcf11b1e591eef48916f0c79ceaaf74d98a9fa96b0992",
"src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/references/generation-batch.md": "f16d3c57d45e63bcaffd3871aa41c76972ee11f7dbf1e12b45d9b170303cad9b",
"src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/references/source-map.md": "91d26181f246307f2113f94b12ac9587d689b654dcd216e9f14b11cf5ebb650c",
@@ -89,7 +89,12 @@
}
],
"applicability": {
- "scope": "MLX-LM generation, batch-row and prefix-cache diagnosis; model/backend/runtime compatibility requires separate qualification."
+ "scope": "MLX-LM generation, batch-row and prefix-cache diagnosis; model/backend/runtime compatibility requires separate qualification.",
+ "prerequisites": "Access to the independently pinned MLX-LM generation/cache contracts and the selected entry path; model, backend and runtime compatibility is separately established for any native claim.",
+ "input_contract": "MLX-LM/MLX revisions and builds, backend/device, model/tokenizer/adapter/draft identities, entry point, tokenized prompts, cache classes, batching/prefill limits, per-request sampling/processors/stops, seed and KV quantization; failure traces include UIDs, stages, lengths, terminal reason and next-request result.",
+ "produced_artifact": "An entry-path decision, UID/stage/cache-row table, prefix identity and cache-class assumptions, terminal/cleanup trace and actual evidence status with a targeted verification plan.",
+ "failure_modes": "Ignoring server batching gates, filtering only some per-request row state, reusing a text-only prefix key or treating a stop flag as immediate MLX work cancellation can corrupt lifecycle reasoning.",
+ "refusal_point": "Stop at the first unsupported contract or missing required fixture/backend; retain a targeted plan without inferring paged allocation, radix sharing or instantaneous prefill interruption."
},
"limitations": [
"Source-derived candidate; no local native-backend, model, service or efficacy validation.",
@@ -98,14 +103,25 @@
"provenance_readsets": [
"references/source-map.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Resolve the library versus server entry and its batching gates before transferring BatchGenerator behavior.",
+ "Keep UID, caches, tokens, samplers, processors, lengths, stop matchers and pending outputs aligned under the same removal selection, including prompt-stage requests.",
+ "Check exact token/model/adapter/draft identity and cache-type copy/trim semantics, then trace cooperative removal through survivor and next-request behavior."
+ ]
},
"applicability": {
- "scope": "MLX-LM generation, batch-row and prefix-cache diagnosis; model/backend/runtime compatibility requires separate qualification."
+ "scope": "MLX-LM generation, batch-row and prefix-cache diagnosis; model/backend/runtime compatibility requires separate qualification.",
+ "prerequisites": "Access to the independently pinned MLX-LM generation/cache contracts and the selected entry path; model, backend and runtime compatibility is separately established for any native claim.",
+ "input_contract": "MLX-LM/MLX revisions and builds, backend/device, model/tokenizer/adapter/draft identities, entry point, tokenized prompts, cache classes, batching/prefill limits, per-request sampling/processors/stops, seed and KV quantization; failure traces include UIDs, stages, lengths, terminal reason and next-request result.",
+ "produced_artifact": "An entry-path decision, UID/stage/cache-row table, prefix identity and cache-class assumptions, terminal/cleanup trace and actual evidence status with a targeted verification plan.",
+ "failure_modes": "Ignoring server batching gates, filtering only some per-request row state, reusing a text-only prefix key or treating a stop flag as immediate MLX work cancellation can corrupt lifecycle reasoning.",
+ "refusal_point": "Stop at the first unsupported contract or missing required fixture/backend; retain a targeted plan without inferring paged allocation, radix sharing or instantaneous prefill interruption."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Resolve the library versus server entry and its batching gates before transferring BatchGenerator behavior.",
+ "Keep UID, caches, tokens, samplers, processors, lengths, stop matchers and pending outputs aligned under the same removal selection, including prompt-stage requests.",
+ "Check exact token/model/adapter/draft identity and cache-type copy/trim semantics, then trace cooperative removal through survivor and next-request behavior."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-memory-management.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-memory-management.json
index cb3f9a019..3d892ae4c 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-memory-management.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-memory-management.json
@@ -30,8 +30,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/hardware/host/host-memory-management/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/hardware/host/host-memory-management/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/hardware/host/host-memory-management/SKILL.md",
"src/inferrsi/data/skill_library/hardware/host/host-memory-management/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-overhead-optimization.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-overhead-optimization.json
index 41a8e3862..8d5b691b8 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-overhead-optimization.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-overhead-optimization.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/hardware/host/host-overhead-optimization/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/hardware/host/host-overhead-optimization/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/hardware/host/host-overhead-optimization/SKILL.md",
"src/inferrsi/data/skill_library/hardware/host/host-overhead-optimization/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/husky.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/husky.json
index aa920afde..0945d60d3 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/husky.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/husky.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/repositories/husky/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/repositories/husky/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/repositories/husky/SKILL.md",
"src/inferrsi/data/skill_library/repositories/husky/metadata.json",
@@ -44,7 +44,7 @@
"checkout_ref": "checkout:husky",
"source_files": {
"src/inferrsi/data/skill_library/repositories/husky/SKILL.md": "ac86afdabf9adffb684cb635971934438724bdfde8628ec549727ea23c40cf3b",
- "src/inferrsi/data/skill_library/repositories/husky/metadata.json": "9b720bc6fcb65b6dbddba36d6187b4bee86346a1bc633f55a75bd43dbe4fa04e",
+ "src/inferrsi/data/skill_library/repositories/husky/metadata.json": "b1cb0a5be02d1170125bfc3e16c044b9c1463a611a14d36fac40638524fc2e52",
"src/inferrsi/data/skill_library/repositories/husky/references/author-report.md": "4c1351ca42b2535a7c03c4da87706040650ef10d86eadcdb496e881a147fc0f7",
"src/inferrsi/data/skill_library/repositories/husky/references/repo-provenance.md": "37de912e2e2a36b44233560e6a8368b994428bff587ed4f04b1cb5fbaba0a9e7",
"src/inferrsi/data/skill_library/repositories/husky/references/source-availability.json": "d3523e53c89d4aee5bfaa17db84501aa4c4eddd59bcee486df5d48d3aa48173d",
@@ -105,7 +105,12 @@
"method": "Single coherent research workflow, bounded source summary, original verification questions and offline schema checks. Repository/environment/native stages remain unavailable."
},
"applicability": {
- "scope": "Review Husky document claims and specify missing evidence; common catalog scope is not inference backend compatibility."
+ "scope": "Review Husky document claims and specify missing evidence; common catalog scope is not inference backend compatibility.",
+ "prerequisites": "Access to the bounded official-document inventory and verification protocol; implementation and native reproduction remain unavailable until separately supplied and authenticated.",
+ "input_contract": "The exact claim, source/view date, workload and comparison intent, supporting evidence classes and available implementation/model/output/measurement artifacts; unknown identity, licensing and availability remain explicit.",
+ "produced_artifact": "A reviewable claim/evidence request recording source availability, mechanism hypothesis, counter-signal, missing identities/artifacts, proposed falsifying check and actual result or bounded INCONCLUSIVE conclusion.",
+ "failure_modes": "Treating a public-page reading as code inspection, proposed extensions as shipped behavior, summed independent ratios as one gain, or a structurally valid self-reported card as authentic evidence can overstate qualification.",
+ "refusal_point": "Stop at missing source/receipts, unsafe or private data, unsupported resources or the authorized budget; unavailable artifacts yield an unverified/INCONCLUSIVE research conclusion."
},
"limitations": [
"No canonical public repository, implementation revision, retained receipts or source-code license was identified; unknowns remain unknown.",
@@ -119,14 +124,25 @@
"references/repo-provenance.md",
"references/source-availability.json"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Separate author reports, inspected implementation, executed experiments and new interpretations, including negative findings and future work.",
+ "Request immutable identities and artifacts, then state one mechanism, competing explanation and smallest ablation with frozen semantic requirements.",
+ "Use the offline card auditor only to identify structural gaps; it authenticates neither artifacts nor execution and leaves qualification unadjudicated."
+ ]
},
"applicability": {
- "scope": "Review Husky document claims and specify missing evidence; common catalog scope is not inference backend compatibility."
+ "scope": "Review Husky document claims and specify missing evidence; common catalog scope is not inference backend compatibility.",
+ "prerequisites": "Access to the bounded official-document inventory and verification protocol; implementation and native reproduction remain unavailable until separately supplied and authenticated.",
+ "input_contract": "The exact claim, source/view date, workload and comparison intent, supporting evidence classes and available implementation/model/output/measurement artifacts; unknown identity, licensing and availability remain explicit.",
+ "produced_artifact": "A reviewable claim/evidence request recording source availability, mechanism hypothesis, counter-signal, missing identities/artifacts, proposed falsifying check and actual result or bounded INCONCLUSIVE conclusion.",
+ "failure_modes": "Treating a public-page reading as code inspection, proposed extensions as shipped behavior, summed independent ratios as one gain, or a structurally valid self-reported card as authentic evidence can overstate qualification.",
+ "refusal_point": "Stop at missing source/receipts, unsafe or private data, unsupported resources or the authorized budget; unavailable artifacts yield an unverified/INCONCLUSIVE research conclusion."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Separate author reports, inspected implementation, executed experiments and new interpretations, including negative findings and future work.",
+ "Request immutable identities and artifacts, then state one mechanism, competing explanation and smallest ablation with frozen semantic requirements.",
+ "Use the offline card auditor only to identify structural gaps; it authenticates neither artifacts nor execution and leaves qualification unadjudicated."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-benchmark-design.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-benchmark-design.json
index 6b3675e24..d62f4ccd9 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-benchmark-design.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-benchmark-design.json
@@ -30,8 +30,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/methods/inference-benchmark-design/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/methods/inference-benchmark-design/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/methods/inference-benchmark-design/SKILL.md",
"src/inferrsi/data/skill_library/methods/inference-benchmark-design/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-optimization-casebook.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-optimization-casebook.json
index a26eedb1d..144785d86 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-optimization-casebook.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-optimization-casebook.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/methods/inference-optimization-casebook/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/methods/inference-optimization-casebook/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/methods/inference-optimization-casebook/SKILL.md",
"src/inferrsi/data/skill_library/methods/inference-optimization-casebook/cases/admission-drain-occupancy.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kernel-writing-guide.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kernel-writing-guide.json
index 2d46438a2..1ef4c67c5 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kernel-writing-guide.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kernel-writing-guide.json
@@ -39,8 +39,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/methods/kernel-writing-guide/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/methods/kernel-writing-guide/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/methods/kernel-writing-guide/SKILL.md",
"src/inferrsi/data/skill_library/methods/kernel-writing-guide/cases/causal-conv1d-dispatch-state.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kv-tier-management.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kv-tier-management.json
index 84da69300..d22656682 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kv-tier-management.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kv-tier-management.json
@@ -24,8 +24,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/repositories/lmcache/kv-tier-management/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/repositories/lmcache/kv-tier-management/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/repositories/lmcache/kv-tier-management/SKILL.md",
"src/inferrsi/data/skill_library/repositories/lmcache/kv-tier-management/metadata.json",
@@ -41,7 +41,7 @@
"checkout_ref": "checkout:kv-tier-management",
"source_files": {
"src/inferrsi/data/skill_library/repositories/lmcache/kv-tier-management/SKILL.md": "e4b78f368d70c8abb74a935acb910414015f1aeec1b50f7a295e836822ec4b27",
- "src/inferrsi/data/skill_library/repositories/lmcache/kv-tier-management/metadata.json": "c221189b8d85fac3bd41ef1c7cd560f1aebc8cfa0381c182d1d571a7169be829",
+ "src/inferrsi/data/skill_library/repositories/lmcache/kv-tier-management/metadata.json": "85f7242b83334af11f57799875738b498bfe4680a51f36727792dfbe91806704",
"src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
},
"skill_metadata": {
@@ -69,12 +69,31 @@
"Original typed source inventory is not normalized; retain source identities and limits in the procedure/readsets."
],
"provenance_readsets": [],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "applicability": {
+ "scope": "LMCache multiprocess runtime KV tier policy across engine GPU cache, host L1 and configured L2 adapters, covering store/retention, prefetch admission, eviction and RAM pressure; startup loading and checkpoint recovery have separate owners.",
+ "prerequisites": "The deployment uses the inspected MP-mode semantics rather than single-process knobs; selected GDS/DAX/raw-block paths need their own platform and adapter prerequisites. Lazy offload additionally requires vLLM prefix caching.",
+ "input_contract": "MP versus single-process mode, store/retention policy and rationale, L1 manager capacities and eviction settings, L2 adapter capacities/eviction, chunk-size distribution and prefetch in-flight setting, together with token-level hit and pressure counters.",
+ "produced_artifact": "A mode- and adapter-specific tier-policy choice with retention rationale, an ordered RAM-release plan and a verification record separating L1/L2 token hits, prefetch pressure, store lag and per-manager usage.",
+ "failure_modes": "Mixing MP and in-process controls, treating key-count eviction as a byte fraction for unequal chunks, omitting fs capacity management, or increasing count-based prefetch admission without tracking L1 pressure can make the policy or memory claim unusable."
+ },
+ "recommendations": [
+ "Choose store retention and prefetched-key retention separately; use skip_l1 with noop only for buffer-only operation and account for failed stores remaining in L1.",
+ "Tune the count-based prefetch in-flight limit against L1 allocation failures and in-flight bytes; eager or warm prefetch can raise pressure before useful reuse.",
+ "Record per-adapter capacity/eviction and actual chunk-size uniformity; report token-level L1/L2 hits and sum per-l1_tag bytes rather than ratios."
+ ]
+ },
+ "applicability": {
+ "scope": "LMCache multiprocess runtime KV tier policy across engine GPU cache, host L1 and configured L2 adapters, covering store/retention, prefetch admission, eviction and RAM pressure; startup loading and checkpoint recovery have separate owners.",
+ "prerequisites": "The deployment uses the inspected MP-mode semantics rather than single-process knobs; selected GDS/DAX/raw-block paths need their own platform and adapter prerequisites. Lazy offload additionally requires vLLM prefix caching.",
+ "input_contract": "MP versus single-process mode, store/retention policy and rationale, L1 manager capacities and eviction settings, L2 adapter capacities/eviction, chunk-size distribution and prefetch in-flight setting, together with token-level hit and pressure counters.",
+ "produced_artifact": "A mode- and adapter-specific tier-policy choice with retention rationale, an ordered RAM-release plan and a verification record separating L1/L2 token hits, prefetch pressure, store lag and per-manager usage.",
+ "failure_modes": "Mixing MP and in-process controls, treating key-count eviction as a byte fraction for unequal chunks, omitting fs capacity management, or increasing count-based prefetch admission without tracking L1 pressure can make the policy or memory claim unusable."
},
- "applicability": {},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Choose store retention and prefetched-key retention separately; use skip_l1 with noop only for buffer-only operation and account for failed stores remaining in L1.",
+ "Tune the count-based prefetch in-flight limit against L1 allocation failures and in-flight bytes; eager or warm prefetch can raise pressure before useful reuse.",
+ "Record per-adapter capacity/eviction and actual chunk-size uniformity; report token-level L1/L2 hits and sum per-l1_tag bytes rather than ratios."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lazy-evaluation-stream-lifetimes.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lazy-evaluation-stream-lifetimes.json
index 314288ef3..90c6e821d 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lazy-evaluation-stream-lifetimes.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lazy-evaluation-stream-lifetimes.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/SKILL.md",
"src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/metadata.json",
@@ -42,7 +42,7 @@
"checkout_ref": "checkout:lazy-evaluation-stream-lifetimes",
"source_files": {
"src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/SKILL.md": "a95569bcc0da6e715fd74783a6a1e0489e92b680c582f15dafbf98df589d505a",
- "src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/metadata.json": "9c2293c8956508de987c886c7224d673d22912e7e915b026b05abdd287e97401",
+ "src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/metadata.json": "cfa57ac908025135113d04c3d64b9967f8d9a3af4a00707673d94b8227c85808",
"src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/references/execution-lifetime.md": "c4bc41e08aca68f661325f61e6f0db1760a079dc362f295c847bdc374fee3bf6",
"src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/references/source-map.md": "798b9cb89ed6b885eb47464ca9f137d2802a54571b9327e631d7742d3e5abd5d",
"src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/references/validation.md": "ebfc00bf7d6ab0a1dc415953ae84d56ccdc713d5614aff85e433c4a843793bff",
@@ -94,7 +94,12 @@
}
],
"applicability": {
- "scope": "MLX evaluation/stream/buffer diagnosis at the pinned implementation; qualify the actual backend separately."
+ "scope": "MLX evaluation/stream/buffer diagnosis at the pinned implementation; qualify the actual backend separately.",
+ "prerequisites": "Access to the pinned MLX execution/lifetime contract and a bounded reproduction with an identified evaluation/read boundary; device-dependent conclusions need the actual selected backend.",
+ "input_contract": "MLX revision/build, backend/device, Python/C++ entry, array shapes/dtypes, primitive streams, evaluation/read boundary and transformation context; timing needs warmup/work boundaries, while memory needs live references, in-flight buffers, allocator cache and synchronized observations.",
+ "produced_artifact": "A producer/evaluation/dependency-wait/backend/availability-cleanup trace, ownership table, observed symptom and one discriminating check with build-bound actual evidence status.",
+ "failure_modes": "Timing graph construction or submission as execution, assuming evaluated means completed, synchronizing only the default stream, or treating graph detachment/Python deletion as reusable storage can misdiagnose time or memory.",
+ "refusal_point": "Stop when the trace answers the question; if the required backend is absent, retain the source-supported diagnosis and exact runtime gap rather than substituting CPU import success."
},
"limitations": [
"Source-derived candidate; no local native-backend, model, service or efficacy validation.",
@@ -103,14 +108,25 @@
"provenance_readsets": [
"references/source-map.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Separate graph construction, async_eval submission, eval and host reads, and record which stream the measurement actually synchronizes.",
+ "Preserve the dependency-wait path and selected backend buffer-retention/donation contract before changing streams or assuming storage can be reused.",
+ "Distinguish live, in-flight and allocator-cached memory; inspect the real error/cleanup path before proposing a leak fix or hiding the failure."
+ ]
},
"applicability": {
- "scope": "MLX evaluation/stream/buffer diagnosis at the pinned implementation; qualify the actual backend separately."
+ "scope": "MLX evaluation/stream/buffer diagnosis at the pinned implementation; qualify the actual backend separately.",
+ "prerequisites": "Access to the pinned MLX execution/lifetime contract and a bounded reproduction with an identified evaluation/read boundary; device-dependent conclusions need the actual selected backend.",
+ "input_contract": "MLX revision/build, backend/device, Python/C++ entry, array shapes/dtypes, primitive streams, evaluation/read boundary and transformation context; timing needs warmup/work boundaries, while memory needs live references, in-flight buffers, allocator cache and synchronized observations.",
+ "produced_artifact": "A producer/evaluation/dependency-wait/backend/availability-cleanup trace, ownership table, observed symptom and one discriminating check with build-bound actual evidence status.",
+ "failure_modes": "Timing graph construction or submission as execution, assuming evaluated means completed, synchronizing only the default stream, or treating graph detachment/Python deletion as reusable storage can misdiagnose time or memory.",
+ "refusal_point": "Stop when the trace answers the question; if the required backend is absent, retain the source-supported diagnosis and exact runtime gap rather than substituting CPU import success."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Separate graph construction, async_eval submission, eval and host reads, and record which stream the measurement actually synchronizes.",
+ "Preserve the dependency-wait path and selected backend buffer-retention/donation contract before changing streams or assuming storage can be reused.",
+ "Distinguish live, in-flight and allocator-cached memory; inspect the real error/cleanup path before proposing a leak fix or hiding the failure."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lily.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lily.json
index 14e81969f..1abdbd6a3 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lily.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lily.json
@@ -25,8 +25,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/repositories/lily/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/repositories/lily/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/repositories/lily/LICENSE",
"src/inferrsi/data/skill_library/repositories/lily/NOTICE.md",
@@ -61,7 +61,7 @@
"src/inferrsi/data/skill_library/repositories/lily/LICENSE": "b9e2b50a48a09fb39dafa71ab1a11a6e87389addda17faa99d55f5f74a0001ac",
"src/inferrsi/data/skill_library/repositories/lily/NOTICE.md": "7a01e666044fb8aca1fd88c636bca43bcc95b8b1fecbe101dce4c7c80fbb5759",
"src/inferrsi/data/skill_library/repositories/lily/SKILL.md": "db815674d5d36964902041d310d3df844d00ab2e6bef0ce4b72ed2696ab2b9aa",
- "src/inferrsi/data/skill_library/repositories/lily/metadata.json": "6d9d54e84f957881cc26894eb82ae666f106cfc6c505924f9a6fb6b35eab457f",
+ "src/inferrsi/data/skill_library/repositories/lily/metadata.json": "52f01d58613e1a93e9cfd958ceca9ab5061baa07c5f94572393520434892b420",
"src/inferrsi/data/skill_library/repositories/lily/references/repo-provenance.md": "96182f6cb3666f89adf1aed6e216c0a84ce3a04a0685a4257e286100ba8ee367",
"src/inferrsi/data/skill_library/repositories/lily/references/repo-routing-metadata.json": "f2f3becb51056e03081c7c4140e19572a6318c3d63015884df56eb49d38a00fc",
"src/inferrsi/data/skill_library/repositories/lily/references/source-identity.json": "4595559c11611c855be150919fca3eeb89e7f96c7a8dd133b9f79a5cd8e05e62",
@@ -177,7 +177,12 @@
"method": "User-approved native Rust/Metal and partial drafting adaptation; stable-ID lanes, integrated source graph and assertion/helper verification. Missing compiler/backend/checkpoint remains explicit."
},
"applicability": {
- "scope": "Exact Lily source/checkpoint/layout/API/cache/qualification investigation; common catalog classification is not CPU inference or cross-device support."
+ "scope": "Exact Lily source/checkpoint/layout/API/cache/qualification investigation; common catalog classification is not CPU inference or cross-device support.",
+ "prerequisites": "Access to the exact Lily source/model-layout/API contracts; native checks require the matching Rust/Metal/checkpoint environment and frozen comparison requirements.",
+ "input_contract": "Source revision, Cargo lock/toolchain, macOS/chip/GPU family, checkpoint and tensor payload identities, affine configuration/layout, tokenizer/template/stops, sequence/output budgets, greedy policy, input history, prefix-cache state and quality/measurement requirements.",
+ "produced_artifact": "A source/model/device-bound representation, serving-cache or qualification diagnosis with the observation, invariant, alternate cause, smallest falsifier, actual result and remaining native gap.",
+ "failure_modes": "Equal bit width, file size, API names or prompt_cache_key do not establish tensor/token/state equivalence; ignoring the extra lookahead slot can understate request capacity, and green CPU references do not replace matched-model Metal goldens.",
+ "refusal_point": "Stop at unsupported formats/APIs, missing compiler/backend/checkpoint, unfrozen gates or the authorized budget; do not weaken ignored/golden tests to obtain a green result."
},
"limitations": [
"Cargo/Rust1.92, supported M5/macOS26/Metal and model resources are unavailable; native evidence remains BLOCKED_REQUIRED_BACKEND.",
@@ -190,14 +195,25 @@
"references/repo-provenance.md",
"references/source-identity.json"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Resolve exact affine representation and strict API semantics, then trace token frontier, cache key and restore/release ownership.",
+ "Retain the source's extra request lookahead slot in capacity arithmetic rather than counting only prompt and visible output.",
+ "Separate reported benchmark version from the checked-out runner and distinguish offline screening, native compilation, Metal operators and matched-model goldens."
+ ]
},
"applicability": {
- "scope": "Exact Lily source/checkpoint/layout/API/cache/qualification investigation; common catalog classification is not CPU inference or cross-device support."
+ "scope": "Exact Lily source/checkpoint/layout/API/cache/qualification investigation; common catalog classification is not CPU inference or cross-device support.",
+ "prerequisites": "Access to the exact Lily source/model-layout/API contracts; native checks require the matching Rust/Metal/checkpoint environment and frozen comparison requirements.",
+ "input_contract": "Source revision, Cargo lock/toolchain, macOS/chip/GPU family, checkpoint and tensor payload identities, affine configuration/layout, tokenizer/template/stops, sequence/output budgets, greedy policy, input history, prefix-cache state and quality/measurement requirements.",
+ "produced_artifact": "A source/model/device-bound representation, serving-cache or qualification diagnosis with the observation, invariant, alternate cause, smallest falsifier, actual result and remaining native gap.",
+ "failure_modes": "Equal bit width, file size, API names or prompt_cache_key do not establish tensor/token/state equivalence; ignoring the extra lookahead slot can understate request capacity, and green CPU references do not replace matched-model Metal goldens.",
+ "refusal_point": "Stop at unsupported formats/APIs, missing compiler/backend/checkpoint, unfrozen gates or the authorized budget; do not weaken ignored/golden tests to obtain a green result."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Resolve exact affine representation and strict API semantics, then trace token frontier, cache key and restore/release ownership.",
+ "Retain the source's extra request lookahead slot in capacity arithmetic rather than counting only prompt and visible output.",
+ "Separate reported benchmark version from the checked-out runner and distinguish offline screening, native compilation, Metal operators and matched-model goldens."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lithos-metal.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lithos-metal.json
index 1ebd25bb4..3ea9ef1a7 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lithos-metal.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lithos-metal.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/repositories/lithos-metal/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/repositories/lithos-metal/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/repositories/lithos-metal/LICENSE",
"src/inferrsi/data/skill_library/repositories/lithos-metal/SKILL.md",
@@ -60,7 +60,7 @@
"source_files": {
"src/inferrsi/data/skill_library/repositories/lithos-metal/LICENSE": "cfc7749b96f63bd31c3c42b5c471bf756814053e847c10f3eb003417bc523d30",
"src/inferrsi/data/skill_library/repositories/lithos-metal/SKILL.md": "4c71ba2f26c275829466bba2412738aa27799842e642c75cdfd36d9ed8713f7f",
- "src/inferrsi/data/skill_library/repositories/lithos-metal/metadata.json": "ea2d4e73e33a49433a2834d19489653f828e56f69c4975cc32cf508176eb24ae",
+ "src/inferrsi/data/skill_library/repositories/lithos-metal/metadata.json": "c5fa717d6f79d54e6aa2b0c5b672e80ac7412fbe5901ae24fda78a24c584ba67",
"src/inferrsi/data/skill_library/repositories/lithos-metal/references/mlx-license.md": "ccfab7ccb2ea306f71531c8ca77bb55507606cd90768b1e32b8b52ab5b48cf01",
"src/inferrsi/data/skill_library/repositories/lithos-metal/references/notices.md": "b2c9d2481ef2c9d16b3240c951ae65db417e5d309cd06b7917db9d84fe175886",
"src/inferrsi/data/skill_library/repositories/lithos-metal/references/package-api.md": "4f3ecfabafb2a15eddf21ccf964b8c37635bbb7862418a2c404e7049c0fa2898",
@@ -132,7 +132,11 @@
}
],
"applicability": {
- "scope": "Pinned Lithos Metal 0.1.2 source/API contracts; Apple silicon Metal runtime; model and recipe qualification required separately."
+ "scope": "Pinned Lithos Metal 0.1.2 source/API contracts; Apple silicon Metal runtime; model and recipe qualification required separately.",
+ "prerequisites": "Access to the pinned package/API and selected compiler, generation or serving contracts; revision/API drift needs a fresh bounded inspection, and efficacy claims need campaign-owned gates and matching native/model evidence.",
+ "input_contract": "Actual source/package version, backend/configuration, chip/core count and OS, target/draft checkpoint identities, context and prompt-token IDs, sampling/numerical options and the affected recipe/runtime/session/cache boundary.",
+ "produced_artifact": "A routed region/dependency, committed-state or cache-identity/lifecycle ledger, followed by the smallest supported finding, missing checks and next bounded action.",
+ "failure_modes": "Treating session adoption/cache hits as generation correctness, fewer dispatches as faster rounds, another chip recipe as the default, CPU cache checks as GPU rollback or health liveness as loaded/warm model state can yield unsupported conclusions."
},
"limitations": [
"Source-derived candidate; empirical efficacy validation not_run.",
@@ -147,14 +151,24 @@
"disable-model-invocation": true,
"metadata": {
"disco-role": "operating"
- }
+ },
+ "recommendations": [
+ "Follow session replacement across recipe geometry, committed generation state and serving adoption/cache identity before interpreting a cache hit.",
+ "Keep an unfused/native or target-only control with matching decoded weights, numerical mode, row bounds and sampling policy; verify dependencies and persistent state before timing.",
+ "Distinguish package/CPU checks, synthetic Metal execution, model continuation and HTTP workload evidence; chip recipes and target-compute projections do not establish another device or E2E rate."
+ ]
},
"applicability": {
- "scope": "Pinned Lithos Metal 0.1.2 source/API contracts; Apple silicon Metal runtime; model and recipe qualification required separately."
+ "scope": "Pinned Lithos Metal 0.1.2 source/API contracts; Apple silicon Metal runtime; model and recipe qualification required separately.",
+ "prerequisites": "Access to the pinned package/API and selected compiler, generation or serving contracts; revision/API drift needs a fresh bounded inspection, and efficacy claims need campaign-owned gates and matching native/model evidence.",
+ "input_contract": "Actual source/package version, backend/configuration, chip/core count and OS, target/draft checkpoint identities, context and prompt-token IDs, sampling/numerical options and the affected recipe/runtime/session/cache boundary.",
+ "produced_artifact": "A routed region/dependency, committed-state or cache-identity/lifecycle ledger, followed by the smallest supported finding, missing checks and next bounded action.",
+ "failure_modes": "Treating session adoption/cache hits as generation correctness, fewer dispatches as faster rounds, another chip recipe as the default, CPU cache checks as GPU rollback or health liveness as loaded/warm model state can yield unsupported conclusions."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Follow session replacement across recipe geometry, committed generation state and serving adoption/cache identity before interpreting a cache hit.",
+ "Keep an unfused/native or target-only control with matching decoded weights, numerical mode, row bounds and sampling policy; verify dependencies and persistent state before timing.",
+ "Distinguish package/CPU checks, synthetic Metal execution, model continuation and HTTP workload evidence; chip recipes and target-compute projections do not establish another device or E2E rate."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/local-dream.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/local-dream.json
index b4d2dcca2..baf434174 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/local-dream.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/local-dream.json
@@ -29,8 +29,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/repositories/local-dream/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/repositories/local-dream/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/repositories/local-dream/SKILL.md",
"src/inferrsi/data/skill_library/repositories/local-dream/metadata.json",
@@ -53,7 +53,7 @@
"checkout_ref": "checkout:local-dream",
"source_files": {
"src/inferrsi/data/skill_library/repositories/local-dream/SKILL.md": "e51fd3c71d693070abeba57ca67dcca82c7347cb94eb98f43e7c455bcacbaa46",
- "src/inferrsi/data/skill_library/repositories/local-dream/metadata.json": "ab33602252d0f5bb0797c3e1f12466b55a82aeaff10cd24e42a99a5224c8e593",
+ "src/inferrsi/data/skill_library/repositories/local-dream/metadata.json": "63b71cc6a237d8c4e00db875d15594df09d6e3cf87a1cb25183a311b4d480804",
"src/inferrsi/data/skill_library/repositories/local-dream/references/android-lifecycle.md": "f85e5fce39f5de5f88548fd8c21978240c810f899e03ff05ed72eb3508ee5e36",
"src/inferrsi/data/skill_library/repositories/local-dream/references/native-contracts.md": "40206892f5e421f9d8edf303870f5c4aefa3e87123537178d45f438f3d04b7ce",
"src/inferrsi/data/skill_library/repositories/local-dream/references/packaging-and-abi.md": "9a19bca59e19c3bfbfc1a15f5fc756c13d954ace367812aeb000551f4484ad06",
@@ -154,7 +154,11 @@
}
],
"applicability": {
- "scope": "Selected Android app/process and native diffusion boundary contracts; common is a discovery category, not CPU/device compatibility or native qualification."
+ "scope": "Selected Android app/process and native diffusion boundary contracts; common is a discovery category, not CPU/device compatibility or native qualification.",
+ "prerequisites": "Access to the pinned Android lifecycle/native/packaging contracts and identified installed artifacts; stronger conclusions require an authorized, frozen workload on the actual selected Android/ARM64 backend.",
+ "input_contract": "Source revision, installed APK/version/flavor/ABI, native executable and loaded DiT library, QNN/Hexagon/MNN builds, model/export/weights/tokenizer identities, requested and serving configuration, and lifecycle/readiness/generation/cache observations.",
+ "produced_artifact": "A bounded failed-boundary report with counter-signals, identities observed versus declared, the smallest missing discriminator and retained INCONCLUSIVE native scope.",
+ "failure_modes": "Reading UI flags or equal declared configurations as actual dispatch/readiness, treating cancellation/SSE progress as native completion, or interpreting model-directory caches as authenticated model-revision namespaces can misidentify stale state."
},
"limitations": [
"Source-derived candidate; required Android ARM64/MNN/QNN/Hexagon/model execution BLOCKED_REQUIRED_BACKEND / not_run; qualification INCONCLUSIVE.",
@@ -169,14 +173,24 @@
"references/repo-provenance.md",
"references/source-identity.json"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Compare requested and serving model/backend/dimensions/bind address through stop/reopen and health observations; never enable host mode as a diagnostic default.",
+ "Trace the actual MNN/QNN/Anima/DiT path and separate native library/ABI identity from UI runOnCpu flags and declared SDK labels.",
+ "Keep progress, cancellation, native completion, release and saved final output separate; check prompt-cache mode/version/dimensions and textual-inversion caller guards before proposing cache changes."
+ ]
},
"applicability": {
- "scope": "Selected Android app/process and native diffusion boundary contracts; common is a discovery category, not CPU/device compatibility or native qualification."
+ "scope": "Selected Android app/process and native diffusion boundary contracts; common is a discovery category, not CPU/device compatibility or native qualification.",
+ "prerequisites": "Access to the pinned Android lifecycle/native/packaging contracts and identified installed artifacts; stronger conclusions require an authorized, frozen workload on the actual selected Android/ARM64 backend.",
+ "input_contract": "Source revision, installed APK/version/flavor/ABI, native executable and loaded DiT library, QNN/Hexagon/MNN builds, model/export/weights/tokenizer identities, requested and serving configuration, and lifecycle/readiness/generation/cache observations.",
+ "produced_artifact": "A bounded failed-boundary report with counter-signals, identities observed versus declared, the smallest missing discriminator and retained INCONCLUSIVE native scope.",
+ "failure_modes": "Reading UI flags or equal declared configurations as actual dispatch/readiness, treating cancellation/SSE progress as native completion, or interpreting model-directory caches as authenticated model-revision namespaces can misidentify stale state."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Compare requested and serving model/backend/dimensions/bind address through stop/reopen and health observations; never enable host mode as a diagnostic default.",
+ "Trace the actual MNN/QNN/Anima/DiT path and separate native library/ABI identity from UI runOnCpu flags and declared SDK labels.",
+ "Keep progress, cancellation, native completion, release and saved final output separate; check prompt-cache mode/version/dimensions and textual-inversion caller guards before proposing cache changes."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mistral-rs.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mistral-rs.json
index 4a2326614..99ddd5f13 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mistral-rs.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mistral-rs.json
@@ -26,8 +26,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/repositories/mistral-rs/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/repositories/mistral-rs/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/repositories/mistral-rs/LICENSE",
"src/inferrsi/data/skill_library/repositories/mistral-rs/SKILL.md",
@@ -67,7 +67,7 @@
"source_files": {
"src/inferrsi/data/skill_library/repositories/mistral-rs/LICENSE": "b40bb09c7b3b6f697a8920cf9d2cb3ff8cfcf77a0fcb6d264a92842b43c4dcae",
"src/inferrsi/data/skill_library/repositories/mistral-rs/SKILL.md": "8917b30fb86437dea7508275ccce1992194d6e0c4858d537a410bf1f900953d0",
- "src/inferrsi/data/skill_library/repositories/mistral-rs/metadata.json": "5e88e2ca64b848b8177336351e03d1fbcb5c495fd54217daa83710ef6faccff0",
+ "src/inferrsi/data/skill_library/repositories/mistral-rs/metadata.json": "3ca0ca5f3a3c9e4c3e2c7d4587940b83123cf05f4e3213b305ccdd1ae89cd679",
"src/inferrsi/data/skill_library/repositories/mistral-rs/references/apache-component-license.md": "c71d239df91726fc519c6eb72d318ec65820627232b2f796219e87dcf35d0ab4",
"src/inferrsi/data/skill_library/repositories/mistral-rs/references/paged-component-notice.md": "a1718d8c5a25f15e61bea98a158880e96faa5b7174e82c4e1470fbacb23a3559",
"src/inferrsi/data/skill_library/repositories/mistral-rs/references/repo-provenance.md": "fab87dacbe0a7298e42137cab7409f443ece0dfefba60961040031e36d96adc8",
@@ -175,7 +175,11 @@
"method": "Stable-ID bounded decode/prefill/qualification author lanes, whole-graph reconciliation, hard cases and offline helper/static/scaffold/wheel checks; compiler and actual Apple evidence not_run."
},
"applicability": {
- "scope": "Selected native Metal paged decode/prefill layout/metadata/workspace and preparation/evidence investigation. Common catalog categorization is not CPU inference or hardware certification."
+ "scope": "Selected native Metal paged decode/prefill layout/metadata/workspace and preparation/evidence investigation. Common catalog categorization is not CPU inference or hardware certification.",
+ "prerequisites": "Access to the pinned Rust/Candle/MSL decode, prefill and qualification contracts; native comparisons need the matched Rust/Apple/Candle/shader stack and separately authorized frozen references.",
+ "input_contract": "mistral.rs/Candle/native artifact revisions, Rust toolchain, Apple SDK/OS/device, actual model/cache dtype and representation, shape/strides/offsets, block table, context and execution limits.",
+ "produced_artifact": "A source/build-bound route and invariant report with actual versus declared facts, safe support result, counter-signal and smallest next native check; unavailable Metal/kernel/model evidence remains blocked.",
+ "failure_modes": "Equating element counts with physical alias/layout identity, transferring decode geometry to prefill, or treating prewarm/pipeline preparation as actual dispatch can misclassify eligibility and native correctness."
},
"limitations": [
"BLOCKED_REQUIRED_BACKEND: actual Rust/Candle/Apple Metal compilation, native tensor/cache/dispatch/parity/model/quality/service/performance evidence not_run.",
@@ -188,14 +192,24 @@
"references/repo-provenance.md",
"references/source-identity.json"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Select decode, prefill or qualification by the failed contract, then follow cache/metadata bytes from producer through view to native consumer.",
+ "Keep static predicates, pipeline preparation, kernel dispatch, numerical correctness and model/service outcomes as separate evidence steps; retain skipped or gated tests.",
+ "Use the matched Candle revision and native shader/build inputs; broad workspace tests can require model fixtures, tokens and downloads."
+ ]
},
"applicability": {
- "scope": "Selected native Metal paged decode/prefill layout/metadata/workspace and preparation/evidence investigation. Common catalog categorization is not CPU inference or hardware certification."
+ "scope": "Selected native Metal paged decode/prefill layout/metadata/workspace and preparation/evidence investigation. Common catalog categorization is not CPU inference or hardware certification.",
+ "prerequisites": "Access to the pinned Rust/Candle/MSL decode, prefill and qualification contracts; native comparisons need the matched Rust/Apple/Candle/shader stack and separately authorized frozen references.",
+ "input_contract": "mistral.rs/Candle/native artifact revisions, Rust toolchain, Apple SDK/OS/device, actual model/cache dtype and representation, shape/strides/offsets, block table, context and execution limits.",
+ "produced_artifact": "A source/build-bound route and invariant report with actual versus declared facts, safe support result, counter-signal and smallest next native check; unavailable Metal/kernel/model evidence remains blocked.",
+ "failure_modes": "Equating element counts with physical alias/layout identity, transferring decode geometry to prefill, or treating prewarm/pipeline preparation as actual dispatch can misclassify eligibility and native correctness."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Select decode, prefill or qualification by the failed contract, then follow cache/metadata bytes from producer through view to native consumer.",
+ "Keep static predicates, pipeline preparation, kernel dispatch, numerical correctness and model/service outcomes as separate evidence steps; retain skipped or gated tests.",
+ "Use the matched Candle revision and native shader/build inputs; broad workspace tests can require model fixtures, tokens and downloads."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlc-llm.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlc-llm.json
index 1792ecdab..0a5db775c 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlc-llm.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlc-llm.json
@@ -27,8 +27,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/repositories/mlc-llm/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/repositories/mlc-llm/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/repositories/mlc-llm/LICENSE",
"src/inferrsi/data/skill_library/repositories/mlc-llm/SKILL.md",
@@ -71,7 +71,7 @@
"source_files": {
"src/inferrsi/data/skill_library/repositories/mlc-llm/LICENSE": "c71d239df91726fc519c6eb72d318ec65820627232b2f796219e87dcf35d0ab4",
"src/inferrsi/data/skill_library/repositories/mlc-llm/SKILL.md": "e0fd1accb4216191c56e7e4a8074633f36d9063727c2546e0d2eed14220d8cf4",
- "src/inferrsi/data/skill_library/repositories/mlc-llm/metadata.json": "c9927e284e44aa09733424c024409ec8efc9593520418bb7d2e53ef301363681",
+ "src/inferrsi/data/skill_library/repositories/mlc-llm/metadata.json": "f20ca326ca0966e8d56b5fa698f971001a4af77370a145631e98476601610fa9",
"src/inferrsi/data/skill_library/repositories/mlc-llm/references/mlc-notice.md": "3a7dfb37bb96ce6f5a1cc0d6d4c8da3b8a35e6ec4692647ba0a37f6155c7e694",
"src/inferrsi/data/skill_library/repositories/mlc-llm/references/repo-provenance.md": "423f21e1de6d7ab2e11c0bdf540357efde025ae4535bdf76ed9770b1a59ba35a",
"src/inferrsi/data/skill_library/repositories/mlc-llm/references/repo-routing-metadata.json": "5309c27e7b60c4fe7aaef28bf6911e7e3c936b58b1a907aa8486bf0e40b4c0d0",
@@ -186,7 +186,11 @@
}
],
"applicability": {
- "scope": "Selected artifact/Apple target/pinned TVM Metal consumer and KV configuration source contracts. Common catalog categorization is not CPU inference or Apple qualification."
+ "scope": "Selected artifact/Apple target/pinned TVM Metal consumer and KV configuration source contracts. Common catalog categorization is not CPU inference or Apple qualification.",
+ "prerequisites": "Access to the pinned MLC and exact TVM fork/gitlink contracts; native checks require separately prepared matching MLC/TVM/FFI and the applicable Apple SDK/device stack.",
+ "input_contract": "MLC revision, TVM gitlink/build, model/tokenizer/quantized-parameter identity, target/host/SDK, compiled artifact/format, runtime device and the actual compiled/runtime configuration sources.",
+ "produced_artifact": "An identity and actual-versus-declared boundary report for artifact, Metal runtime or KV-limit ownership, with a negative signal and smallest missing native discriminator.",
+ "failure_modes": "Treating a schema hash as payload authentication, an opt-in WebLLM artifact contract as a native-engine ABI, serializable Metal fallback as runnable, or automatic capacity estimates as free memory can create unsupported qualification claims."
},
"limitations": [
"Source-derived candidate; original compiled CPU/native tests, compiler/live API/CLI/model/Metal/fit/quality/service/performance execution not_run; required native BLOCKED_REQUIRED_BACKEND.",
@@ -200,14 +204,24 @@
"references/repo-provenance.md",
"references/source-identity.json"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Identify the interface's actual consumer before using a valid artifact manifest as evidence of native model/task support.",
+ "Separate codegen/export from runtime factory/load, pipeline creation, dispatch and stream completion; a serializable kind=metal fallback can still be non-runnable.",
+ "Trace runtime preference, compile-time bounds, nested workspace inputs and explicit overrides separately before interpreting a KV or working-set estimate."
+ ]
},
"applicability": {
- "scope": "Selected artifact/Apple target/pinned TVM Metal consumer and KV configuration source contracts. Common catalog categorization is not CPU inference or Apple qualification."
+ "scope": "Selected artifact/Apple target/pinned TVM Metal consumer and KV configuration source contracts. Common catalog categorization is not CPU inference or Apple qualification.",
+ "prerequisites": "Access to the pinned MLC and exact TVM fork/gitlink contracts; native checks require separately prepared matching MLC/TVM/FFI and the applicable Apple SDK/device stack.",
+ "input_contract": "MLC revision, TVM gitlink/build, model/tokenizer/quantized-parameter identity, target/host/SDK, compiled artifact/format, runtime device and the actual compiled/runtime configuration sources.",
+ "produced_artifact": "An identity and actual-versus-declared boundary report for artifact, Metal runtime or KV-limit ownership, with a negative signal and smallest missing native discriminator.",
+ "failure_modes": "Treating a schema hash as payload authentication, an opt-in WebLLM artifact contract as a native-engine ABI, serializable Metal fallback as runnable, or automatic capacity estimates as free memory can create unsupported qualification claims."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Identify the interface's actual consumer before using a valid artifact manifest as evidence of native model/task support.",
+ "Separate codegen/export from runtime factory/load, pipeline creation, dispatch and stream completion; a serializable kind=metal fallback can still be non-runnable.",
+ "Trace runtime preference, compile-time bounds, nested workspace inputs and explicit overrides separately before interpreting a KV or working-set estimate."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlx-swift-lm.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlx-swift-lm.json
index e42d9f9e3..2120572f0 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlx-swift-lm.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlx-swift-lm.json
@@ -27,8 +27,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/repositories/mlx-swift-lm/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/repositories/mlx-swift-lm/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/repositories/mlx-swift-lm/LICENSE",
"src/inferrsi/data/skill_library/repositories/mlx-swift-lm/SKILL.md",
@@ -62,7 +62,7 @@
"source_files": {
"src/inferrsi/data/skill_library/repositories/mlx-swift-lm/LICENSE": "7b1d86acac816ce8192209d7e01141c129b4d48d474c15cc5fb4af62f497c323",
"src/inferrsi/data/skill_library/repositories/mlx-swift-lm/SKILL.md": "501f73e7b327b04b6b80501b42b20dab1ac15c875817e6f4c4b66fe3e69b4fc4",
- "src/inferrsi/data/skill_library/repositories/mlx-swift-lm/metadata.json": "014131f345a28d3367313ed5a4cce858fe4c93d4d47a5f03eb89f609592e032f",
+ "src/inferrsi/data/skill_library/repositories/mlx-swift-lm/metadata.json": "6d5bd7d5f3566b75c8c0b53be95d524db039a517566826c211c978987d78b253",
"src/inferrsi/data/skill_library/repositories/mlx-swift-lm/references/repo-provenance.md": "f1b15edf89c597db0c3fc0bff7db6b2451924ce9273c39623bc0cbdccb568d31",
"src/inferrsi/data/skill_library/repositories/mlx-swift-lm/references/repo-routing-metadata.json": "8fc804347771cd768a7ed027f2041d96c9832fe5f8c5869e6593a3021b1bef8b",
"src/inferrsi/data/skill_library/repositories/mlx-swift-lm/references/scope-and-notices.md": "18e56805a8f275d35af3d8e42f3aa1525871b43400921a5d2dca4897c2047214",
@@ -175,7 +175,11 @@
"method": "Stable-ID lifecycle and recurrent-kernel author lanes, integrated root/local references, hard cases and offline support/static/scaffold/package checks; native Swift/Xcode/MLX compiler/tests not_run."
},
"applicability": {
- "scope": "Selected Swift ownership/cancellation/settling and actual GatedDelta Metal/ops contracts. Common categorization is not CPU inference or Apple qualification."
+ "scope": "Selected Swift ownership/cancellation/settling and actual GatedDelta Metal/ops contracts. Common categorization is not CPU inference or Apple qualification.",
+ "prerequisites": "Access to the Swift lifecycle and selected GatedDelta contracts, with resolved MLX/native identities for a comparison; native checks require a prepared matching Swift/Xcode/Apple stack and separate authorization.",
+ "input_contract": "Exact source revision, resolved Swift/MLX dependencies, Xcode/SDK/OS/device, API family and native artifact identity; the selected question supplies lifecycle ownership/completion observations or recurrent producer shape/dtype/layout and actual consumer/fallback.",
+ "produced_artifact": "An identity and actual-versus-declared report for generation ownership or recurrent kernel invariants, with safe support result, counter-signal and smallest separately authorized native check.",
+ "failure_modes": "Conflating lexical isolation with exclusive access across await or completed task join, treating useKernel as observed dispatch, calling an ops fallback CPU execution, or assuming a manifest range is a resolved dependency can invalidate the conclusion."
},
"limitations": [
"BLOCKED_REQUIRED_BACKEND: native Swift/MLX compiler, host-control tests and required Apple Metal cancellation/backgrounding/kernel/cache/model/fit/quality/service/performance evidence not_run.",
@@ -189,14 +193,24 @@
"references/repo-provenance.md",
"references/source-identity.json"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Resolve mutable-context/cache ownership and task settlement separately from lexical isolation and queued GPU work before cache reuse or producer shutdown.",
+ "Follow recurrent shape/dtype/layout to the actual Metal or ops consumer; predicate eligibility and useKernel alone do not prove dispatch.",
+ "Record the actual dependency resolution and use the documented Xcode macOS test route; the source states swift test does not work for this package."
+ ]
},
"applicability": {
- "scope": "Selected Swift ownership/cancellation/settling and actual GatedDelta Metal/ops contracts. Common categorization is not CPU inference or Apple qualification."
+ "scope": "Selected Swift ownership/cancellation/settling and actual GatedDelta Metal/ops contracts. Common categorization is not CPU inference or Apple qualification.",
+ "prerequisites": "Access to the Swift lifecycle and selected GatedDelta contracts, with resolved MLX/native identities for a comparison; native checks require a prepared matching Swift/Xcode/Apple stack and separate authorization.",
+ "input_contract": "Exact source revision, resolved Swift/MLX dependencies, Xcode/SDK/OS/device, API family and native artifact identity; the selected question supplies lifecycle ownership/completion observations or recurrent producer shape/dtype/layout and actual consumer/fallback.",
+ "produced_artifact": "An identity and actual-versus-declared report for generation ownership or recurrent kernel invariants, with safe support result, counter-signal and smallest separately authorized native check.",
+ "failure_modes": "Conflating lexical isolation with exclusive access across await or completed task join, treating useKernel as observed dispatch, calling an ops fallback CPU execution, or assuming a manifest range is a resolved dependency can invalidate the conclusion."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Resolve mutable-context/cache ownership and task settlement separately from lexical isolation and queued GPU work before cache reuse or producer shutdown.",
+ "Follow recurrent shape/dtype/layout to the actual Metal or ops consumer; predicate eligibility and useKernel alone do not prove dispatch.",
+ "Record the actual dependency resolution and use the documented Xcode macOS test route; the source states swift test does not work for this package."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/npu-skill-factory.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/npu-skill-factory.json
index 822382292..1feb8312a 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/npu-skill-factory.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/npu-skill-factory.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/workflows/npu-skill-factory/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/workflows/npu-skill-factory/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/skill-index.json",
"src/inferrsi/data/skill_library/workflows/npu-skill-factory/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/optimization-search-policy.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/optimization-search-policy.json
index 3acce4fec..99382ef56 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/optimization-search-policy.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/optimization-search-policy.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/methods/optimization-search-policy/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/methods/optimization-search-policy/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/methods/optimization-search-policy/SKILL.md",
"src/inferrsi/data/skill_library/methods/optimization-search-policy/cases/sarathi-chunk-budget.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/os-process-placement.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/os-process-placement.json
index d8a00b0fb..c07bff462 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/os-process-placement.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/os-process-placement.json
@@ -24,8 +24,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/hardware/host/os-process-placement/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/hardware/host/os-process-placement/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/hardware/host/os-process-placement/SKILL.md",
"src/inferrsi/data/skill_library/hardware/host/os-process-placement/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/performance-ceiling-analysis.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/performance-ceiling-analysis.json
index 968956ca4..aeb500c1b 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/performance-ceiling-analysis.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/performance-ceiling-analysis.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/SKILL.md",
"src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/plan-activation-residency.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/plan-activation-residency.json
index 9c43c1ff9..9693249f5 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/plan-activation-residency.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/plan-activation-residency.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/methods/plan-activation-residency/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/methods/plan-activation-residency/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/methods/plan-activation-residency/SKILL.md",
"src/inferrsi/data/skill_library/methods/plan-activation-residency/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/project-maya.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/project-maya.json
index d1a1ecc97..616aed8d2 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/project-maya.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/project-maya.json
@@ -25,8 +25,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/repositories/project-maya/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/repositories/project-maya/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/repositories/project-maya/LICENSE",
"src/inferrsi/data/skill_library/repositories/project-maya/SKILL.md",
@@ -63,7 +63,7 @@
"source_files": {
"src/inferrsi/data/skill_library/repositories/project-maya/LICENSE": "574cffb7c57769a9271043ac0c2b07c5e7e7621cfb81081b901d8981e791d121",
"src/inferrsi/data/skill_library/repositories/project-maya/SKILL.md": "c619a3472be02147a847a24ca97c56afd8b56b6ffd22ce66fcae453155e34bab",
- "src/inferrsi/data/skill_library/repositories/project-maya/metadata.json": "c9f7879ca512c07e9bdf6d1d1a52de055a0fa210b8838c205e27a34436324834",
+ "src/inferrsi/data/skill_library/repositories/project-maya/metadata.json": "c48af69c430f23a8b39a2dbc09495bef54b48147c64068b39cfb3551f4e94c63",
"src/inferrsi/data/skill_library/repositories/project-maya/references/license-boundaries.md": "090839df90d6e96f55794c2ad821d7154dc6a6aa4211cdbb08a1e5f6f2e52a92",
"src/inferrsi/data/skill_library/repositories/project-maya/references/repo-provenance.md": "7bb6b6c14d1a9b77e906804505c723569f952e5d47559e636d2e6621994216bf",
"src/inferrsi/data/skill_library/repositories/project-maya/references/repo-routing-metadata.json": "0acbb111e92316e74d537200cd033f3d7d0f69972bf0953c8cea9dfb2ecb4a34",
@@ -225,7 +225,12 @@
"method": "Stable-ID isolated author lanes, integrated runtime graph, assertion/helper/static and selected direct-compiler CPU verification; required accelerator/model blocks retained."
},
"applicability": {
- "scope": "Model-specific dispatch/residency, API/native state and qualification investigation. Common catalog placement is not CPU inference or cross-device certification."
+ "scope": "Model-specific dispatch/residency, API/native state and qualification investigation. Common catalog placement is not CPU inference or cross-device certification.",
+ "prerequisites": "Access to the selected model-family runtime/state/qualification contract; native changes need frozen correctness/comparison requirements and the matching authorized CUDA/HIP/model resources.",
+ "input_contract": "Full source revision, actual executable/model family, model and asset identities, representation, compiler/runtime/device, rendered workload, sampling policy, single/parallel route, exact tokens/images and relevant completion observations.",
+ "produced_artifact": "A route/identity/evidence-class report with invariant, proposed check, actual result, counter-signal, missing resource and next authorized action; missing native requirements remain INCONCLUSIVE.",
+ "failure_modes": "Transferring coverage through a common facade or quantization name, confusing API history with native KV/private state, double-counting unified RAM/VRAM, or repairing a slot header to appear compatible can invalidate diagnosis.",
+ "refusal_point": "When required CUDA/HIP/model evidence is unavailable, retain BLOCKED_REQUIRED_BACKEND and an unverified/INCONCLUSIVE conclusion; exit-zero skips or CPU simulations cannot clear it."
},
"limitations": [
"BLOCKED_REQUIRED_BACKEND: no actual CUDA/HIP/model/device, quantized payload, vision, parallel/state/restore or cancellation-cleanup evidence.",
@@ -238,14 +243,25 @@
"references/repo-provenance.md",
"references/source-identity.json"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Select the family-specific route and preserve the explicitly shallow Qwen coverage rather than transferring deeper GLM/DeepSeek/MiniMax evidence.",
+ "Separate shared weights, private conversation state, API history and native KV; count unified physical memory once.",
+ "Preserve exact token/image identities and completion boundaries before reuse, then test one falsifiable hypothesis without modifying slot headers to bypass compatibility."
+ ]
},
"applicability": {
- "scope": "Model-specific dispatch/residency, API/native state and qualification investigation. Common catalog placement is not CPU inference or cross-device certification."
+ "scope": "Model-specific dispatch/residency, API/native state and qualification investigation. Common catalog placement is not CPU inference or cross-device certification.",
+ "prerequisites": "Access to the selected model-family runtime/state/qualification contract; native changes need frozen correctness/comparison requirements and the matching authorized CUDA/HIP/model resources.",
+ "input_contract": "Full source revision, actual executable/model family, model and asset identities, representation, compiler/runtime/device, rendered workload, sampling policy, single/parallel route, exact tokens/images and relevant completion observations.",
+ "produced_artifact": "A route/identity/evidence-class report with invariant, proposed check, actual result, counter-signal, missing resource and next authorized action; missing native requirements remain INCONCLUSIVE.",
+ "failure_modes": "Transferring coverage through a common facade or quantization name, confusing API history with native KV/private state, double-counting unified RAM/VRAM, or repairing a slot header to appear compatible can invalidate diagnosis.",
+ "refusal_point": "When required CUDA/HIP/model evidence is unavailable, retain BLOCKED_REQUIRED_BACKEND and an unverified/INCONCLUSIVE conclusion; exit-zero skips or CPU simulations cannot clear it."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Select the family-specific route and preserve the explicitly shallow Qwen coverage rather than transferring deeper GLM/DeepSeek/MiniMax evidence.",
+ "Separate shared weights, private conversation state, API history and native KV; count unified physical memory once.",
+ "Preserve exact token/image identities and completion boundaries before reuse, then test one falsifiable hypothesis without modifying slot headers to bypass compatibility."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/repository-evolution-analysis.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/repository-evolution-analysis.json
index 2c630b6c4..84f9811fa 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/repository-evolution-analysis.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/repository-evolution-analysis.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/methods/repository-evolution-analysis/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/methods/repository-evolution-analysis/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/methods/repository-evolution-analysis/SKILL.md",
"src/inferrsi/data/skill_library/methods/repository-evolution-analysis/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/rsi-evidence.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/rsi-evidence.json
index fccdde4cf..de8a66bfc 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/rsi-evidence.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/rsi-evidence.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/workflows/rsi-evidence/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/workflows/rsi-evidence/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/skill-index.json",
"src/inferrsi/data/skill_library/workflows/rsi-evidence/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/skill-evolution-playbook.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/skill-evolution-playbook.json
index 4be510eb3..2ca4217a2 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/skill-evolution-playbook.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/skill-evolution-playbook.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/methods/skill-evolution-playbook/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/methods/skill-evolution-playbook/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/methods/skill-evolution-playbook/SKILL.md",
"src/inferrsi/data/skill_library/methods/skill-evolution-playbook/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/splash.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/splash.json
index e1ab587f6..578955d20 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/splash.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/splash.json
@@ -25,8 +25,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/repositories/splash/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/repositories/splash/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/repositories/splash/LICENSE",
"src/inferrsi/data/skill_library/repositories/splash/SKILL.md",
@@ -61,7 +61,7 @@
"src/inferrsi/data/skill_library/repositories/splash/LICENSE": "cfc7749b96f63bd31c3c42b5c471bf756814053e847c10f3eb003417bc523d30",
"src/inferrsi/data/skill_library/repositories/splash/SKILL.md": "619b87c282dcec4befa319085764fb53fa506a05c6a7112500bcca662f059ece",
"src/inferrsi/data/skill_library/repositories/splash/THIRD_PARTY_NOTICES": "2fedf1277bf732f8435c3e3be2798f52f25f7e2bd4155865cb366da55b88e50b",
- "src/inferrsi/data/skill_library/repositories/splash/metadata.json": "d4c679f474c31549156aad417e23c2802b9f4bc904472e5cb16916f2996b50ba",
+ "src/inferrsi/data/skill_library/repositories/splash/metadata.json": "9d0613387fdaee61de7e79146e2a896f19ac7f5c1d355f74afe767074f64dd2a",
"src/inferrsi/data/skill_library/repositories/splash/references/repo-provenance.md": "20e3462f1b6a8752eb0fa6f36e76867f142a49c1953f498f895e29c4fd4f0204",
"src/inferrsi/data/skill_library/repositories/splash/references/repo-routing-metadata.json": "287ae591f42949b8d33312659ec8303c47d5c73093c64c125edb1cf6ea8de8b9",
"src/inferrsi/data/skill_library/repositories/splash/references/source-identity.json": "2f778882ec440538a46f48332650412b446722246c8dc5690083f19da7dce61d",
@@ -189,7 +189,12 @@
"method": "User-approved native Metal/C++ and isolated source-Python adaptation; stable-ID lanes, integrated graph, assertion cases and create/verify stages. Required-backend block disables automatic admission."
},
"applicability": {
- "scope": "Pinned Splash representation, planning, state/serving and verification investigation; common discovery does not qualify a different Mac/model/backend."
+ "scope": "Pinned Splash representation, planning, state/serving and verification investigation; common discovery does not qualify a different Mac/model/backend.",
+ "prerequisites": "Access to the pinned representation/planning, serving-cache or qualification route; native checks need matching Apple/backend/model resources, an authorized workload and frozen semantic/measurement controls.",
+ "input_contract": "Source/build, macOS/device, target and compatible draft artifacts/formats, tokenizer/template/vision selection, context/cache/memory controls, KV/recurrent precision, ANE and verifier-row policy, concurrency/priority/sampling, quality controls and cache state.",
+ "produced_artifact": "An execution-identity report with observation, source-backed invariant, alternate explanation, smallest discriminating check, actual result and unresolved Metal/ANE/model/disk-state gap.",
+ "failure_modes": "Treating an API alias as new weights, equal bit width as loader/kernel compatibility, or a fixed seed as fixed cache/batch/adaptive-verification scheduling can invalidate the comparison; CPU cache/codec checks do not prove GPU state.",
+ "refusal_point": "Stop at unsupported formats, unavailable backend/model resources, unsafe service exposure or the authorized budget."
},
"limitations": [
"Required Metal/ANE/model/actual device and disk-state execution remains BLOCKED_REQUIRED_BACKEND; Python policy/codec checks do not clear it.",
@@ -201,14 +206,25 @@
"references/repo-provenance.md",
"references/source-identity.json"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Resolve representation, memory/cache owner, validity key and reclaim/rollback boundary before tuning a numeric control.",
+ "Freeze semantic and workload controls beyond the seed, and retain actual-path fallbacks, admission waits, evictions, cancellations and errors.",
+ "Inspect options without starting the normal serve path, which can download target/draft weights; installation and serving require separate task authority."
+ ]
},
"applicability": {
- "scope": "Pinned Splash representation, planning, state/serving and verification investigation; common discovery does not qualify a different Mac/model/backend."
+ "scope": "Pinned Splash representation, planning, state/serving and verification investigation; common discovery does not qualify a different Mac/model/backend.",
+ "prerequisites": "Access to the pinned representation/planning, serving-cache or qualification route; native checks need matching Apple/backend/model resources, an authorized workload and frozen semantic/measurement controls.",
+ "input_contract": "Source/build, macOS/device, target and compatible draft artifacts/formats, tokenizer/template/vision selection, context/cache/memory controls, KV/recurrent precision, ANE and verifier-row policy, concurrency/priority/sampling, quality controls and cache state.",
+ "produced_artifact": "An execution-identity report with observation, source-backed invariant, alternate explanation, smallest discriminating check, actual result and unresolved Metal/ANE/model/disk-state gap.",
+ "failure_modes": "Treating an API alias as new weights, equal bit width as loader/kernel compatibility, or a fixed seed as fixed cache/batch/adaptive-verification scheduling can invalidate the comparison; CPU cache/codec checks do not prove GPU state.",
+ "refusal_point": "Stop at unsupported formats, unavailable backend/model resources, unsafe service exposure or the authorized budget."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Resolve representation, memory/cache owner, validity key and reclaim/rollback boundary before tuning a numeric control.",
+ "Freeze semantic and workload controls beyond the seed, and retain actual-path fallbacks, admission waits, evictions, cancellations and errors.",
+ "Inspect options without starting the normal serve path, which can download target/draft weights; installation and serving require separate task authority."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/storage-io-measurement.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/storage-io-measurement.json
index cfe190d5c..4c2c81c2c 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/storage-io-measurement.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/storage-io-measurement.json
@@ -24,8 +24,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/hardware/host/storage-io-measurement/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/hardware/host/storage-io-measurement/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/hardware/host/storage-io-measurement/SKILL.md",
"src/inferrsi/data/skill_library/hardware/host/storage-io-measurement/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-input-pipeline.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-input-pipeline.json
index b1fa42bc6..ede094a22 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-input-pipeline.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-input-pipeline.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/methods/training-input-pipeline/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/methods/training-input-pipeline/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/methods/training-input-pipeline/SKILL.md",
"src/inferrsi/data/skill_library/methods/training-input-pipeline/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-resource-budget.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-resource-budget.json
index b2164f169..e191fddb1 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-resource-budget.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-resource-budget.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/methods/training-resource-budget/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/methods/training-resource-budget/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/methods/training-resource-budget/SKILL.md",
"src/inferrsi/data/skill_library/methods/training-resource-budget/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/video-diffusion-optimization.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/video-diffusion-optimization.json
index d00fa08ff..c00ffdadd 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/video-diffusion-optimization.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/video-diffusion-optimization.json
@@ -26,8 +26,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/methods/video-diffusion-optimization/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/methods/video-diffusion-optimization/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/methods/video-diffusion-optimization/SKILL.md",
"src/inferrsi/data/skill_library/methods/video-diffusion-optimization/assets/campaign.template.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/vllm-metal.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/vllm-metal.json
index c3248f455..135cea491 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/vllm-metal.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/vllm-metal.json
@@ -26,8 +26,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/repositories/vllm-metal/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/repositories/vllm-metal/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/repositories/vllm-metal/LICENSE",
"src/inferrsi/data/skill_library/repositories/vllm-metal/SKILL.md",
@@ -65,7 +65,7 @@
"source_files": {
"src/inferrsi/data/skill_library/repositories/vllm-metal/LICENSE": "c71d239df91726fc519c6eb72d318ec65820627232b2f796219e87dcf35d0ab4",
"src/inferrsi/data/skill_library/repositories/vllm-metal/SKILL.md": "5f83b1b989e8dc239a6f34a6b3bd63872533ecc1404483a69dabfb1865648439",
- "src/inferrsi/data/skill_library/repositories/vllm-metal/metadata.json": "63d51d1ace6b0ba503c28751a8ffbf9224038395f4146a94c775c628d8075470",
+ "src/inferrsi/data/skill_library/repositories/vllm-metal/metadata.json": "55652ec6de744d92044951e25ea89aa18c92b14846dd4b597444d773bb8c31e3",
"src/inferrsi/data/skill_library/repositories/vllm-metal/references/repo-provenance.md": "970bda4d1081704aa82c94d6544a98bbf479430c4fc2c82a1a82f206043aa72a",
"src/inferrsi/data/skill_library/repositories/vllm-metal/references/repo-routing-metadata.json": "230a2b383bcf2855af4de50374c41f11b0b0841268ae45e424c1c97f679b477a",
"src/inferrsi/data/skill_library/repositories/vllm-metal/references/scope-and-notices.md": "aaef98eaea4cff56f39852e1d131163ebc01334e79034b1890f67a2660727d3e",
@@ -185,7 +185,11 @@
"method": "Stable-ID manual author sequencing plus bounded cache author lane; integrated graph followed by assertion/helper/static and selected pure Python source tests; Apple native blocks retained."
},
"applicability": {
- "scope": "Selected plugin preparation, KV/offload ownership and speculative/deferred/dispatch evidence investigation. Common catalog scope does not certify CPU inference or cross-backend qualification."
+ "scope": "Selected plugin preparation, KV/offload ownership and speculative/deferred/dispatch evidence investigation. Common catalog scope does not certify CPU inference or cross-backend qualification.",
+ "prerequisites": "Access to the plugin-specific platform/ABI, cache-lifecycle and qualification contracts; actual engine changes require separate authority, resource ownership, frozen gates and cleanup, with a supported artifact pairing.",
+ "input_contract": "Source/package/runtime/native artifact identities, Python/OS/chip/device, backend/executor, checkpoint/tokenizer/draft identities, format/dtype, effective feature flags and workload.",
+ "produced_artifact": "An identity/route/invariant report with actual safe-check result, counter-signal and next authorized native check; partial CPU evidence stays separate from unverified transfers, sleep/wake, model and HTTP outcomes.",
+ "failure_modes": "Using lazy import success as native registration or ABI evidence, treating a feature switch as dispatch, or conflating cache capacity with physical residency and kernel time with HTTP time can overstate plugin behavior."
},
"limitations": [
"BLOCKED_REQUIRED_BACKEND: actual Apple MLX/Metal and MPS kernels/model loading, cache transfers/sleep/wake and real worker dispatch are not verified.",
@@ -198,14 +202,24 @@
"references/repo-provenance.md",
"references/source-identity.json"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Resolve platform registration, backend/runner and native ABI pairing before capacity tuning; a package visibility check establishes only its reported software facts.",
+ "Keep MLX and MPS, full-attention and recurrent state, logical cache capacity and physical residency as separate contracts.",
+ "Preserve failed paths and distinguish parser/API checks from real transfers, sleep/wake, model output and HTTP performance evidence."
+ ]
},
"applicability": {
- "scope": "Selected plugin preparation, KV/offload ownership and speculative/deferred/dispatch evidence investigation. Common catalog scope does not certify CPU inference or cross-backend qualification."
+ "scope": "Selected plugin preparation, KV/offload ownership and speculative/deferred/dispatch evidence investigation. Common catalog scope does not certify CPU inference or cross-backend qualification.",
+ "prerequisites": "Access to the plugin-specific platform/ABI, cache-lifecycle and qualification contracts; actual engine changes require separate authority, resource ownership, frozen gates and cleanup, with a supported artifact pairing.",
+ "input_contract": "Source/package/runtime/native artifact identities, Python/OS/chip/device, backend/executor, checkpoint/tokenizer/draft identities, format/dtype, effective feature flags and workload.",
+ "produced_artifact": "An identity/route/invariant report with actual safe-check result, counter-signal and next authorized native check; partial CPU evidence stays separate from unverified transfers, sleep/wake, model and HTTP outcomes.",
+ "failure_modes": "Using lazy import success as native registration or ABI evidence, treating a feature switch as dispatch, or conflating cache capacity with physical residency and kernel time with HTTP time can overstate plugin behavior."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Resolve platform registration, backend/runner and native ABI pairing before capacity tuning; a package visibility check establishes only its reported software facts.",
+ "Keep MLX and MPS, full-attention and recurrent state, logical cache capacity and physical residency as separate contracts.",
+ "Preserve failed paths and distinguish parser/API checks from real transfers, sleep/wake, model output and HTTP performance evidence."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/weight-loading-optimization.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/weight-loading-optimization.json
index 5fe7522d4..1a9752e29 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/weight-loading-optimization.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/weight-loading-optimization.json
@@ -29,8 +29,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/methods/weight-loading-optimization/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/methods/weight-loading-optimization/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/methods/weight-loading-optimization/SKILL.md",
"src/inferrsi/data/skill_library/methods/weight-loading-optimization/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-afd-p2p-exchanges.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-afd-p2p-exchanges.json
index 646537a8f..3b456e047 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-afd-p2p-exchanges.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-afd-p2p-exchanges.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/SKILL.md",
"src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/metadata.json",
@@ -42,7 +42,7 @@
"checkout_ref": "checkout:audit-afd-p2p-exchanges",
"source_files": {
"src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/SKILL.md": "a46110891eca860915e314d6921a22856d631453b22b17837fe30262d6780535",
- "src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/metadata.json": "4e6100dfb9fc7bc00b19915f4fd38ad509704f2a609c658adf4d632d4e05b6c1",
+ "src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/metadata.json": "ae8c7582265dd6902cc4172293fc709244cc915df9390672fde25687b9e79b4b",
"src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/references/afd-plugin-LICENSE.md": "0e8566ea362276878e1f38f97da49964dbea301c15d4aa9d727f97a4c39fd943",
"src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/references/source-notes.md": "5073d87450cb60d1b245308dff8807e73749c07de976d50f2fa26694fad81a72",
"src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
@@ -92,7 +92,12 @@
}
],
"applicability": {
- "scope": "Source-based CUDA P2pNccl rank/payload audit; selected eager dense-FFN path with compute_gate_on_attention=false. Repository identity afd_plugin; runtime/backend compatibility unverified."
+ "scope": "Source-based CUDA P2pNccl rank/payload audit; selected eager dense-FFN path with compute_gate_on_attention=false. Repository identity afd_plugin; runtime/backend compatibility unverified.",
+ "prerequisites": "Access to the pinned source ledger and caller-level evidence for the selected eager dense FFN path with Attention-side gating disabled; any proposed cross-stream reuse needs an established communicator completion contract.",
+ "input_contract": "Plugin and dependency revisions, connector and role, A/F worker counts, DP/PCP/TP coordinates, worker/device ranks, subgroup roster, stage, per-peer token counts, payload shape/dtype/device, transfer context and returned slices.",
+ "produced_artifact": "A bounded rank-namespace and per-peer payload ownership ledger, the earliest evidenced mismatch or missing edge, and a proposed verification plan.",
+ "failure_modes": "Inferring data peers from control senders, accepting equal total row counts despite a peer permutation, or treating an even-split fallback or Python return as proof of ownership/completion can produce a wrong exchange diagnosis.",
+ "refusal_point": "Stop on A < F, invalid role ranks, duplicate/missing assignments or an unresolved rank namespace; return the conflicting callers and ledger entries without inventing ranks or attributing the fault to networking."
},
"limitations": [
"Source-derived candidate; native exchange, model quality, backend compatibility, performance and agent efficacy not_run.",
@@ -107,14 +112,25 @@
"../../../methods/inference-benchmark-design/SKILL.md",
"../../../methods/dense-performance-evidence/SKILL.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Record role, world and subgroup-local ranks separately; reconstruct the actual subgroup roster before interpreting send/receive peers.",
+ "Track distinct row identities and stage-specific peer lengths through concatenation and splitting; use same-total swapped lengths as a countercheck.",
+ "Establish the last communicator/stream completion obligation before reusing or releasing a receive buffer."
+ ]
},
"applicability": {
- "scope": "Source-based CUDA P2pNccl rank/payload audit; selected eager dense-FFN path with compute_gate_on_attention=false. Repository identity afd_plugin; runtime/backend compatibility unverified."
+ "scope": "Source-based CUDA P2pNccl rank/payload audit; selected eager dense-FFN path with compute_gate_on_attention=false. Repository identity afd_plugin; runtime/backend compatibility unverified.",
+ "prerequisites": "Access to the pinned source ledger and caller-level evidence for the selected eager dense FFN path with Attention-side gating disabled; any proposed cross-stream reuse needs an established communicator completion contract.",
+ "input_contract": "Plugin and dependency revisions, connector and role, A/F worker counts, DP/PCP/TP coordinates, worker/device ranks, subgroup roster, stage, per-peer token counts, payload shape/dtype/device, transfer context and returned slices.",
+ "produced_artifact": "A bounded rank-namespace and per-peer payload ownership ledger, the earliest evidenced mismatch or missing edge, and a proposed verification plan.",
+ "failure_modes": "Inferring data peers from control senders, accepting equal total row counts despite a peer permutation, or treating an even-split fallback or Python return as proof of ownership/completion can produce a wrong exchange diagnosis.",
+ "refusal_point": "Stop on A < F, invalid role ranks, duplicate/missing assignments or an unresolved rank namespace; return the conflicting callers and ledger entries without inventing ranks or attributing the fault to networking."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Record role, world and subgroup-local ranks separately; reconstruct the actual subgroup roster before interpreting send/receive peers.",
+ "Track distinct row identities and stage-specific peer lengths through concatenation and splitting; use same-total swapped lengths as a countercheck.",
+ "Establish the last communicator/stream completion obligation before reusing or releasing a receive buffer."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-flare-session-lifecycle.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-flare-session-lifecycle.json
index c96c05cf1..943853aee 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-flare-session-lifecycle.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-flare-session-lifecycle.json
@@ -29,8 +29,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/LICENSE",
"src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/SKILL.md",
@@ -52,7 +52,7 @@
"source_files": {
"src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/LICENSE": "cfc7749b96f63bd31c3c42b5c471bf756814053e847c10f3eb003417bc523d30",
"src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/SKILL.md": "02a5f7c18e44201215ecec17c6ac96b27be9148bcc189df287f39d1b372b7721",
- "src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/metadata.json": "58716737446f46bd1de77aa323c5652f97277d44dd952c82b42399b2cb769b87",
+ "src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/metadata.json": "9c667c7735ebe4fb07247e098218f6b03a1c1fafa776f1372f8d948ea8ea756b",
"src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/references/diagnostics.md": "0908a0eae53a2e10775a8b6ee3e3944b2fa410174402b5a1a8c98eb674dd4b4d",
"src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/references/lifecycle.md": "107ebe3c22e5dba8007c3b8ea61149f10bc55c51b557a4230f3e5de859bb0f7c",
"src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/references/source-ledger.json": "61e51a97ffd7c1fa7866bdd440b024d0c5e0065635ac8d0ba53d2c69c7f9e470",
@@ -126,7 +126,12 @@
"method": "AREX scope/ground/construct/verify and repository evidence/usability methods adapted to InferRSI native owners and packaging; no DisCo runtime or managed import."
},
"applicability": {
- "scope": "Pinned Flare complete-session restore and admission audit for its NVIDIA SM120/SM121 research prototype. CUDA catalog scope is content classification, not device or backend qualification."
+ "scope": "Pinned Flare complete-session restore and admission audit for its NVIDIA SM120/SM121 research prototype. CUDA catalog scope is content classification, not device or backend qualification.",
+ "prerequisites": "Access to package/source identity and bounded restore traces, with preparation compatibility established before cache-state diagnosis. Running, freeing resources or changing configuration requires separate authority.",
+ "input_contract": "Execution ABI and prepared-contract version, ordinary/speculative configuration, model/layout identity, request IDs and consumed-token endpoints, GPU/host prefixes, continuation row, ticket/waiter identities, transfer events, fences and resource counts.",
+ "produced_artifact": "A checkpoint/ticket ownership table, five-outcome admission ledger and diagnosis identifying the earliest unsupported transition and smallest missing observation.",
+ "failure_modes": "A KV-only lookup hit can omit continuation state; admission can be mistaken for actual reuse; client cancellation can be mistaken for transfer completion; mixing absolute timestamps from client and native clocks can corrupt the diagnosis.",
+ "refusal_point": "Stop with an unresolved edge when package identity, layout, transfer completion or clock provenance is unknown."
},
"limitations": [
"Source-only candidate; native Flare/model/GPU/service and skill efficacy were not evaluated.",
@@ -139,14 +144,25 @@
"references/source-notes.md",
"references/source-ledger.json"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Pair the immutable KV parent chain with the complete continuation row at the same consumed-token endpoint before calling a checkpoint restorable.",
+ "Distinguish lookup outcomes, destination reservation, ticket creation, transfer completion, publication and actual reused-token evidence.",
+ "Follow scheduler-owned tickets through waiter cancellation and check write/use fences before allowing storage reuse."
+ ]
},
"applicability": {
- "scope": "Pinned Flare complete-session restore and admission audit for its NVIDIA SM120/SM121 research prototype. CUDA catalog scope is content classification, not device or backend qualification."
+ "scope": "Pinned Flare complete-session restore and admission audit for its NVIDIA SM120/SM121 research prototype. CUDA catalog scope is content classification, not device or backend qualification.",
+ "prerequisites": "Access to package/source identity and bounded restore traces, with preparation compatibility established before cache-state diagnosis. Running, freeing resources or changing configuration requires separate authority.",
+ "input_contract": "Execution ABI and prepared-contract version, ordinary/speculative configuration, model/layout identity, request IDs and consumed-token endpoints, GPU/host prefixes, continuation row, ticket/waiter identities, transfer events, fences and resource counts.",
+ "produced_artifact": "A checkpoint/ticket ownership table, five-outcome admission ledger and diagnosis identifying the earliest unsupported transition and smallest missing observation.",
+ "failure_modes": "A KV-only lookup hit can omit continuation state; admission can be mistaken for actual reuse; client cancellation can be mistaken for transfer completion; mixing absolute timestamps from client and native clocks can corrupt the diagnosis.",
+ "refusal_point": "Stop with an unresolved edge when package identity, layout, transfer completion or clock provenance is unknown."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Pair the immutable KV parent chain with the complete continuation row at the same consumed-token endpoint before calling a checkpoint restorable.",
+ "Distinguish lookup outcomes, destination reservation, ticket creation, transfer completion, publication and actual reused-token evidence.",
+ "Follow scheduler-owned tickets through waiter cancellation and check write/use fences before allowing storage reuse."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/diagnose-attention-bottlenecks.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/diagnose-attention-bottlenecks.json
index 98d7ef8a0..38600a9a9 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/diagnose-attention-bottlenecks.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/diagnose-attention-bottlenecks.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/hardware/cuda/diagnose-attention-bottlenecks/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/hardware/cuda/diagnose-attention-bottlenecks/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/hardware/cuda/diagnose-attention-bottlenecks/SKILL.md",
"src/inferrsi/data/skill_library/hardware/cuda/diagnose-attention-bottlenecks/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/host-overhead-profiling.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/host-overhead-profiling.json
index d9278ddbe..1e9737954 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/host-overhead-profiling.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/host-overhead-profiling.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/hardware/cuda/host-overhead-profiling/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/hardware/cuda/host-overhead-profiling/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/hardware/cuda/host-overhead-profiling/SKILL.md",
"src/inferrsi/data/skill_library/hardware/cuda/host-overhead-profiling/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/reason-hopper-kernels.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/reason-hopper-kernels.json
index 7ae3cae3b..47f5ae5c1 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/reason-hopper-kernels.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/reason-hopper-kernels.json
@@ -36,8 +36,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/95b1e4bac05edd101392a70920fc2295074f0848/src/inferrsi/data/skill_library/hardware/cuda/reason-hopper-kernels/SKILL.md",
- "commit": "95b1e4bac05edd101392a70920fc2295074f0848",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/hardware/cuda/reason-hopper-kernels/SKILL.md",
+ "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
"paths": [
"src/inferrsi/data/skill_library/hardware/cuda/reason-hopper-kernels/SKILL.md",
"src/inferrsi/data/skill_library/hardware/cuda/reason-hopper-kernels/metadata.json",
From 6b8093dbf1ca055f11ce9242bb5f88fe9ff4990a Mon Sep 17 00:00:00 2001
From: Dong1017 <55723162+Dong1017@users.noreply.github.com>
Date: Sat, 31 Oct 2026 14:03:06 +0800
Subject: [PATCH 06/11] docs: regenerate skill projections from published
source history
Keep the canonical source snapshot reachable as a preceding commit.
Agent-assisted-by: Codex / gpt-6-astra [effort=xhigh] (author)
Signed-off-by: Dong1017 <55723162+Dong1017@users.noreply.github.com>
---
docs/skill-coverage.md | 4 +--
.../repo_packs/inferrsi/knowledge/README.md | 4 +--
.../ascend-aclgraph-capture-replay.json | 4 +--
.../ascend/skills/ascend-device-health.json | 4 +--
.../skills/ascend-distributed-parallel.json | 4 +--
.../ascend/skills/ascend-graph-mode.json | 4 +--
.../skills/ascend-model-adaptation.json | 4 +--
.../skills/ascend-operator-profiling.json | 4 +--
.../ascend/skills/ascend-operator-tuning.json | 4 +--
.../skills/ascend-optimization-casebook.json | 4 +--
.../skills/ascend-performance-analysis.json | 4 +--
.../ascend-video-diffusion-optimization.json | 4 +--
.../ascend/skills/ascendc-tiling-design.json | 4 +--
.../skills/cann-attribution-overlap.json | 4 +--
.../torchnpu-device-memory-lifecycle.json | 4 +--
...torchnpu-pinned-host-memory-lifecycle.json | 4 +--
.../skills/assess-serving-preemption.json | 4 +--
.../assess-split-backward-schedule.json | 4 +--
.../skills/audit-fixed-workload-claim.json | 4 +--
.../audit-h3-metal-fasth3-semantics.json | 4 +--
.../skills/audit-h3-runtime-residency.json | 4 +--
.../skills/audit-llama-runtime-state.json | 4 +--
.../audit-recurrent-kv-finalization.json | 4 +--
.../common/skills/audit-sd-cpp-residency.json | 4 +--
.../audit-sglang-diffusion-lifetimes.json | 4 +--
.../skills/audit-vllm-kv-lifetimes.json | 4 +--
.../bound-windowed-attention-replay.json | 4 +--
.../budget-speculative-verification.json | 4 +--
.../skills/calibrate-serving-simulation.json | 4 +--
.../skills/cgroup-resource-controls.json | 4 +--
.../checkpoint-recovery-consistency.json | 4 +--
.../skills/choose-parallel-training-plan.json | 4 +--
.../skills/cite-check-downstream-pr.json | 4 +--
.../common/skills/claim-contract.json | 4 +--
.../knowledge/common/skills/comfy-ui.json | 4 +--
.../common/skills/competitor-scan.json | 4 +--
.../skills/dcp-async-save-performance.json | 4 +--
.../skills/dense-performance-evidence.json | 4 +--
.../diagnose-vllm-admission-pressure.json | 4 +--
.../skills/diffusers-invocation-state.json | 4 +--
.../common/skills/draft-acceptance-issue.json | 4 +--
.../knowledge/common/skills/dwarfstar.json | 4 +--
.../evaluate-elastic-serving-placement.json | 4 +--
.../skills/generation-cache-lifecycle.json | 4 +--
.../common/skills/host-memory-management.json | 4 +--
.../skills/host-overhead-optimization.json | 4 +--
.../knowledge/common/skills/husky.json | 4 +--
.../skills/inference-benchmark-design.json | 4 +--
.../inference-optimization-casebook.json | 4 +--
.../common/skills/kernel-writing-guide.json | 4 +--
.../common/skills/kv-tier-management.json | 4 +--
.../lazy-evaluation-stream-lifetimes.json | 4 +--
.../knowledge/common/skills/lily.json | 4 +--
.../knowledge/common/skills/lithos-metal.json | 4 +--
.../knowledge/common/skills/local-dream.json | 4 +--
.../knowledge/common/skills/mistral-rs.json | 4 +--
.../knowledge/common/skills/mlc-llm.json | 4 +--
.../knowledge/common/skills/mlx-swift-lm.json | 4 +--
.../common/skills/npu-skill-factory.json | 4 +--
.../skills/optimization-search-policy.json | 4 +--
.../common/skills/os-process-placement.json | 4 +--
.../skills/performance-ceiling-analysis.json | 32 +++++++++++++++----
.../skills/plan-activation-residency.json | 4 +--
.../knowledge/common/skills/project-maya.json | 4 +--
.../skills/repository-evolution-analysis.json | 4 +--
.../knowledge/common/skills/rsi-evidence.json | 4 +--
.../skills/skill-evolution-playbook.json | 4 +--
.../knowledge/common/skills/splash.json | 4 +--
.../common/skills/storage-io-measurement.json | 4 +--
.../skills/training-input-pipeline.json | 4 +--
.../skills/training-resource-budget.json | 4 +--
.../skills/video-diffusion-optimization.json | 4 +--
.../knowledge/common/skills/vllm-metal.json | 4 +--
.../skills/weight-loading-optimization.json | 4 +--
.../cuda/skills/audit-afd-p2p-exchanges.json | 4 +--
.../skills/audit-flare-session-lifecycle.json | 4 +--
.../diagnose-attention-bottlenecks.json | 4 +--
.../cuda/skills/host-overhead-profiling.json | 4 +--
.../cuda/skills/reason-hopper-kernels.json | 4 +--
79 files changed, 181 insertions(+), 163 deletions(-)
diff --git a/docs/skill-coverage.md b/docs/skill-coverage.md
index 247c8d7eb..48fab8258 100644
--- a/docs/skill-coverage.md
+++ b/docs/skill-coverage.md
@@ -4,7 +4,7 @@ Generated from the registry, the skill sidecars and the two maintained
matrices. **Do not edit by hand**; regenerate after any registry, skill,
matrix or frozen-plan change.
-- Input digest: `sha256:aac486c04d4b627a74f2078c266811cefa42aeef8ea920806e2a889fb19851b7`
+- Input digest: `sha256:92e3d3809e7c4ad1549be166745e7e7b0b3b9a65d97e3d49c2172c8e0bec878e`
- Regenerate: `python tools/build_skill_coverage.py`
- Verify without writing: `python tools/build_skill_coverage.py --check`
@@ -130,7 +130,7 @@ optimization is fully covered or that any procedure has measured benefit.
| [inference-optimization-casebook](../src/inferrsi/data/skill_library/methods/inference-optimization-casebook/SKILL.md) | common | How to structure and use a casebook of past optimization decisions as a decision-support tool. Extracts the casebook ... | **unknown** | — | 1 Accelerator · Optimize | 1 · Systems sources (NVIDIA/TensorRT-LLM @ f33e0e5fd) | candidate · not_run · 0 readset(s) | no-applicability, no-eval |
| [kernel-writing-guide](../src/inferrsi/data/skill_library/methods/kernel-writing-guide/SKILL.md) | common | Decision guide for writing custom inference kernels: when to write one, which framework to choose (Triton, CuTe DSL ... | **unknown** | — | 1 Accelerator · Optimize / Diagnose / Configure / Verify | 1 · Systems sources (Dao-AILab/causal-conv1d @ `cd81f0413cad2fc1e6f17e785ac39f59aae690cd` ...)
1 · Systems sources (NVIDIA/TensorRT-LLM @ f33e0e5fd)
1 · Systems sources (huggingface/transformers @ `0dad7b822255a0ae261ec45ae937371e859ffd1a` ...)
1 · Systems sources (mindspore-ai/akg @ `5aa15f3fe4d856eb60cd3b27447e04f5f1fb4dff` ...)
1 · Systems sources (sgl-project/sglang @ `16176278d0180f1a1a578e53e6eef432577876c2` ...)
1 · Systems sources (vllm-project/vllm @ `537a79f0fc5ba564d2bef43d54915b2f80f9f77d` (root ...) | candidate · not_run · 7 readset(s) | no-applicability, no-eval |
| [optimization-search-policy](../src/inferrsi/data/skill_library/methods/optimization-search-policy/SKILL.md) | common | Policy for how an optimization agent should select, order, and stop optimization experiments. Covers candidate ... | **unknown** | — | 1 Accelerator · Optimize | 2 · Methodology and corpus references (JiusiServe/vllm-evolve @ 43cc82c)
2 · Methodology and corpus references (RRSI — arXiv 2609.24972v3)
2 · Methodology and corpus references (arXiv 2608.01867 (CRISP))
2 · Methodology and corpus references (microsoft/SkillOpt @ 343db22) | candidate · not_run · 9 readset(s) | no-applicability, no-eval |
-| [performance-ceiling-analysis](../src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/SKILL.md) | common | Estimate optimization headroom before modifying an inference system. Produces a three-level ceiling model (analytical ... | inference serving systems (LLM, diffusion, audio) | — | 1 Accelerator · Diagnose / Verify
5 OS · Configure | 1 · Systems sources (NVIDIA/TensorRT-LLM @ f33e0e5fd)
1 · Systems sources (ebugger/Empirical-Roofline-Toolkit @ 0771fa9)
1 · Systems sources (hahnyuan/LLM-Viewer @ 1893e4b) | candidate · not_run · 16 readset(s) | no-eval |
+| [performance-ceiling-analysis](../src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/SKILL.md) | common | Estimate optimization headroom before modifying an inference system. Produces a three-level ceiling model (analytical ... | inference serving systems (LLM, diffusion, audio) | — | 1 Accelerator · Diagnose / Verify
5 OS · Configure | 1 · Systems sources (Mooncake — FAST '25 proceedings, pp. 155-170)
1 · Systems sources (NVIDIA/TensorRT-LLM @ f33e0e5fd)
1 · Systems sources (ebugger/Empirical-Roofline-Toolkit @ 0771fa9)
1 · Systems sources (hahnyuan/LLM-Viewer @ 1893e4b) | candidate · not_run · 17 readset(s) | no-eval |
| [plan-activation-residency](../src/inferrsi/data/skill_library/methods/plan-activation-residency/SKILL.md) | common | Propose training activation and gradient memory reductions by distinguishing lifetime reordering, address packing and ... | **unknown** | — | 1 Accelerator · Diagnose / Optimize / Verify | 1 · Systems sources (Checkmate — [1910.02653v3](https://arxiv.org/abs/1910.02653v3)) | candidate · not_run · 1 readset(s) | no-applicability, no-eval |
| [repository-evolution-analysis](../src/inferrsi/data/skill_library/methods/repository-evolution-analysis/SKILL.md) | common | Analyze a repository's commit history to build a capability-evolution model: what the system can do, how it got there ... | any software repository with a recoverable git history | — | — | 2 · Methodology and corpus references (LearningToCommit/LearningToCommit @ 7e5a1d7)
2 · Methodology and corpus references (yetone/magpie @ 77800032) | candidate · not_run · 0 readset(s) | no-SearchSpace-layer, no-eval |
| [skill-evolution-playbook](../src/inferrsi/data/skill_library/methods/skill-evolution-playbook/SKILL.md) | common | Practical design choices for evolving agent skills from execution traces: when one trajectory should update a skill ... | **unknown** | — | — | 2 · Methodology and corpus references (Qwen-Applications/Trace2Skill @ 3d0b52a)
2 · Methodology and corpus references (microsoft/SkillOpt @ 343db22) | candidate · not_run · 0 readset(s) | no-SearchSpace-layer, no-applicability, no-eval |
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/README.md b/src/inferrsi/data/repo_packs/inferrsi/knowledge/README.md
index 745c2a8d3..4fcff0249 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/README.md
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/README.md
@@ -15,7 +15,7 @@ These are knowledge abstracts (`--kind knowledge`), with namespaced package
IDs distinct from explicitly selected `checkout_ref` skills. Ascend/CUDA
procedures use their corresponding backend; shared methods use `common`.
-Source snapshot: [224f6534dc5596ff78fd2f8329d73658c6bc78da](https://github.com/InferMatrixLab/InferRSI/tree/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library).
+Source snapshot: [a7f23dfef4770e8542f8390709b029fc934928e4](https://github.com/InferMatrixLab/InferRSI/tree/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library).
```bash
inferrsi knowledge search 'profiling' --repo-type inferrsi --limit 50
@@ -29,7 +29,7 @@ contain the current input paths and bytes (Git checkout filters are respected).
```bash
python tools/build_skill_index.py --source-revision
-python tools/build_skill_index.py --source-revision 224f6534dc5596ff78fd2f8329d73658c6bc78da --check
+python tools/build_skill_index.py --source-revision a7f23dfef4770e8542f8390709b029fc934928e4 --check
```
`--check` detects missing, stale and unexpected generated records without writing.
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-aclgraph-capture-replay.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-aclgraph-capture-replay.json
index 467c280e6..36d1943b5 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-aclgraph-capture-replay.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-aclgraph-capture-replay.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/hardware/ascend/ascend-aclgraph-capture-replay/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/hardware/ascend/ascend-aclgraph-capture-replay/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-aclgraph-capture-replay/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-aclgraph-capture-replay/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-device-health.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-device-health.json
index c772ccb0f..64c151619 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-device-health.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-device-health.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/hardware/ascend/ascend-device-health/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/hardware/ascend/ascend-device-health/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-device-health/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-device-health/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-distributed-parallel.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-distributed-parallel.json
index 41bc38666..39fe6821f 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-distributed-parallel.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-distributed-parallel.json
@@ -27,8 +27,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/hardware/ascend/ascend-distributed-parallel/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/hardware/ascend/ascend-distributed-parallel/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-distributed-parallel/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-distributed-parallel/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-graph-mode.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-graph-mode.json
index 256e50fd9..753ce0763 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-graph-mode.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-graph-mode.json
@@ -26,8 +26,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/hardware/ascend/ascend-graph-mode/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/hardware/ascend/ascend-graph-mode/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-graph-mode/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-graph-mode/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-model-adaptation.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-model-adaptation.json
index 9e7b5a5f0..c0da75751 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-model-adaptation.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-model-adaptation.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/hardware/ascend/ascend-model-adaptation/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/hardware/ascend/ascend-model-adaptation/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-model-adaptation/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-model-adaptation/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-profiling.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-profiling.json
index d3cd21640..18e566431 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-profiling.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-profiling.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-profiling/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-profiling/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-profiling/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-profiling/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-tuning.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-tuning.json
index d95afd452..f3adf32b8 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-tuning.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-tuning.json
@@ -24,8 +24,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-tuning/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-tuning/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-tuning/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-tuning/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-optimization-casebook.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-optimization-casebook.json
index c0943f86d..66096ed72 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-optimization-casebook.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-optimization-casebook.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/hardware/ascend/ascend-optimization-casebook/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/hardware/ascend/ascend-optimization-casebook/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-optimization-casebook/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-optimization-casebook/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-performance-analysis.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-performance-analysis.json
index fb47f4d6f..badfcf4c5 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-performance-analysis.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-performance-analysis.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/hardware/ascend/ascend-performance-analysis/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/hardware/ascend/ascend-performance-analysis/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-performance-analysis/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-performance-analysis/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-video-diffusion-optimization.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-video-diffusion-optimization.json
index ae629ffa8..8dbcf642e 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-video-diffusion-optimization.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-video-diffusion-optimization.json
@@ -26,8 +26,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/hardware/ascend/ascend-video-diffusion-optimization/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/hardware/ascend/ascend-video-diffusion-optimization/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-video-diffusion-optimization/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-video-diffusion-optimization/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascendc-tiling-design.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascendc-tiling-design.json
index ada79f388..f9b6799d6 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascendc-tiling-design.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascendc-tiling-design.json
@@ -24,8 +24,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/hardware/ascend/ascendc-tiling-design/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/hardware/ascend/ascendc-tiling-design/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascendc-tiling-design/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascendc-tiling-design/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/cann-attribution-overlap.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/cann-attribution-overlap.json
index acb5add78..db98eb823 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/cann-attribution-overlap.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/cann-attribution-overlap.json
@@ -24,8 +24,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/hardware/ascend/cann-attribution-overlap/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/hardware/ascend/cann-attribution-overlap/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/cann-attribution-overlap/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/cann-attribution-overlap/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-device-memory-lifecycle.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-device-memory-lifecycle.json
index 34da136ca..c15d230fe 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-device-memory-lifecycle.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-device-memory-lifecycle.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/hardware/ascend/torchnpu-device-memory-lifecycle/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/hardware/ascend/torchnpu-device-memory-lifecycle/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/torchnpu-device-memory-lifecycle/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/torchnpu-device-memory-lifecycle/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-pinned-host-memory-lifecycle.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-pinned-host-memory-lifecycle.json
index 573bf36a4..d6c1aaa18 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-pinned-host-memory-lifecycle.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-pinned-host-memory-lifecycle.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/hardware/ascend/torchnpu-pinned-host-memory-lifecycle/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/hardware/ascend/torchnpu-pinned-host-memory-lifecycle/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/torchnpu-pinned-host-memory-lifecycle/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/torchnpu-pinned-host-memory-lifecycle/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-serving-preemption.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-serving-preemption.json
index 493140b27..ab7bd4668 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-serving-preemption.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-serving-preemption.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/methods/assess-serving-preemption/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/methods/assess-serving-preemption/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/methods/assess-serving-preemption/SKILL.md",
"src/inferrsi/data/skill_library/methods/assess-serving-preemption/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-split-backward-schedule.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-split-backward-schedule.json
index cf9137b3b..b31436fde 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-split-backward-schedule.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-split-backward-schedule.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/methods/assess-split-backward-schedule/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/methods/assess-split-backward-schedule/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/methods/assess-split-backward-schedule/SKILL.md",
"src/inferrsi/data/skill_library/methods/assess-split-backward-schedule/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-fixed-workload-claim.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-fixed-workload-claim.json
index 524ec576b..dc95214e5 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-fixed-workload-claim.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-fixed-workload-claim.json
@@ -20,8 +20,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/methods/audit-fixed-workload-claim/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/methods/audit-fixed-workload-claim/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/methods/audit-fixed-workload-claim/SKILL.md",
"src/inferrsi/data/skill_library/methods/audit-fixed-workload-claim/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-metal-fasth3-semantics.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-metal-fasth3-semantics.json
index 8c330b417..d578f60e2 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-metal-fasth3-semantics.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-metal-fasth3-semantics.json
@@ -29,8 +29,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/LICENSE",
"src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-runtime-residency.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-runtime-residency.json
index 3f4f8ccdd..b34f75b7d 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-runtime-residency.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-runtime-residency.json
@@ -30,8 +30,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/LICENSE",
"src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-llama-runtime-state.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-llama-runtime-state.json
index b5910acf6..470b7eeb5 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-llama-runtime-state.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-llama-runtime-state.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/SKILL.md",
"src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-recurrent-kv-finalization.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-recurrent-kv-finalization.json
index ed4ff460f..e1b53cfd2 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-recurrent-kv-finalization.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-recurrent-kv-finalization.json
@@ -26,8 +26,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/LICENSE",
"src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sd-cpp-residency.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sd-cpp-residency.json
index 6bc677079..f6d61e148 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sd-cpp-residency.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sd-cpp-residency.json
@@ -25,8 +25,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/LICENSE",
"src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sglang-diffusion-lifetimes.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sglang-diffusion-lifetimes.json
index 6a2493a65..a11d61ab5 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sglang-diffusion-lifetimes.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sglang-diffusion-lifetimes.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/LICENSE",
"src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-vllm-kv-lifetimes.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-vllm-kv-lifetimes.json
index 9e9196a20..5a13147cf 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-vllm-kv-lifetimes.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-vllm-kv-lifetimes.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/SKILL.md",
"src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/bound-windowed-attention-replay.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/bound-windowed-attention-replay.json
index 138ada970..90eb3eee0 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/bound-windowed-attention-replay.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/bound-windowed-attention-replay.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/SKILL.md",
"src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/budget-speculative-verification.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/budget-speculative-verification.json
index c88908d7a..fc97f0365 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/budget-speculative-verification.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/budget-speculative-verification.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/SKILL.md",
"src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/calibrate-serving-simulation.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/calibrate-serving-simulation.json
index dfa546a59..3fdbb750d 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/calibrate-serving-simulation.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/calibrate-serving-simulation.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/methods/calibrate-serving-simulation/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/methods/calibrate-serving-simulation/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/methods/calibrate-serving-simulation/SKILL.md",
"src/inferrsi/data/skill_library/methods/calibrate-serving-simulation/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cgroup-resource-controls.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cgroup-resource-controls.json
index 291752a82..da420d71d 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cgroup-resource-controls.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cgroup-resource-controls.json
@@ -25,8 +25,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/hardware/host/cgroup-resource-controls/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/hardware/host/cgroup-resource-controls/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/hardware/host/cgroup-resource-controls/SKILL.md",
"src/inferrsi/data/skill_library/hardware/host/cgroup-resource-controls/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/checkpoint-recovery-consistency.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/checkpoint-recovery-consistency.json
index 2cb453fde..a21ab3a02 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/checkpoint-recovery-consistency.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/checkpoint-recovery-consistency.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/methods/checkpoint-recovery-consistency/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/methods/checkpoint-recovery-consistency/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/methods/checkpoint-recovery-consistency/SKILL.md",
"src/inferrsi/data/skill_library/methods/checkpoint-recovery-consistency/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/choose-parallel-training-plan.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/choose-parallel-training-plan.json
index cfc1ead65..7e43beb7d 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/choose-parallel-training-plan.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/choose-parallel-training-plan.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/methods/choose-parallel-training-plan/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/methods/choose-parallel-training-plan/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/methods/choose-parallel-training-plan/SKILL.md",
"src/inferrsi/data/skill_library/methods/choose-parallel-training-plan/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cite-check-downstream-pr.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cite-check-downstream-pr.json
index ef2f13585..a7dffc283 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cite-check-downstream-pr.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cite-check-downstream-pr.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/workflows/cite-check-downstream-pr/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/workflows/cite-check-downstream-pr/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/skill-index.json",
"src/inferrsi/data/skill_library/workflows/cite-check-downstream-pr/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/claim-contract.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/claim-contract.json
index 94270e56f..2a622fe88 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/claim-contract.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/claim-contract.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/workflows/claim-contract/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/workflows/claim-contract/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/skill-index.json",
"src/inferrsi/data/skill_library/workflows/claim-contract/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/comfy-ui.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/comfy-ui.json
index f2574de51..b022ae568 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/comfy-ui.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/comfy-ui.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/repositories/comfy-ui/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/repositories/comfy-ui/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/repositories/comfy-ui/LICENSE",
"src/inferrsi/data/skill_library/repositories/comfy-ui/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/competitor-scan.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/competitor-scan.json
index 06d75ab0e..50544030a 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/competitor-scan.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/competitor-scan.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/workflows/competitor-scan/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/workflows/competitor-scan/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/skill-index.json",
"src/inferrsi/data/skill_library/workflows/competitor-scan/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dcp-async-save-performance.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dcp-async-save-performance.json
index 357849cc7..a891567b4 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dcp-async-save-performance.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dcp-async-save-performance.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/SKILL.md",
"src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dense-performance-evidence.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dense-performance-evidence.json
index 4faa1834e..3a5024c0d 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dense-performance-evidence.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dense-performance-evidence.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/methods/dense-performance-evidence/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/methods/dense-performance-evidence/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/methods/dense-performance-evidence/SKILL.md",
"src/inferrsi/data/skill_library/methods/dense-performance-evidence/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diagnose-vllm-admission-pressure.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diagnose-vllm-admission-pressure.json
index 9cfacf268..4033ccc66 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diagnose-vllm-admission-pressure.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diagnose-vllm-admission-pressure.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/SKILL.md",
"src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diffusers-invocation-state.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diffusers-invocation-state.json
index 4edb012d7..0286c16cf 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diffusers-invocation-state.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diffusers-invocation-state.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/SKILL.md",
"src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/draft-acceptance-issue.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/draft-acceptance-issue.json
index 7bdaf64fc..7c39bde78 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/draft-acceptance-issue.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/draft-acceptance-issue.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/workflows/draft-acceptance-issue/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/workflows/draft-acceptance-issue/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/skill-index.json",
"src/inferrsi/data/skill_library/workflows/draft-acceptance-issue/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dwarfstar.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dwarfstar.json
index d814af8a8..e50958232 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dwarfstar.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dwarfstar.json
@@ -25,8 +25,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/repositories/dwarfstar/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/repositories/dwarfstar/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/repositories/dwarfstar/LICENSE",
"src/inferrsi/data/skill_library/repositories/dwarfstar/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/evaluate-elastic-serving-placement.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/evaluate-elastic-serving-placement.json
index ae3161a5d..b3929362e 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/evaluate-elastic-serving-placement.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/evaluate-elastic-serving-placement.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/methods/evaluate-elastic-serving-placement/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/methods/evaluate-elastic-serving-placement/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/methods/evaluate-elastic-serving-placement/SKILL.md",
"src/inferrsi/data/skill_library/methods/evaluate-elastic-serving-placement/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/generation-cache-lifecycle.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/generation-cache-lifecycle.json
index 7018b6ebd..681117f60 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/generation-cache-lifecycle.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/generation-cache-lifecycle.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/SKILL.md",
"src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-memory-management.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-memory-management.json
index 3d892ae4c..0e0eee119 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-memory-management.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-memory-management.json
@@ -30,8 +30,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/hardware/host/host-memory-management/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/hardware/host/host-memory-management/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/hardware/host/host-memory-management/SKILL.md",
"src/inferrsi/data/skill_library/hardware/host/host-memory-management/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-overhead-optimization.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-overhead-optimization.json
index 8d5b691b8..6ce57fba4 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-overhead-optimization.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-overhead-optimization.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/hardware/host/host-overhead-optimization/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/hardware/host/host-overhead-optimization/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/hardware/host/host-overhead-optimization/SKILL.md",
"src/inferrsi/data/skill_library/hardware/host/host-overhead-optimization/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/husky.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/husky.json
index 0945d60d3..05a8f448a 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/husky.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/husky.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/repositories/husky/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/repositories/husky/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/repositories/husky/SKILL.md",
"src/inferrsi/data/skill_library/repositories/husky/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-benchmark-design.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-benchmark-design.json
index d62f4ccd9..415d03961 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-benchmark-design.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-benchmark-design.json
@@ -30,8 +30,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/methods/inference-benchmark-design/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/methods/inference-benchmark-design/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/methods/inference-benchmark-design/SKILL.md",
"src/inferrsi/data/skill_library/methods/inference-benchmark-design/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-optimization-casebook.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-optimization-casebook.json
index 144785d86..ab8a80244 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-optimization-casebook.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-optimization-casebook.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/methods/inference-optimization-casebook/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/methods/inference-optimization-casebook/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/methods/inference-optimization-casebook/SKILL.md",
"src/inferrsi/data/skill_library/methods/inference-optimization-casebook/cases/admission-drain-occupancy.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kernel-writing-guide.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kernel-writing-guide.json
index 1ef4c67c5..8e8626e6a 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kernel-writing-guide.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kernel-writing-guide.json
@@ -39,8 +39,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/methods/kernel-writing-guide/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/methods/kernel-writing-guide/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/methods/kernel-writing-guide/SKILL.md",
"src/inferrsi/data/skill_library/methods/kernel-writing-guide/cases/causal-conv1d-dispatch-state.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kv-tier-management.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kv-tier-management.json
index d22656682..f09aca89f 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kv-tier-management.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kv-tier-management.json
@@ -24,8 +24,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/repositories/lmcache/kv-tier-management/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/repositories/lmcache/kv-tier-management/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/repositories/lmcache/kv-tier-management/SKILL.md",
"src/inferrsi/data/skill_library/repositories/lmcache/kv-tier-management/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lazy-evaluation-stream-lifetimes.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lazy-evaluation-stream-lifetimes.json
index 90c6e821d..baf15822e 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lazy-evaluation-stream-lifetimes.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lazy-evaluation-stream-lifetimes.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/SKILL.md",
"src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lily.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lily.json
index 1abdbd6a3..7bad127d3 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lily.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lily.json
@@ -25,8 +25,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/repositories/lily/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/repositories/lily/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/repositories/lily/LICENSE",
"src/inferrsi/data/skill_library/repositories/lily/NOTICE.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lithos-metal.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lithos-metal.json
index 3ea9ef1a7..73a0d5000 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lithos-metal.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lithos-metal.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/repositories/lithos-metal/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/repositories/lithos-metal/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/repositories/lithos-metal/LICENSE",
"src/inferrsi/data/skill_library/repositories/lithos-metal/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/local-dream.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/local-dream.json
index baf434174..7d3f5719f 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/local-dream.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/local-dream.json
@@ -29,8 +29,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/repositories/local-dream/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/repositories/local-dream/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/repositories/local-dream/SKILL.md",
"src/inferrsi/data/skill_library/repositories/local-dream/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mistral-rs.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mistral-rs.json
index 99ddd5f13..1a3eafe8e 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mistral-rs.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mistral-rs.json
@@ -26,8 +26,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/repositories/mistral-rs/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/repositories/mistral-rs/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/repositories/mistral-rs/LICENSE",
"src/inferrsi/data/skill_library/repositories/mistral-rs/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlc-llm.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlc-llm.json
index 0a5db775c..aa435a5c8 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlc-llm.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlc-llm.json
@@ -27,8 +27,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/repositories/mlc-llm/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/repositories/mlc-llm/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/repositories/mlc-llm/LICENSE",
"src/inferrsi/data/skill_library/repositories/mlc-llm/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlx-swift-lm.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlx-swift-lm.json
index 2120572f0..b6d2effd3 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlx-swift-lm.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlx-swift-lm.json
@@ -27,8 +27,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/repositories/mlx-swift-lm/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/repositories/mlx-swift-lm/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/repositories/mlx-swift-lm/LICENSE",
"src/inferrsi/data/skill_library/repositories/mlx-swift-lm/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/npu-skill-factory.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/npu-skill-factory.json
index 1feb8312a..646bd2108 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/npu-skill-factory.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/npu-skill-factory.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/workflows/npu-skill-factory/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/workflows/npu-skill-factory/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/skill-index.json",
"src/inferrsi/data/skill_library/workflows/npu-skill-factory/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/optimization-search-policy.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/optimization-search-policy.json
index 99382ef56..4ad246ad3 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/optimization-search-policy.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/optimization-search-policy.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/methods/optimization-search-policy/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/methods/optimization-search-policy/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/methods/optimization-search-policy/SKILL.md",
"src/inferrsi/data/skill_library/methods/optimization-search-policy/cases/sarathi-chunk-budget.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/os-process-placement.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/os-process-placement.json
index c07bff462..2e1a0ab3a 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/os-process-placement.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/os-process-placement.json
@@ -24,8 +24,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/hardware/host/os-process-placement/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/hardware/host/os-process-placement/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/hardware/host/os-process-placement/SKILL.md",
"src/inferrsi/data/skill_library/hardware/host/os-process-placement/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/performance-ceiling-analysis.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/performance-ceiling-analysis.json
index aeb500c1b..bb08f01fb 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/performance-ceiling-analysis.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/performance-ceiling-analysis.json
@@ -22,11 +22,12 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/SKILL.md",
"src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/metadata.json",
+ "src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/references/mooncake-prefix-reuse-feasibility.md",
"src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/references/papers-20261015/INTEGRATION_NOTES.md",
"src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/references/papers-20261015/MECHANISMS.md",
"src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/references/papers-20261015/README.md",
@@ -48,7 +49,7 @@
"src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/scripts/roofline.py",
"src/inferrsi/data/skill_library/skill-index.json"
],
- "content_sha256": "c5f78961c23e3c8d659f5675b522518ab06e1f2a8ea8bc74c5d428ffb41693d5"
+ "content_sha256": "4810761a3ab1e5e6e000060d9ad140a0e8892937486f2ab384d50868c22b0f11"
}
],
"index_type": "packaged-skill",
@@ -57,8 +58,9 @@
},
"checkout_ref": "checkout:performance-ceiling-analysis",
"source_files": {
- "src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/SKILL.md": "c5f78961c23e3c8d659f5675b522518ab06e1f2a8ea8bc74c5d428ffb41693d5",
- "src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/metadata.json": "60b5b24ead590b0fb222398fba47822bf6723690da683e256a450bb5c0eeb806",
+ "src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/SKILL.md": "4810761a3ab1e5e6e000060d9ad140a0e8892937486f2ab384d50868c22b0f11",
+ "src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/metadata.json": "3fcd2126846cf42ee77b1def39fc05ebfc54cf7428e0a6844b0ade1acf9f47ea",
+ "src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/references/mooncake-prefix-reuse-feasibility.md": "e64b62fded4490708414a898f540b0319c54d2b205fd5d060acaab419734d100",
"src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/references/papers-20261015/INTEGRATION_NOTES.md": "083e462e0a204dce599362d46712fcfb392d053c5cda56216e22e28d4b8fa0fa",
"src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/references/papers-20261015/MECHANISMS.md": "742c249082ceca230326f4407a40130fe624553b796ae5a157a8a4c15a7e1398",
"src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/references/papers-20261015/README.md": "17b2d1c7c14124d59ac8a0d5ad4200ad2803d991e784a1a98cd97188cc364999",
@@ -125,6 +127,19 @@
],
"type": "git",
"revision_precision": "prefix"
+ },
+ {
+ "type": "paper",
+ "id": "usenix:fast25-qin",
+ "version": "FAST '25 proceedings, February 25-27, 2025, pp. 155-170",
+ "url": "https://www.usenix.org/system/files/fast25-qin.pdf",
+ "content_sha256": "543e56c541523c5e807d2231370f96749b2294bc901266397a7265b18ae2c130",
+ "sections": [
+ "2.2, Table 1, Equations 1-2",
+ "3.1 and 5.3 (storage-scope limits only)"
+ ],
+ "readset": "references/mooncake-prefix-reuse-feasibility.md",
+ "license": "UNKNOWN"
}
],
"verification": {
@@ -151,9 +166,11 @@
],
"empirical_validation": "not_run",
"limitations": [
- "Source-derived candidate; no local efficacy validation."
+ "Source-derived candidate; no local efficacy validation.",
+ "Mooncake's prefix-reuse condition is an analytical approximation; current cache validity, placement, effective rates, overlap and SLO outcomes remain unverified."
],
"provenance_readsets": [
+ "references/mooncake-prefix-reuse-feasibility.md",
"references/papers-20261015/cases/decision-cases.md",
"references/papers-20261015/govern-optimization-experiments/procedure.md",
"references/papers-20261015/govern-optimization-experiments/references/source-notes.md",
@@ -187,7 +204,8 @@
"The imported status describes this index, not the original skill; its declared status is candidate.",
"Reading or indexing does not authorize execution, validate efficacy, promote a candidate or grant a license.",
"The backend categorizes the procedure's hardware dependency; it does not certify compatibility, and skills.discovery is a discovery scenario.",
- "Source-derived candidate; no local efficacy validation."
+ "Source-derived candidate; no local efficacy validation.",
+ "Mooncake's prefix-reuse condition is an analytical approximation; current cache validity, placement, effective rates, overlap and SLO outcomes remain unverified."
],
"verification": {
"locally_verified": false,
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/plan-activation-residency.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/plan-activation-residency.json
index 9693249f5..4dcb0dc4b 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/plan-activation-residency.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/plan-activation-residency.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/methods/plan-activation-residency/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/methods/plan-activation-residency/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/methods/plan-activation-residency/SKILL.md",
"src/inferrsi/data/skill_library/methods/plan-activation-residency/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/project-maya.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/project-maya.json
index 616aed8d2..4f2588d8f 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/project-maya.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/project-maya.json
@@ -25,8 +25,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/repositories/project-maya/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/repositories/project-maya/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/repositories/project-maya/LICENSE",
"src/inferrsi/data/skill_library/repositories/project-maya/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/repository-evolution-analysis.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/repository-evolution-analysis.json
index 84f9811fa..3f66326e2 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/repository-evolution-analysis.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/repository-evolution-analysis.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/methods/repository-evolution-analysis/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/methods/repository-evolution-analysis/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/methods/repository-evolution-analysis/SKILL.md",
"src/inferrsi/data/skill_library/methods/repository-evolution-analysis/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/rsi-evidence.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/rsi-evidence.json
index de8a66bfc..29b03048f 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/rsi-evidence.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/rsi-evidence.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/workflows/rsi-evidence/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/workflows/rsi-evidence/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/skill-index.json",
"src/inferrsi/data/skill_library/workflows/rsi-evidence/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/skill-evolution-playbook.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/skill-evolution-playbook.json
index 2ca4217a2..2929f3f3b 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/skill-evolution-playbook.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/skill-evolution-playbook.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/methods/skill-evolution-playbook/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/methods/skill-evolution-playbook/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/methods/skill-evolution-playbook/SKILL.md",
"src/inferrsi/data/skill_library/methods/skill-evolution-playbook/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/splash.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/splash.json
index 578955d20..2c920a600 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/splash.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/splash.json
@@ -25,8 +25,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/repositories/splash/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/repositories/splash/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/repositories/splash/LICENSE",
"src/inferrsi/data/skill_library/repositories/splash/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/storage-io-measurement.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/storage-io-measurement.json
index 4c2c81c2c..430810584 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/storage-io-measurement.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/storage-io-measurement.json
@@ -24,8 +24,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/hardware/host/storage-io-measurement/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/hardware/host/storage-io-measurement/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/hardware/host/storage-io-measurement/SKILL.md",
"src/inferrsi/data/skill_library/hardware/host/storage-io-measurement/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-input-pipeline.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-input-pipeline.json
index ede094a22..be9f0ab60 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-input-pipeline.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-input-pipeline.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/methods/training-input-pipeline/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/methods/training-input-pipeline/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/methods/training-input-pipeline/SKILL.md",
"src/inferrsi/data/skill_library/methods/training-input-pipeline/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-resource-budget.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-resource-budget.json
index e191fddb1..e3482f0f0 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-resource-budget.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-resource-budget.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/methods/training-resource-budget/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/methods/training-resource-budget/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/methods/training-resource-budget/SKILL.md",
"src/inferrsi/data/skill_library/methods/training-resource-budget/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/video-diffusion-optimization.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/video-diffusion-optimization.json
index c00ffdadd..24af5a0d3 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/video-diffusion-optimization.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/video-diffusion-optimization.json
@@ -26,8 +26,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/methods/video-diffusion-optimization/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/methods/video-diffusion-optimization/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/methods/video-diffusion-optimization/SKILL.md",
"src/inferrsi/data/skill_library/methods/video-diffusion-optimization/assets/campaign.template.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/vllm-metal.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/vllm-metal.json
index 135cea491..c6b525a1b 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/vllm-metal.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/vllm-metal.json
@@ -26,8 +26,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/repositories/vllm-metal/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/repositories/vllm-metal/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/repositories/vllm-metal/LICENSE",
"src/inferrsi/data/skill_library/repositories/vllm-metal/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/weight-loading-optimization.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/weight-loading-optimization.json
index 1a9752e29..3e3110f54 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/weight-loading-optimization.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/weight-loading-optimization.json
@@ -29,8 +29,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/methods/weight-loading-optimization/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/methods/weight-loading-optimization/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/methods/weight-loading-optimization/SKILL.md",
"src/inferrsi/data/skill_library/methods/weight-loading-optimization/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-afd-p2p-exchanges.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-afd-p2p-exchanges.json
index 3b456e047..b60cefdb7 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-afd-p2p-exchanges.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-afd-p2p-exchanges.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/SKILL.md",
"src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-flare-session-lifecycle.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-flare-session-lifecycle.json
index 943853aee..69b4d7bb3 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-flare-session-lifecycle.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-flare-session-lifecycle.json
@@ -29,8 +29,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/LICENSE",
"src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/diagnose-attention-bottlenecks.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/diagnose-attention-bottlenecks.json
index 38600a9a9..f8aca5f03 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/diagnose-attention-bottlenecks.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/diagnose-attention-bottlenecks.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/hardware/cuda/diagnose-attention-bottlenecks/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/hardware/cuda/diagnose-attention-bottlenecks/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/hardware/cuda/diagnose-attention-bottlenecks/SKILL.md",
"src/inferrsi/data/skill_library/hardware/cuda/diagnose-attention-bottlenecks/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/host-overhead-profiling.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/host-overhead-profiling.json
index 1e9737954..cf0b05595 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/host-overhead-profiling.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/host-overhead-profiling.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/hardware/cuda/host-overhead-profiling/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/hardware/cuda/host-overhead-profiling/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/hardware/cuda/host-overhead-profiling/SKILL.md",
"src/inferrsi/data/skill_library/hardware/cuda/host-overhead-profiling/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/reason-hopper-kernels.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/reason-hopper-kernels.json
index 47f5ae5c1..731bc6296 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/reason-hopper-kernels.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/reason-hopper-kernels.json
@@ -36,8 +36,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/224f6534dc5596ff78fd2f8329d73658c6bc78da/src/inferrsi/data/skill_library/hardware/cuda/reason-hopper-kernels/SKILL.md",
- "commit": "224f6534dc5596ff78fd2f8329d73658c6bc78da",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/hardware/cuda/reason-hopper-kernels/SKILL.md",
+ "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
"paths": [
"src/inferrsi/data/skill_library/hardware/cuda/reason-hopper-kernels/SKILL.md",
"src/inferrsi/data/skill_library/hardware/cuda/reason-hopper-kernels/metadata.json",
From ed5d9072ed416bf2c8c78eaa7ad89875cab00c23 Mon Sep 17 00:00:00 2001
From: Dong1017 <55723162+Dong1017@users.noreply.github.com>
Date: Sun, 1 Nov 2026 18:46:49 +0800
Subject: [PATCH 07/11] test: retain independent MLX lifecycle discovery
coverage
Keep query propagation checks and separately require the canonical MLX lifecycle resource in a scoped research query, including installed CLI discovery.
Agent-assisted-by: Codex / gpt-6-astra [effort=xhigh] (author)
Signed-off-by: Dong1017 <55723162+Dong1017@users.noreply.github.com>
---
tests/check_installed.py | 15 +++++++++++++++
tests/test_research.py | 16 ++++++++++++++++
2 files changed, 31 insertions(+)
diff --git a/tests/check_installed.py b/tests/check_installed.py
index e15715a06..a8772b76c 100644
--- a/tests/check_installed.py
+++ b/tests/check_installed.py
@@ -309,6 +309,21 @@
assert set(research["profile"]["knowledge_ids"]) <= {
item["record"]["id"] for item in research["items"]
}
+output = io.StringIO()
+with contextlib.redirect_stdout(output):
+ assert main([
+ "knowledge", "research", "MiniMax-H3", "--framework", "mlx",
+ "--mechanism", "lazy evaluation", "--repo-type", "inferrsi",
+ "--scenario", "skills.discovery", "--limit", "12",
+ ]) == 0
+mlx_research = json.loads(output.getvalue())
+mlx_lifecycle = next(
+ item for item in mlx_research["items"]
+ if item["record"]["id"] == "inferrsi.common.skills.lazy_evaluation_stream_lifetimes"
+)
+assert mlx_lifecycle["record"]["skill_documents"]["original"] == mlx_document
+assert {"group": "framework", "query": "mlx"} in mlx_lifecycle["matched_queries"]
+assert {"group": "mechanism", "query": "lazy evaluation"} in mlx_lifecycle["matched_queries"]
assert not inference_modules & sys.modules.keys()
print(f"Installed package: {inferrsi.__file__}; {len(records)} source-grounded records available.")
diff --git a/tests/test_research.py b/tests/test_research.py
index 3da4ba48e..31931da7b 100644
--- a/tests/test_research.py
+++ b/tests/test_research.py
@@ -54,6 +54,22 @@ def test_h3_retrieves_framework_and_task_knowledge_with_source_boundaries():
assert result["profile"]["output_modalities"] == ["video", "audio"]
+def test_scoped_mlx_research_retains_the_lifecycle_resource():
+ result = research_knowledge(
+ "MiniMax-H3", frameworks=["mlx"], mechanisms=["lazy evaluation"],
+ repo="inferrsi", scenario="skills.discovery", limit=12,
+ )
+ lifecycle = next(
+ item for item in result["items"]
+ if item["record"]["id"] == "inferrsi.common.skills.lazy_evaluation_stream_lifetimes"
+ )
+ assert lifecycle["record"]["skill_documents"]["original"] == (
+ "data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/SKILL.md"
+ )
+ assert {"group": "framework", "query": "mlx"} in lifecycle["matched_queries"]
+ assert {"group": "mechanism", "query": "lazy evaluation"} in lifecycle["matched_queries"]
+
+
def test_independent_queries_are_unioned_fairly_and_exact_search_is_unchanged(monkeypatch):
records = [
{"id": name, "repo": "vllm_omni", "backend": "common", "scenario": "diffusion.*",
From 9e9f6b7b14eee7b71fc472617b53053f542d3772 Mon Sep 17 00:00:00 2001
From: Dong1017 <55723162+Dong1017@users.noreply.github.com>
Date: Sun, 1 Nov 2026 22:29:13 +0800
Subject: [PATCH 08/11] docs: regenerate combined skill discovery projections
Keep the canonical combined-source snapshot reachable in the preceding commit.
Agent-assisted-by: Codex / gpt-6-astra [effort=xhigh] (author)
Signed-off-by: Dong1017 <55723162+Dong1017@users.noreply.github.com>
---
docs/skill-coverage.md | 38 ++++----
.../repo_packs/inferrsi/knowledge/README.md | 4 +-
.../ascend-aclgraph-capture-replay.json | 4 +-
.../ascend/skills/ascend-device-health.json | 4 +-
.../skills/ascend-distributed-parallel.json | 4 +-
.../ascend/skills/ascend-graph-mode.json | 4 +-
.../skills/ascend-model-adaptation.json | 4 +-
.../skills/ascend-operator-profiling.json | 4 +-
.../ascend/skills/ascend-operator-tuning.json | 4 +-
.../skills/ascend-optimization-casebook.json | 4 +-
.../skills/ascend-performance-analysis.json | 4 +-
.../ascend-video-diffusion-optimization.json | 4 +-
.../ascend/skills/ascendc-tiling-design.json | 4 +-
.../skills/cann-attribution-overlap.json | 4 +-
.../torchnpu-device-memory-lifecycle.json | 4 +-
...torchnpu-pinned-host-memory-lifecycle.json | 4 +-
.../skills/assess-serving-preemption.json | 49 ++++++++--
.../assess-split-backward-schedule.json | 47 ++++++++--
.../skills/audit-fixed-workload-claim.json | 45 +++++++--
.../audit-h3-metal-fasth3-semantics.json | 4 +-
.../skills/audit-h3-runtime-residency.json | 4 +-
.../skills/audit-llama-runtime-state.json | 4 +-
.../audit-recurrent-kv-finalization.json | 4 +-
.../common/skills/audit-sd-cpp-residency.json | 4 +-
.../audit-sglang-diffusion-lifetimes.json | 4 +-
.../skills/audit-vllm-kv-lifetimes.json | 4 +-
.../bound-windowed-attention-replay.json | 4 +-
.../budget-speculative-verification.json | 4 +-
.../skills/calibrate-serving-simulation.json | 47 ++++++++--
.../skills/cgroup-resource-controls.json | 4 +-
.../checkpoint-recovery-consistency.json | 51 +++++++++--
.../skills/choose-parallel-training-plan.json | 47 ++++++++--
.../skills/cite-check-downstream-pr.json | 4 +-
.../common/skills/claim-contract.json | 4 +-
.../knowledge/common/skills/comfy-ui.json | 4 +-
.../common/skills/competitor-scan.json | 4 +-
.../skills/dcp-async-save-performance.json | 4 +-
.../skills/dense-performance-evidence.json | 47 ++++++++--
.../diagnose-vllm-admission-pressure.json | 4 +-
.../skills/diffusers-invocation-state.json | 4 +-
.../common/skills/draft-acceptance-issue.json | 4 +-
.../knowledge/common/skills/dwarfstar.json | 4 +-
.../evaluate-elastic-serving-placement.json | 47 ++++++++--
.../skills/generation-cache-lifecycle.json | 4 +-
.../common/skills/host-memory-management.json | 4 +-
.../skills/host-overhead-optimization.json | 4 +-
.../knowledge/common/skills/husky.json | 4 +-
.../skills/inference-benchmark-design.json | 51 +++++++++--
.../inference-optimization-casebook.json | 49 ++++++++--
.../common/skills/kernel-writing-guide.json | 69 ++++++++++++--
.../common/skills/kv-tier-management.json | 4 +-
.../lazy-evaluation-stream-lifetimes.json | 4 +-
.../knowledge/common/skills/lily.json | 4 +-
.../knowledge/common/skills/lithos-metal.json | 4 +-
.../knowledge/common/skills/local-dream.json | 4 +-
.../knowledge/common/skills/mistral-rs.json | 4 +-
.../knowledge/common/skills/mlc-llm.json | 4 +-
.../knowledge/common/skills/mlx-swift-lm.json | 4 +-
.../common/skills/npu-skill-factory.json | 4 +-
.../skills/optimization-search-policy.json | 91 +++++++++++++------
.../common/skills/os-process-placement.json | 4 +-
.../skills/performance-ceiling-analysis.json | 4 +-
.../skills/plan-activation-residency.json | 49 ++++++++--
.../knowledge/common/skills/project-maya.json | 4 +-
.../skills/repository-evolution-analysis.json | 4 +-
.../knowledge/common/skills/rsi-evidence.json | 4 +-
.../skills/skill-evolution-playbook.json | 51 +++++++++--
.../knowledge/common/skills/splash.json | 4 +-
.../common/skills/storage-io-measurement.json | 4 +-
.../skills/training-input-pipeline.json | 51 +++++++++--
.../skills/training-resource-budget.json | 51 +++++++++--
.../skills/video-diffusion-optimization.json | 4 +-
.../knowledge/common/skills/vllm-metal.json | 4 +-
.../skills/weight-loading-optimization.json | 59 ++++++++++--
.../cuda/skills/audit-afd-p2p-exchanges.json | 4 +-
.../skills/audit-flare-session-lifecycle.json | 4 +-
.../diagnose-attention-bottlenecks.json | 4 +-
.../cuda/skills/host-overhead-profiling.json | 4 +-
.../cuda/skills/reason-hopper-kernels.json | 4 +-
79 files changed, 920 insertions(+), 263 deletions(-)
diff --git a/docs/skill-coverage.md b/docs/skill-coverage.md
index 48fab8258..9f8223d37 100644
--- a/docs/skill-coverage.md
+++ b/docs/skill-coverage.md
@@ -4,7 +4,7 @@ Generated from the registry, the skill sidecars and the two maintained
matrices. **Do not edit by hand**; regenerate after any registry, skill,
matrix or frozen-plan change.
-- Input digest: `sha256:92e3d3809e7c4ad1549be166745e7e7b0b3b9a65d97e3d49c2172c8e0bec878e`
+- Input digest: `sha256:37e7219c0b159ecbdf593d1bf903ce83a7de56d618dd22d0feecd83841ea4ecd`
- Regenerate: `python tools/build_skill_coverage.py`
- Verify without writing: `python tools/build_skill_coverage.py --check`
@@ -118,26 +118,26 @@ optimization is fully covered or that any procedure has measured benefit.
| Skill | Backend | Declared task | Applicability | Conditions | SearchSpace layer / stage | Reference Matrix | State | Gaps |
|---|---|---|---|---|---|---|---|---|
-| [assess-serving-preemption](../src/inferrsi/data/skill_library/methods/assess-serving-preemption/SKILL.md) | common | Assess whether LLM serving queue delays justify token-boundary priority preemption, accounting for suspended KV, host ... | **unknown** | — | 1 Accelerator · Diagnose / Configure / Optimize / Verify
2 RAM/Host · Diagnose / Configure / Optimize / Verify | 1 · Systems sources (Orca — OSDI 2022 proceedings) | candidate · not_run · 1 readset(s) | no-applicability, no-eval |
-| [assess-split-backward-schedule](../src/inferrsi/data/skill_library/methods/assess-split-backward-schedule/SKILL.md) | common | Evaluate memory-capped split-backward and V-shaped pipeline scheduling proposals, preserving synchronous update ... | **unknown** | — | 1 Accelerator · Diagnose / Configure / Optimize / Verify | 1 · Systems sources (Zero Bubble — [2401.10241v1](https://arxiv.org/abs/2401.10241v1)) | candidate · not_run · 1 readset(s) | no-applicability, no-eval |
-| [audit-fixed-workload-claim](../src/inferrsi/data/skill_library/methods/audit-fixed-workload-claim/SKILL.md) | common | Audit a performance claim from supplied fixed-workload reports and an explicit measurement contract. Check aggregation ... | **unknown** | — | — | — | candidate · not_run · 0 readset(s) | no-SearchSpace-layer, no-Reference-Matrix-row, no-applicability, no-eval |
-| [calibrate-serving-simulation](../src/inferrsi/data/skill_library/methods/calibrate-serving-simulation/SKILL.md) | common | Build an evidence-gated serving-configuration shortlist from a profiled simulator while separating empirical ... | **unknown** | — | 1 Accelerator · Diagnose / Verify | 1 · Systems sources (Vidur — [2405.05465v2](https://arxiv.org/abs/2405.05465v2)) | candidate · not_run · 1 readset(s) | no-applicability, no-eval |
-| [checkpoint-recovery-consistency](../src/inferrsi/data/skill_library/methods/checkpoint-recovery-consistency/SKILL.md) | common | Assess distributed checkpoint recovery using DLRover memory agreement, persistence completion and replica transport ... | **unknown** | — | 2 RAM/Host · Configure / Verify
3 SSD · Configure / Verify
4 Network · Diagnose / Configure / Verify
5 OS · Diagnose / Configure | 1 · Systems sources (C3 / DLRover FlashCheckpoint @ `e35b1ad90e2928ca6bacf5d22117f71356 ...)
1 · Systems sources (C4 / NVIDIA Resiliency Extension @ ...) | candidate · not_run · 1 readset(s) | no-applicability, no-eval |
-| [choose-parallel-training-plan](../src/inferrsi/data/skill_library/methods/choose-parallel-training-plan/SKILL.md) | common | Compare stage, device-mesh and operator-sharding candidates for static distributed training while accounting for ... | **unknown** | — | 1 Accelerator · Diagnose / Configure / Optimize / Verify
4 Network · Diagnose / Configure / Optimize / Verify | 1 · Systems sources (Alpa — [2201.12023v3](https://arxiv.org/abs/2201.12023v3)) | candidate · not_run · 1 readset(s) | no-applicability, no-eval |
-| [dense-performance-evidence](../src/inferrsi/data/skill_library/methods/dense-performance-evidence/SKILL.md) | common | Taxonomy of performance evidence an optimization agent needs, ordered by collection cost and diagnostic power. Maps ... | **unknown** | — | 1 Accelerator · Observe / Profile / Verify
2 RAM/Host · Verify
4 Network · Verify | 1 · Systems sources (NVIDIA/TensorRT-LLM @ f33e0e5fd)
2 · Methodology and corpus references (JiusiServe/vllm-evolve @ 43cc82c)
2 · Methodology and corpus references (microsoft/SkillOpt @ 343db22) | candidate · not_run · 0 readset(s) | no-applicability |
-| [evaluate-elastic-serving-placement](../src/inferrsi/data/skill_library/methods/evaluate-elastic-serving-placement/SKILL.md) | common | Compare migrating an LLM request between model replicas with changing its sequence-parallel execution group, including ... | **unknown** | — | 1 Accelerator · Diagnose / Configure / Optimize / Verify
2 RAM/Host · Diagnose / Configure / Optimize / Verify
4 Network · Diagnose / Configure / Optimize / Verify | 1 · Systems sources (Llumnix — [2406.03243v1](https://arxiv.org/abs/2406.03243v1)) | candidate · not_run · 1 readset(s) | no-applicability, no-eval |
-| [inference-benchmark-design](../src/inferrsi/data/skill_library/methods/inference-benchmark-design/SKILL.md) | common | Select task quality gates, reference parity checks and performance workloads for text, agentic coding, work agents ... | **unknown** | — | — | 2 · Methodology and corpus references (JiusiServe/vllm-evolve @ 43cc82c)
Task-specific benchmark selection (issue #60) ([Issue #60 report](https://github.com/InferMatrixLab/InferRSI/issues/6 ...) | candidate · not_run · 2 readset(s) | no-SearchSpace-layer, no-applicability |
-| [inference-optimization-casebook](../src/inferrsi/data/skill_library/methods/inference-optimization-casebook/SKILL.md) | common | How to structure and use a casebook of past optimization decisions as a decision-support tool. Extracts the casebook ... | **unknown** | — | 1 Accelerator · Optimize | 1 · Systems sources (NVIDIA/TensorRT-LLM @ f33e0e5fd) | candidate · not_run · 0 readset(s) | no-applicability, no-eval |
-| [kernel-writing-guide](../src/inferrsi/data/skill_library/methods/kernel-writing-guide/SKILL.md) | common | Decision guide for writing custom inference kernels: when to write one, which framework to choose (Triton, CuTe DSL ... | **unknown** | — | 1 Accelerator · Optimize / Diagnose / Configure / Verify | 1 · Systems sources (Dao-AILab/causal-conv1d @ `cd81f0413cad2fc1e6f17e785ac39f59aae690cd` ...)
1 · Systems sources (NVIDIA/TensorRT-LLM @ f33e0e5fd)
1 · Systems sources (huggingface/transformers @ `0dad7b822255a0ae261ec45ae937371e859ffd1a` ...)
1 · Systems sources (mindspore-ai/akg @ `5aa15f3fe4d856eb60cd3b27447e04f5f1fb4dff` ...)
1 · Systems sources (sgl-project/sglang @ `16176278d0180f1a1a578e53e6eef432577876c2` ...)
1 · Systems sources (vllm-project/vllm @ `537a79f0fc5ba564d2bef43d54915b2f80f9f77d` (root ...) | candidate · not_run · 7 readset(s) | no-applicability, no-eval |
-| [optimization-search-policy](../src/inferrsi/data/skill_library/methods/optimization-search-policy/SKILL.md) | common | Policy for how an optimization agent should select, order, and stop optimization experiments. Covers candidate ... | **unknown** | — | 1 Accelerator · Optimize | 2 · Methodology and corpus references (JiusiServe/vllm-evolve @ 43cc82c)
2 · Methodology and corpus references (RRSI — arXiv 2609.24972v3)
2 · Methodology and corpus references (arXiv 2608.01867 (CRISP))
2 · Methodology and corpus references (microsoft/SkillOpt @ 343db22) | candidate · not_run · 9 readset(s) | no-applicability, no-eval |
+| [assess-serving-preemption](../src/inferrsi/data/skill_library/methods/assess-serving-preemption/SKILL.md) | common | Assess whether LLM serving queue delays justify token-boundary priority preemption, accounting for suspended KV, host ... | Diagnosing whether LLM serving queue delay justifies token-boundary priority preemption ... | failure_modes, input_contract, produced_artifact, refusal_point
4 advice | 1 Accelerator · Diagnose / Configure / Optimize / Verify
2 RAM/Host · Diagnose / Configure / Optimize / Verify | 1 · Systems sources (Orca — OSDI 2022 proceedings) | candidate · not_run · 1 readset(s) | no-eval |
+| [assess-split-backward-schedule](../src/inferrsi/data/skill_library/methods/assess-split-backward-schedule/SKILL.md) | common | Evaluate memory-capped split-backward and V-shaped pipeline scheduling proposals, preserving synchronous update ... | Assessing memory-capped split-backward and V-shaped pipeline schedule proposals when a ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator · Diagnose / Configure / Optimize / Verify | 1 · Systems sources (Zero Bubble — [2401.10241v1](https://arxiv.org/abs/2401.10241v1)) | candidate · not_run · 1 readset(s) | no-eval |
+| [audit-fixed-workload-claim](../src/inferrsi/data/skill_library/methods/audit-fixed-workload-claim/SKILL.md) | common | Audit a performance claim from supplied fixed-workload reports and an explicit measurement contract. Check aggregation ... | Auditing a performance claim against supplied fixed-workload reports and an explicit ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | — | — | candidate · not_run · 0 readset(s) | no-SearchSpace-layer, no-Reference-Matrix-row, no-eval |
+| [calibrate-serving-simulation](../src/inferrsi/data/skill_library/methods/calibrate-serving-simulation/SKILL.md) | common | Build an evidence-gated serving-configuration shortlist from a profiled simulator while separating empirical ... | Building an evidence-gated serving-configuration shortlist when a profiled simulator or ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator · Diagnose / Verify | 1 · Systems sources (Vidur — [2405.05465v2](https://arxiv.org/abs/2405.05465v2)) | candidate · not_run · 1 readset(s) | no-eval |
+| [checkpoint-recovery-consistency](../src/inferrsi/data/skill_library/methods/checkpoint-recovery-consistency/SKILL.md) | common | Assess distributed checkpoint recovery using DLRover memory agreement, persistence completion and replica transport ... | Assessing distributed checkpoint recovery through DLRover memory agreement, persistence ... | failure_modes, input_contract, produced_artifact, refusal_point
4 advice | 2 RAM/Host · Configure / Verify
3 SSD · Configure / Verify
4 Network · Diagnose / Configure / Verify
5 OS · Diagnose / Configure | 1 · Systems sources (C3 / DLRover FlashCheckpoint @ `e35b1ad90e2928ca6bacf5d22117f71356 ...)
1 · Systems sources (C4 / NVIDIA Resiliency Extension @ ...) | candidate · not_run · 1 readset(s) | no-eval |
+| [choose-parallel-training-plan](../src/inferrsi/data/skill_library/methods/choose-parallel-training-plan/SKILL.md) | common | Compare stage, device-mesh and operator-sharding candidates for static distributed training while accounting for ... | Comparing stage, device-mesh and operator-sharding candidates for static distributed ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator · Diagnose / Configure / Optimize / Verify
4 Network · Diagnose / Configure / Optimize / Verify | 1 · Systems sources (Alpa — [2201.12023v3](https://arxiv.org/abs/2201.12023v3)) | candidate · not_run · 1 readset(s) | no-eval |
+| [dense-performance-evidence](../src/inferrsi/data/skill_library/methods/dense-performance-evidence/SKILL.md) | common | Taxonomy of performance evidence an optimization agent needs, ordered by collection cost and diagnostic power. Maps ... | Deciding which performance evidence to collect next and which claim each evidence level ... | failure_modes, input_contract, produced_artifact, refusal_point
4 advice | 1 Accelerator · Observe / Profile / Verify
2 RAM/Host · Verify
4 Network · Verify | 1 · Systems sources (NVIDIA/TensorRT-LLM @ f33e0e5fd)
2 · Methodology and corpus references (JiusiServe/vllm-evolve @ 43cc82c)
2 · Methodology and corpus references (microsoft/SkillOpt @ 343db22) | candidate · not_run · 0 readset(s) | — |
+| [evaluate-elastic-serving-placement](../src/inferrsi/data/skill_library/methods/evaluate-elastic-serving-placement/SKILL.md) | common | Compare migrating an LLM request between model replicas with changing its sequence-parallel execution group, including ... | Comparing migration of a running LLM request between compatible model replicas with ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator · Diagnose / Configure / Optimize / Verify
2 RAM/Host · Diagnose / Configure / Optimize / Verify
4 Network · Diagnose / Configure / Optimize / Verify | 1 · Systems sources (Llumnix — [2406.03243v1](https://arxiv.org/abs/2406.03243v1)) | candidate · not_run · 1 readset(s) | no-eval |
+| [inference-benchmark-design](../src/inferrsi/data/skill_library/methods/inference-benchmark-design/SKILL.md) | common | Select task quality gates, reference parity checks and performance workloads for text, agentic coding, work agents ... | Selecting task quality gates, reference parity checks and performance workloads for text ... | failure_modes, input_contract, produced_artifact, refusal_point
4 advice | — | 2 · Methodology and corpus references (JiusiServe/vllm-evolve @ 43cc82c)
Task-specific benchmark selection (issue #60) ([Issue #60 report](https://github.com/InferMatrixLab/InferRSI/issues/6 ...) | candidate · not_run · 2 readset(s) | no-SearchSpace-layer |
+| [inference-optimization-casebook](../src/inferrsi/data/skill_library/methods/inference-optimization-casebook/SKILL.md) | common | How to structure and use a casebook of past optimization decisions as a decision-support tool. Extracts the casebook ... | Structuring and querying a casebook of past optimization decisions as decision support ... | failure_modes, input_contract, produced_artifact, refusal_point
4 advice | 1 Accelerator · Optimize | 1 · Systems sources (NVIDIA/TensorRT-LLM @ f33e0e5fd) | candidate · not_run · 0 readset(s) | no-eval |
+| [kernel-writing-guide](../src/inferrsi/data/skill_library/methods/kernel-writing-guide/SKILL.md) | common | Decision guide for writing custom inference kernels: when to write one, which framework to choose (Triton, CuTe DSL ... | Deciding whether to write a custom inference kernel, routing a kernel pattern to Triton ... | failure_modes, input_contract, produced_artifact, refusal_point
6 advice | 1 Accelerator · Optimize / Diagnose / Configure / Verify | 1 · Systems sources (Dao-AILab/causal-conv1d @ `cd81f0413cad2fc1e6f17e785ac39f59aae690cd` ...)
1 · Systems sources (NVIDIA/TensorRT-LLM @ f33e0e5fd)
1 · Systems sources (huggingface/transformers @ `0dad7b822255a0ae261ec45ae937371e859ffd1a` ...)
1 · Systems sources (mindspore-ai/akg @ `5aa15f3fe4d856eb60cd3b27447e04f5f1fb4dff` ...)
1 · Systems sources (sgl-project/sglang @ `16176278d0180f1a1a578e53e6eef432577876c2` ...)
1 · Systems sources (vllm-project/vllm @ `537a79f0fc5ba564d2bef43d54915b2f80f9f77d` (root ...) | candidate · not_run · 7 readset(s) | no-eval |
+| [optimization-search-policy](../src/inferrsi/data/skill_library/methods/optimization-search-policy/SKILL.md) | common | Policy for how an optimization agent should select, order, and stop optimization experiments. Covers candidate ... | Selecting, ordering and stopping optimization experiments: candidate generation ... | failure_modes, input_contract, produced_artifact, refusal_point
4 advice | 1 Accelerator · Optimize | 2 · Methodology and corpus references (JiusiServe/vllm-evolve @ 43cc82c)
2 · Methodology and corpus references (RRSI — arXiv 2609.24972v3)
2 · Methodology and corpus references (arXiv 2608.01867 (CRISP))
2 · Methodology and corpus references (microsoft/SkillOpt @ 343db22) | candidate · not_run · 9 readset(s) | no-eval |
| [performance-ceiling-analysis](../src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/SKILL.md) | common | Estimate optimization headroom before modifying an inference system. Produces a three-level ceiling model (analytical ... | inference serving systems (LLM, diffusion, audio) | — | 1 Accelerator · Diagnose / Verify
5 OS · Configure | 1 · Systems sources (Mooncake — FAST '25 proceedings, pp. 155-170)
1 · Systems sources (NVIDIA/TensorRT-LLM @ f33e0e5fd)
1 · Systems sources (ebugger/Empirical-Roofline-Toolkit @ 0771fa9)
1 · Systems sources (hahnyuan/LLM-Viewer @ 1893e4b) | candidate · not_run · 17 readset(s) | no-eval |
-| [plan-activation-residency](../src/inferrsi/data/skill_library/methods/plan-activation-residency/SKILL.md) | common | Propose training activation and gradient memory reductions by distinguishing lifetime reordering, address packing and ... | **unknown** | — | 1 Accelerator · Diagnose / Optimize / Verify | 1 · Systems sources (Checkmate — [1910.02653v3](https://arxiv.org/abs/1910.02653v3)) | candidate · not_run · 1 readset(s) | no-applicability, no-eval |
+| [plan-activation-residency](../src/inferrsi/data/skill_library/methods/plan-activation-residency/SKILL.md) | common | Propose training activation and gradient memory reductions by distinguishing lifetime reordering, address packing and ... | Proposing training activation and gradient memory reductions when pressure appears to ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator · Diagnose / Optimize / Verify | 1 · Systems sources (Checkmate — [1910.02653v3](https://arxiv.org/abs/1910.02653v3)) | candidate · not_run · 1 readset(s) | no-eval |
| [repository-evolution-analysis](../src/inferrsi/data/skill_library/methods/repository-evolution-analysis/SKILL.md) | common | Analyze a repository's commit history to build a capability-evolution model: what the system can do, how it got there ... | any software repository with a recoverable git history | — | — | 2 · Methodology and corpus references (LearningToCommit/LearningToCommit @ 7e5a1d7)
2 · Methodology and corpus references (yetone/magpie @ 77800032) | candidate · not_run · 0 readset(s) | no-SearchSpace-layer, no-eval |
-| [skill-evolution-playbook](../src/inferrsi/data/skill_library/methods/skill-evolution-playbook/SKILL.md) | common | Practical design choices for evolving agent skills from execution traces: when one trajectory should update a skill ... | **unknown** | — | — | 2 · Methodology and corpus references (Qwen-Applications/Trace2Skill @ 3d0b52a)
2 · Methodology and corpus references (microsoft/SkillOpt @ 343db22) | candidate · not_run · 0 readset(s) | no-SearchSpace-layer, no-applicability, no-eval |
-| [training-input-pipeline](../src/inferrsi/data/skill_library/methods/training-input-pipeline/SKILL.md) | common | Diagnose training input stalls and choose bounded queue, shard-cache and cache-ownership changes from DALI, Mosaic ... | **unknown** | — | 1 Accelerator · Observe / Configure / Optimize
2 RAM/Host · Observe / Diagnose / Configure / Optimize
3 SSD · Diagnose / Configure / Optimize
4 Network · Configure / Optimize
5 OS · Configure / Optimize | 1 · Systems sources (B1 / NVIDIA DALI @ `6fe47c48d3b0ca335f5e1afd3c77c750b2c7e041` ...)
1 · Systems sources (B2 / Mosaic Streaming @ `d99bf9c7cdf2dd4ed62a4960ed35270b93337a5f` ...)
1 · Systems sources (B3 / FFCV @ `3a12966b3afe3a81733a732e633317d747bfaac7` ...) | candidate · not_run · 1 readset(s) | no-applicability, no-eval |
-| [training-resource-budget](../src/inferrsi/data/skill_library/methods/training-resource-budget/SKILL.md) | common | Choose training offload and resource-budget changes using pinned DeepSpeed configuration, attributed expert diagnosis ... | **unknown** | — | 1 Accelerator · Diagnose / Configure / Optimize
2 RAM/Host · Observe / Diagnose / Configure / Optimize / Verify
3 SSD · Configure / Optimize
4 Network · Observe / Diagnose / Verify
5 OS · Diagnose / Configure | 1 · Systems sources (CR1 / DeepSpeed ZeRO-Infinity @ `5f266e26d216a705d9c64cac88fb813ce ...)
1 · Systems sources (CR5 / Ascend MindSpeed adaptive memory — historical `core_r0.8.0` ...)
1 · Systems sources (ZeRO-Offload — [2101.06840v1](https://arxiv.org/abs/2101.06840v1))
2 · Methodology and corpus references (CR6 / Stas Bekman ML Engineering Open Book @ ...) | candidate · not_run · 9 readset(s) | no-applicability, no-eval |
+| [skill-evolution-playbook](../src/inferrsi/data/skill_library/methods/skill-evolution-playbook/SKILL.md) | common | Practical design choices for evolving agent skills from execution traces: when one trajectory should update a skill ... | Design choices for evolving agent skills from execution traces, drawn from Trace2Skill ... | failure_modes, input_contract, produced_artifact, refusal_point
4 advice | — | 2 · Methodology and corpus references (Qwen-Applications/Trace2Skill @ 3d0b52a)
2 · Methodology and corpus references (microsoft/SkillOpt @ 343db22) | candidate · not_run · 0 readset(s) | no-SearchSpace-layer, no-eval |
+| [training-input-pipeline](../src/inferrsi/data/skill_library/methods/training-input-pipeline/SKILL.md) | common | Diagnose training input stalls and choose bounded queue, shard-cache and cache-ownership changes from DALI, Mosaic ... | Diagnosing training input stalls and choosing bounded queue, shard-cache and ... | failure_modes, input_contract, produced_artifact, refusal_point
4 advice | 1 Accelerator · Observe / Configure / Optimize
2 RAM/Host · Observe / Diagnose / Configure / Optimize
3 SSD · Diagnose / Configure / Optimize
4 Network · Configure / Optimize
5 OS · Configure / Optimize | 1 · Systems sources (B1 / NVIDIA DALI @ `6fe47c48d3b0ca335f5e1afd3c77c750b2c7e041` ...)
1 · Systems sources (B2 / Mosaic Streaming @ `d99bf9c7cdf2dd4ed62a4960ed35270b93337a5f` ...)
1 · Systems sources (B3 / FFCV @ `3a12966b3afe3a81733a732e633317d747bfaac7` ...) | candidate · not_run · 1 readset(s) | no-eval |
+| [training-resource-budget](../src/inferrsi/data/skill_library/methods/training-resource-budget/SKILL.md) | common | Choose training offload and resource-budget changes using pinned DeepSpeed configuration, attributed expert diagnosis ... | Choosing training offload and resource-budget changes from pinned DeepSpeed ... | failure_modes, input_contract, produced_artifact, refusal_point
4 advice | 1 Accelerator · Diagnose / Configure / Optimize
2 RAM/Host · Observe / Diagnose / Configure / Optimize / Verify
3 SSD · Configure / Optimize
4 Network · Observe / Diagnose / Verify
5 OS · Diagnose / Configure | 1 · Systems sources (CR1 / DeepSpeed ZeRO-Infinity @ `5f266e26d216a705d9c64cac88fb813ce ...)
1 · Systems sources (CR5 / Ascend MindSpeed adaptive memory — historical `core_r0.8.0` ...)
1 · Systems sources (ZeRO-Offload — [2101.06840v1](https://arxiv.org/abs/2101.06840v1))
2 · Methodology and corpus references (CR6 / Stas Bekman ML Engineering Open Book @ ...) | candidate · not_run · 9 readset(s) | no-eval |
| [video-diffusion-optimization](../src/inferrsi/data/skill_library/methods/video-diffusion-optimization/SKILL.md) | common | Run and verify goal-driven video diffusion optimization, backend-agnostic. Use when the user asks for latency, RTF ... | Backend-agnostic video diffusion inference and serving optimization | — | — | — | candidate · not_run · 0 readset(s) | no-SearchSpace-layer, no-Reference-Matrix-row, no-eval |
-| [weight-loading-optimization](../src/inferrsi/data/skill_library/methods/weight-loading-optimization/SKILL.md) | common | Optimize the serving-level weight load path from storage to device: judge the load path on four dimensions (source ... | **unknown** | — | 1 Accelerator †guidance: Flare preparation and complete-session source contracts
2 RAM/Host †guidance: Flare preparation and complete-session source contracts
3 SSD · Observe / Profile / Diagnose / Optimize / Verify †guidance: Flare preparation and complete-session source contracts
4 Network · Configure / Optimize / Verify | 1 · Systems sources (Ascend/MindIE-SD @ 8481d39)
1 · Systems sources (NVIDIA/TensorRT-LLM @ f33e0e5fd)
1 · Systems sources (ascend-ai-coding/awesome-ascend-skills @ 75ec89c)
1 · Systems sources (foundation-model-stack/fastsafetensors @ 219e256)
1 · Systems sources (ggml-org/llama.cpp @ `3e390af18de1131dda27913c00511725b4274c8a` (MIT ...)
1 · Systems sources (mindie-agent/mindie-agent @ 6a81b41) | candidate · not_run · 3 readset(s) | no-applicability, no-eval |
+| [weight-loading-optimization](../src/inferrsi/data/skill_library/methods/weight-loading-optimization/SKILL.md) | common | Optimize the serving-level weight load path from storage to device: judge the load path on four dimensions (source ... | Diagnosing the serving-level weight load path from storage to device and choosing changes ... | failure_modes, input_contract, produced_artifact, refusal_point
5 advice | 1 Accelerator †guidance: Flare preparation and complete-session source contracts
2 RAM/Host †guidance: Flare preparation and complete-session source contracts
3 SSD · Observe / Profile / Diagnose / Optimize / Verify †guidance: Flare preparation and complete-session source contracts
4 Network · Configure / Optimize / Verify | 1 · Systems sources (Ascend/MindIE-SD @ 8481d39)
1 · Systems sources (NVIDIA/TensorRT-LLM @ f33e0e5fd)
1 · Systems sources (ascend-ai-coding/awesome-ascend-skills @ 75ec89c)
1 · Systems sources (foundation-model-stack/fastsafetensors @ 219e256)
1 · Systems sources (ggml-org/llama.cpp @ `3e390af18de1131dda27913c00511725b4274c8a` (MIT ...)
1 · Systems sources (mindie-agent/mindie-agent @ 6a81b41) | candidate · not_run · 3 readset(s) | no-eval |
## Registered: Workflows (6)
@@ -221,7 +221,7 @@ an entrypoint routes to procedures below it rather than adding one.
| `no-SearchSpace-layer` | 15: ascend-video-diffusion-optimization, audit-fixed-workload-claim, audit-recurrent-kv-finalization, cite-check-downstream-pr, competitor-scan, draft-acceptance-issue, dwarfstar, husky, inference-benchmark-design, lily, npu-skill-factory, repository-evolution-analysis, skill-evolution-playbook, splash, video-diffusion-optimization |
| `no-Reference-Matrix-row` | 8: ascend-video-diffusion-optimization, audit-fixed-workload-claim, cite-check-downstream-pr, claim-contract, competitor-scan, draft-acceptance-issue, rsi-evidence, video-diffusion-optimization |
| `no-sidecar` | 0: — |
-| `no-applicability` | 37: as listed in the Gaps column above |
+| `no-applicability` | 20: ascend-aclgraph-capture-replay, ascend-device-health, ascend-distributed-parallel, ascend-graph-mode, ascend-model-adaptation, ascend-operator-profiling, ascend-operator-tuning, ascend-optimization-casebook, ascend-performance-analysis, ascendc-tiling-design, cann-attribution-overlap, cgroup-resource-controls, diagnose-attention-bottlenecks, host-memory-management, host-overhead-optimization, host-overhead-profiling, os-process-placement, storage-io-measurement, torchnpu-device-memory-lifecycle, torchnpu-pinned-host-memory-lifecycle |
| `no-eval` | 75: as listed in the Gaps column above |
These are inventory gaps, not defects. A workflow skill may legitimately
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/README.md b/src/inferrsi/data/repo_packs/inferrsi/knowledge/README.md
index 4fcff0249..0645548d5 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/README.md
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/README.md
@@ -15,7 +15,7 @@ These are knowledge abstracts (`--kind knowledge`), with namespaced package
IDs distinct from explicitly selected `checkout_ref` skills. Ascend/CUDA
procedures use their corresponding backend; shared methods use `common`.
-Source snapshot: [a7f23dfef4770e8542f8390709b029fc934928e4](https://github.com/InferMatrixLab/InferRSI/tree/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library).
+Source snapshot: [daa33a4ba588298ff76e3e077d6c894d7750d9bc](https://github.com/InferMatrixLab/InferRSI/tree/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library).
```bash
inferrsi knowledge search 'profiling' --repo-type inferrsi --limit 50
@@ -29,7 +29,7 @@ contain the current input paths and bytes (Git checkout filters are respected).
```bash
python tools/build_skill_index.py --source-revision
-python tools/build_skill_index.py --source-revision a7f23dfef4770e8542f8390709b029fc934928e4 --check
+python tools/build_skill_index.py --source-revision daa33a4ba588298ff76e3e077d6c894d7750d9bc --check
```
`--check` detects missing, stale and unexpected generated records without writing.
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-aclgraph-capture-replay.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-aclgraph-capture-replay.json
index 36d1943b5..ca56dfe40 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-aclgraph-capture-replay.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-aclgraph-capture-replay.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/hardware/ascend/ascend-aclgraph-capture-replay/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/hardware/ascend/ascend-aclgraph-capture-replay/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-aclgraph-capture-replay/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-aclgraph-capture-replay/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-device-health.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-device-health.json
index 64c151619..7395d5531 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-device-health.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-device-health.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/hardware/ascend/ascend-device-health/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/hardware/ascend/ascend-device-health/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-device-health/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-device-health/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-distributed-parallel.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-distributed-parallel.json
index 39fe6821f..cb6c8c4fd 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-distributed-parallel.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-distributed-parallel.json
@@ -27,8 +27,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/hardware/ascend/ascend-distributed-parallel/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/hardware/ascend/ascend-distributed-parallel/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-distributed-parallel/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-distributed-parallel/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-graph-mode.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-graph-mode.json
index 753ce0763..836b58e05 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-graph-mode.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-graph-mode.json
@@ -26,8 +26,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/hardware/ascend/ascend-graph-mode/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/hardware/ascend/ascend-graph-mode/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-graph-mode/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-graph-mode/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-model-adaptation.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-model-adaptation.json
index c0da75751..e55deb1d0 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-model-adaptation.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-model-adaptation.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/hardware/ascend/ascend-model-adaptation/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/hardware/ascend/ascend-model-adaptation/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-model-adaptation/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-model-adaptation/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-profiling.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-profiling.json
index 18e566431..936e75bb5 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-profiling.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-profiling.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-profiling/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-profiling/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-profiling/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-profiling/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-tuning.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-tuning.json
index f3adf32b8..9b9ac9da6 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-tuning.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-tuning.json
@@ -24,8 +24,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-tuning/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-tuning/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-tuning/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-tuning/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-optimization-casebook.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-optimization-casebook.json
index 66096ed72..436cc7e64 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-optimization-casebook.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-optimization-casebook.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/hardware/ascend/ascend-optimization-casebook/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/hardware/ascend/ascend-optimization-casebook/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-optimization-casebook/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-optimization-casebook/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-performance-analysis.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-performance-analysis.json
index badfcf4c5..532c02bae 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-performance-analysis.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-performance-analysis.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/hardware/ascend/ascend-performance-analysis/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/hardware/ascend/ascend-performance-analysis/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-performance-analysis/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-performance-analysis/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-video-diffusion-optimization.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-video-diffusion-optimization.json
index 8dbcf642e..d18b828b2 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-video-diffusion-optimization.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-video-diffusion-optimization.json
@@ -26,8 +26,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/hardware/ascend/ascend-video-diffusion-optimization/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/hardware/ascend/ascend-video-diffusion-optimization/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-video-diffusion-optimization/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-video-diffusion-optimization/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascendc-tiling-design.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascendc-tiling-design.json
index f9b6799d6..549d7438f 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascendc-tiling-design.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascendc-tiling-design.json
@@ -24,8 +24,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/hardware/ascend/ascendc-tiling-design/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/hardware/ascend/ascendc-tiling-design/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascendc-tiling-design/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascendc-tiling-design/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/cann-attribution-overlap.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/cann-attribution-overlap.json
index db98eb823..56e78c7fa 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/cann-attribution-overlap.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/cann-attribution-overlap.json
@@ -24,8 +24,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/hardware/ascend/cann-attribution-overlap/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/hardware/ascend/cann-attribution-overlap/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/cann-attribution-overlap/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/cann-attribution-overlap/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-device-memory-lifecycle.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-device-memory-lifecycle.json
index c15d230fe..4946c2645 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-device-memory-lifecycle.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-device-memory-lifecycle.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/hardware/ascend/torchnpu-device-memory-lifecycle/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/hardware/ascend/torchnpu-device-memory-lifecycle/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/torchnpu-device-memory-lifecycle/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/torchnpu-device-memory-lifecycle/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-pinned-host-memory-lifecycle.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-pinned-host-memory-lifecycle.json
index d6c1aaa18..bb0c2b4cc 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-pinned-host-memory-lifecycle.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-pinned-host-memory-lifecycle.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/hardware/ascend/torchnpu-pinned-host-memory-lifecycle/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/hardware/ascend/torchnpu-pinned-host-memory-lifecycle/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/torchnpu-pinned-host-memory-lifecycle/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/torchnpu-pinned-host-memory-lifecycle/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-serving-preemption.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-serving-preemption.json
index ab7bd4668..7776378c8 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-serving-preemption.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-serving-preemption.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/methods/assess-serving-preemption/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/methods/assess-serving-preemption/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/methods/assess-serving-preemption/SKILL.md",
"src/inferrsi/data/skill_library/methods/assess-serving-preemption/metadata.json",
@@ -34,7 +34,7 @@
"checkout_ref": "checkout:assess-serving-preemption",
"source_files": {
"src/inferrsi/data/skill_library/methods/assess-serving-preemption/SKILL.md": "0a5f48d9352c5bf460775ef47f9136829978811f4a16e84983e299782cae305d",
- "src/inferrsi/data/skill_library/methods/assess-serving-preemption/metadata.json": "77ce5ba16245cbd80d46d0abdf80aaf26fc0924fa2d2a5624ff430de68a1daf0",
+ "src/inferrsi/data/skill_library/methods/assess-serving-preemption/metadata.json": "3e2cea37f5669e7c60f7c01c390a48ea960b15cbd6de8f5340cf8fe2afb9e1a1",
"src/inferrsi/data/skill_library/methods/assess-serving-preemption/references/source-notes.md": "f29fcecfba0490008a354c11b42403983fe165e56ff907ad61cc6040d2b6407f",
"src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
},
@@ -50,6 +50,27 @@
"origin": [
"external"
],
+ "applicability": {
+ "scope": "Diagnosing whether LLM serving queue delay justifies token-boundary priority preemption, including suspended KV, host swap or recomputation and long-request service guarantees; admission-versus-service-order diagnosis, not current scheduler implementation or tenant-fairness certification",
+ "prerequisites": "The packaged source notes are read before quoting a metric or comparing a baseline; installing, benchmarking, changing scheduling or admission, migrating state, editing a repository or running a reproduction needs a separately authorized task",
+ "input_contract": "The late boundary (first token, token gap, completion or a stated cohort's service) with its exact percentile or attainment target and denominator; model and KV format, engine revision, safe interruption boundary, request batch limit, input/output/context distributions, arrival bursts, device live/reserved/allocated bytes, active and suspended requests, host budget, transfer path, recovery time and queue/phase/service timings; per-tenant or priority-class obligations kept apart from aggregate latency",
+ "produced_artifact": "A read-only decision naming the observed failure boundary, the smallest candidate that addresses it, unchanged assumptions, missing evidence, resource cost, affected cohorts, counter-signal and next authorized gate, plus a proposed verification plan; not a validated configuration",
+ "failure_modes": [
+ "An ordinary iteration boundary is taken to permit interrupting a prefill halfway through",
+ "Unknown output length is treated as known from the prompt",
+ "KV is counted for the running batch only, omitting started requests that are suspended",
+ "Reservations are removed because some reserved capacity is not live",
+ "Recompute-based eviction is proposed without a progress invariant, so two jobs can repeatedly displace each other",
+ "Host-device copies or same-host IPC are charged to Network instead of their local resource"
+ ],
+ "refusal_point": "Withholds a preemption recommendation when the apparent gain can be explained by different kernels, unmatched parallelism or an incomparable latency metric; unknown safe boundaries, unsupported state recovery, uncalibrated timing or unclear service obligations are blockers"
+ },
+ "recommendations": [
+ "Name the late boundary and its exact target and denominator before choosing among batching, admission, preemption and state recovery.",
+ "Send prefill interference to choose-prefill-decode-schedule and physical KV slack to diagnose-kv-block-efficiency when those candidates are available; otherwise report the diagnosed boundary and the missing evidence.",
+ "In a later authorized test, isolate service ordering first and state recovery second, and evaluate their interaction only after both are understood.",
+ "Weigh FastServe's same-engine FCFS comparison above its historical-vLLM headline, and do not read its P95 aggregate gain as a starvation or tenant-fairness guarantee."
+ ],
"limitations": [
"Source-derived candidate; no local efficacy validation.",
"Original typed source inventory is not normalized; retain source identities and limits in the procedure/readsets."
@@ -59,10 +80,26 @@
],
"consumption_mode": "packaged"
},
- "applicability": {},
+ "applicability": {
+ "scope": "Diagnosing whether LLM serving queue delay justifies token-boundary priority preemption, including suspended KV, host swap or recomputation and long-request service guarantees; admission-versus-service-order diagnosis, not current scheduler implementation or tenant-fairness certification",
+ "prerequisites": "The packaged source notes are read before quoting a metric or comparing a baseline; installing, benchmarking, changing scheduling or admission, migrating state, editing a repository or running a reproduction needs a separately authorized task",
+ "input_contract": "The late boundary (first token, token gap, completion or a stated cohort's service) with its exact percentile or attainment target and denominator; model and KV format, engine revision, safe interruption boundary, request batch limit, input/output/context distributions, arrival bursts, device live/reserved/allocated bytes, active and suspended requests, host budget, transfer path, recovery time and queue/phase/service timings; per-tenant or priority-class obligations kept apart from aggregate latency",
+ "produced_artifact": "A read-only decision naming the observed failure boundary, the smallest candidate that addresses it, unchanged assumptions, missing evidence, resource cost, affected cohorts, counter-signal and next authorized gate, plus a proposed verification plan; not a validated configuration",
+ "failure_modes": [
+ "An ordinary iteration boundary is taken to permit interrupting a prefill halfway through",
+ "Unknown output length is treated as known from the prompt",
+ "KV is counted for the running batch only, omitting started requests that are suspended",
+ "Reservations are removed because some reserved capacity is not live",
+ "Recompute-based eviction is proposed without a progress invariant, so two jobs can repeatedly displace each other",
+ "Host-device copies or same-host IPC are charged to Network instead of their local resource"
+ ],
+ "refusal_point": "Withholds a preemption recommendation when the apparent gain can be explained by different kernels, unmatched parallelism or an incomparable latency metric; unknown safe boundaries, unsupported state recovery, uncalibrated timing or unclear service obligations are blockers"
+ },
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Name the late boundary and its exact target and denominator before choosing among batching, admission, preemption and state recovery.",
+ "Send prefill interference to choose-prefill-decode-schedule and physical KV slack to diagnose-kv-block-efficiency when those candidates are available; otherwise report the diagnosed boundary and the missing evidence.",
+ "In a later authorized test, isolate service ordering first and state recovery second, and evaluate their interaction only after both are understood.",
+ "Weigh FastServe's same-engine FCFS comparison above its historical-vLLM headline, and do not read its P95 aggregate gain as a starvation or tenant-fairness guarantee."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-split-backward-schedule.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-split-backward-schedule.json
index b31436fde..b0aa98710 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-split-backward-schedule.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-split-backward-schedule.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/methods/assess-split-backward-schedule/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/methods/assess-split-backward-schedule/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/methods/assess-split-backward-schedule/SKILL.md",
"src/inferrsi/data/skill_library/methods/assess-split-backward-schedule/metadata.json",
@@ -34,7 +34,7 @@
"checkout_ref": "checkout:assess-split-backward-schedule",
"source_files": {
"src/inferrsi/data/skill_library/methods/assess-split-backward-schedule/SKILL.md": "1e64f28ea71672fb53c56f6108b87864738c6d03d6dcd32eedaef9e0625e96e3",
- "src/inferrsi/data/skill_library/methods/assess-split-backward-schedule/metadata.json": "c6acae57d70190cea0e18c71c79127962181e58169e011df75d8635b3db4daa7",
+ "src/inferrsi/data/skill_library/methods/assess-split-backward-schedule/metadata.json": "a089dd2c2e5a4b41e970e426c63b5b8f65acc77d2e5a40a44b38543b5d730a00",
"src/inferrsi/data/skill_library/methods/assess-split-backward-schedule/references/source-notes.md": "61b1609f71c37b61d136b8c4d39f793de87cffb1cf31c8ec737a1c3fa6fb1b4e",
"src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
},
@@ -49,6 +49,26 @@
"origin": [
"external"
],
+ "applicability": {
+ "scope": "Assessing memory-capped split-backward and V-shaped pipeline schedule proposals when a pipeline trace suggests input-gradient propagation waits behind deferrable weight-gradient work or an optimizer barrier, while preserving synchronous update, global-batch and optimizer-validation semantics",
+ "prerequisites": "The packaged evidence notes are read before citing the paper; installing or running frameworks, benchmarking schedules, changing training semantics, configuring hardware or modifying repositories needs separate authorization",
+ "input_contract": "Stage and microbatch counts, microbatch and global batch, precision, optimizer and clipping/NaN behavior; F, input-gradient B, weight-gradient W and transfer timings; per-pass retained and released bytes; total device peak and margin; collective ordering; kernel-splitting support; calculated bubble rate kept apart from measured idle time and useful samples/s",
+ "produced_artifact": "A candidate schedule and fallback with per-pass order rationale, peak-memory and transfer costs, unchanged semantic requirements, counter-signals and a future verification proposal; a reviewable plan, not an enabled schedule",
+ "failure_modes": [
+ "Completion of B is equated with releasing all activations, so W inputs retained by deferral go uncounted",
+ "The equal-time zero-bubble construction replaces observed F/B/W and transfer times, or kernel-splitting and synchronization overhead is left out",
+ "The schedule is chosen by minimum modeled bubble rate instead of whole-step throughput under the memory and quality constraint",
+ "pMB activation capacity is read as total memory, hiding HBM, workspace, buffers or per-rank imbalance",
+ "Removing the optimizer barrier drops global norm, clipping or NaN decisions, rollback and re-update, or the correctness of work that depends on a tentative update",
+ "Optimizer post-validation is conflated with accepting stale-gradient delayed updates"
+ ],
+ "refusal_point": "Withholds an enablement recommendation under frequent clipping/NaNs, small available memory, strongly unequal passes, unsupported split kernels, hidden collectives or dynamic shapes, and retains the synchronized fallback while optimizer reversibility and rollback coverage are unverified"
+ },
+ "recommendations": [
+ "Compare ordinary 1F1B with a memory-capped split schedule first; add V-shaped two-chunk placement only when memory permits, and at matched global batch also try smaller microbatches with more microbatches.",
+ "Decide optimizer post-validation separately from the schedule change.",
+ "In a later authorized comparison, hold model, global batch and precision fixed and compare real with calculated idle intervals."
+ ],
"limitations": [
"Source-derived candidate; no local efficacy validation.",
"Original typed source inventory is not normalized; retain source identities and limits in the procedure/readsets."
@@ -58,10 +78,25 @@
],
"consumption_mode": "packaged"
},
- "applicability": {},
+ "applicability": {
+ "scope": "Assessing memory-capped split-backward and V-shaped pipeline schedule proposals when a pipeline trace suggests input-gradient propagation waits behind deferrable weight-gradient work or an optimizer barrier, while preserving synchronous update, global-batch and optimizer-validation semantics",
+ "prerequisites": "The packaged evidence notes are read before citing the paper; installing or running frameworks, benchmarking schedules, changing training semantics, configuring hardware or modifying repositories needs separate authorization",
+ "input_contract": "Stage and microbatch counts, microbatch and global batch, precision, optimizer and clipping/NaN behavior; F, input-gradient B, weight-gradient W and transfer timings; per-pass retained and released bytes; total device peak and margin; collective ordering; kernel-splitting support; calculated bubble rate kept apart from measured idle time and useful samples/s",
+ "produced_artifact": "A candidate schedule and fallback with per-pass order rationale, peak-memory and transfer costs, unchanged semantic requirements, counter-signals and a future verification proposal; a reviewable plan, not an enabled schedule",
+ "failure_modes": [
+ "Completion of B is equated with releasing all activations, so W inputs retained by deferral go uncounted",
+ "The equal-time zero-bubble construction replaces observed F/B/W and transfer times, or kernel-splitting and synchronization overhead is left out",
+ "The schedule is chosen by minimum modeled bubble rate instead of whole-step throughput under the memory and quality constraint",
+ "pMB activation capacity is read as total memory, hiding HBM, workspace, buffers or per-rank imbalance",
+ "Removing the optimizer barrier drops global norm, clipping or NaN decisions, rollback and re-update, or the correctness of work that depends on a tentative update",
+ "Optimizer post-validation is conflated with accepting stale-gradient delayed updates"
+ ],
+ "refusal_point": "Withholds an enablement recommendation under frequent clipping/NaNs, small available memory, strongly unequal passes, unsupported split kernels, hidden collectives or dynamic shapes, and retains the synchronized fallback while optimizer reversibility and rollback coverage are unverified"
+ },
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Compare ordinary 1F1B with a memory-capped split schedule first; add V-shaped two-chunk placement only when memory permits, and at matched global batch also try smaller microbatches with more microbatches.",
+ "Decide optimizer post-validation separately from the schedule change.",
+ "In a later authorized comparison, hold model, global batch and precision fixed and compare real with calculated idle intervals."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-fixed-workload-claim.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-fixed-workload-claim.json
index dc95214e5..fbcb6c5f5 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-fixed-workload-claim.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-fixed-workload-claim.json
@@ -20,8 +20,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/methods/audit-fixed-workload-claim/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/methods/audit-fixed-workload-claim/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/methods/audit-fixed-workload-claim/SKILL.md",
"src/inferrsi/data/skill_library/methods/audit-fixed-workload-claim/metadata.json",
@@ -37,7 +37,7 @@
"checkout_ref": "checkout:audit-fixed-workload-claim",
"source_files": {
"src/inferrsi/data/skill_library/methods/audit-fixed-workload-claim/SKILL.md": "2be89b1340a3deb978c8e1792d18aeecbe8f5de81d4daf542389f53826b8ebee",
- "src/inferrsi/data/skill_library/methods/audit-fixed-workload-claim/metadata.json": "81edd75fa16939807548d0944e9203cfd7c0cb17e63463872d96ca2c1273d1d8",
+ "src/inferrsi/data/skill_library/methods/audit-fixed-workload-claim/metadata.json": "d36e8a6148b7ab830f1ddf6e573a865965502709a835ce460c37e3d4c386de95",
"src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
},
"skill_metadata": {
@@ -53,14 +53,47 @@
"measurement"
],
"empirical_validation": "not_run",
+ "applicability": {
+ "scope": "Auditing a performance claim against supplied fixed-workload reports and an explicit measurement contract: aggregation, retained observations, comparability and quality evidence",
+ "input_contract": "Baseline and candidate fixed-workload reports with request observations grouped by trial and workload; the measurement contract stating metric, unit, timing boundary, improvement direction, aggregation rule and order, fixed exclusions, threshold and comparison operator; the declared quality and completion evidence; the requested output schema",
+ "produced_artifact": "An audit conclusion showing baseline and candidate aggregates, the calculation, the threshold comparison and the evidence or missing information that controls the verdict; it interprets supplied evidence and does not itself establish a measured improvement",
+ "failure_modes": [
+ "A pooled request median silently replaces a median of per-trial means, or the reverse, reweighting trials whose request counts differ",
+ "Late requests, failures or outliers are dropped, or requests move between trials, outside the exclusions the contract fixes",
+ "Values are rounded before the threshold comparison",
+ "A missing rule that permits different verdicts is resolved toward the favorable one",
+ "A speed improvement is read as task success, or a quality pass as equivalent measured work",
+ "Different workloads or incomplete observations are repaired by a favorable aggregate"
+ ],
+ "refusal_point": "Does not replace missing evidence with a pass, does not support a percentage claim on an undefined denominator, and does not launch a new benchmark or alter the contract to resolve an unfavorable result; a failed threshold is reported apart from an unverified comparison or missing quality evidence"
+ },
+ "recommendations": [
+ "Recover the contract's metric, direction, aggregation order and comparison operator before reading the report's own verdict.",
+ "Treat a named minimum bound as >= and a maximum bound as <= unless the contract requires a strict comparison.",
+ "Keep separate regimes and metrics separate unless the contract defines their aggregation."
+ ],
"limitations": [
"Guidance for supplied reports; no local efficacy validation or measured performance gain is established."
]
},
- "applicability": {},
+ "applicability": {
+ "scope": "Auditing a performance claim against supplied fixed-workload reports and an explicit measurement contract: aggregation, retained observations, comparability and quality evidence",
+ "input_contract": "Baseline and candidate fixed-workload reports with request observations grouped by trial and workload; the measurement contract stating metric, unit, timing boundary, improvement direction, aggregation rule and order, fixed exclusions, threshold and comparison operator; the declared quality and completion evidence; the requested output schema",
+ "produced_artifact": "An audit conclusion showing baseline and candidate aggregates, the calculation, the threshold comparison and the evidence or missing information that controls the verdict; it interprets supplied evidence and does not itself establish a measured improvement",
+ "failure_modes": [
+ "A pooled request median silently replaces a median of per-trial means, or the reverse, reweighting trials whose request counts differ",
+ "Late requests, failures or outliers are dropped, or requests move between trials, outside the exclusions the contract fixes",
+ "Values are rounded before the threshold comparison",
+ "A missing rule that permits different verdicts is resolved toward the favorable one",
+ "A speed improvement is read as task success, or a quality pass as equivalent measured work",
+ "Different workloads or incomplete observations are repaired by a favorable aggregate"
+ ],
+ "refusal_point": "Does not replace missing evidence with a pass, does not support a percentage claim on an undefined denominator, and does not launch a new benchmark or alter the contract to resolve an unfavorable result; a failed threshold is reported apart from an unverified comparison or missing quality evidence"
+ },
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Recover the contract's metric, direction, aggregation order and comparison operator before reading the report's own verdict.",
+ "Treat a named minimum bound as >= and a maximum bound as <= unless the contract requires a strict comparison.",
+ "Keep separate regimes and metrics separate unless the contract defines their aggregation."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-metal-fasth3-semantics.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-metal-fasth3-semantics.json
index d578f60e2..5499c5247 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-metal-fasth3-semantics.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-metal-fasth3-semantics.json
@@ -29,8 +29,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/LICENSE",
"src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-runtime-residency.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-runtime-residency.json
index b34f75b7d..d47eaa5a7 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-runtime-residency.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-runtime-residency.json
@@ -30,8 +30,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/LICENSE",
"src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-llama-runtime-state.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-llama-runtime-state.json
index 470b7eeb5..88e489b35 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-llama-runtime-state.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-llama-runtime-state.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/SKILL.md",
"src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-recurrent-kv-finalization.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-recurrent-kv-finalization.json
index e1b53cfd2..e2e13984b 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-recurrent-kv-finalization.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-recurrent-kv-finalization.json
@@ -26,8 +26,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/LICENSE",
"src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sd-cpp-residency.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sd-cpp-residency.json
index f6d61e148..99ae7ca7f 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sd-cpp-residency.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sd-cpp-residency.json
@@ -25,8 +25,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/LICENSE",
"src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sglang-diffusion-lifetimes.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sglang-diffusion-lifetimes.json
index a11d61ab5..4f0e4108d 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sglang-diffusion-lifetimes.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sglang-diffusion-lifetimes.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/LICENSE",
"src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-vllm-kv-lifetimes.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-vllm-kv-lifetimes.json
index 5a13147cf..f7d867115 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-vllm-kv-lifetimes.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-vllm-kv-lifetimes.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/SKILL.md",
"src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/bound-windowed-attention-replay.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/bound-windowed-attention-replay.json
index 90eb3eee0..828a9fefb 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/bound-windowed-attention-replay.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/bound-windowed-attention-replay.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/SKILL.md",
"src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/budget-speculative-verification.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/budget-speculative-verification.json
index fc97f0365..61a5d12ac 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/budget-speculative-verification.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/budget-speculative-verification.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/SKILL.md",
"src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/calibrate-serving-simulation.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/calibrate-serving-simulation.json
index 3fdbb750d..f17baeb43 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/calibrate-serving-simulation.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/calibrate-serving-simulation.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/methods/calibrate-serving-simulation/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/methods/calibrate-serving-simulation/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/methods/calibrate-serving-simulation/SKILL.md",
"src/inferrsi/data/skill_library/methods/calibrate-serving-simulation/metadata.json",
@@ -34,7 +34,7 @@
"checkout_ref": "checkout:calibrate-serving-simulation",
"source_files": {
"src/inferrsi/data/skill_library/methods/calibrate-serving-simulation/SKILL.md": "ac773c6f1c110a664dfca28a3ae9129b10c07614e60ca3640dea776220e7b352",
- "src/inferrsi/data/skill_library/methods/calibrate-serving-simulation/metadata.json": "261ce9f644ea6eebb90c830f41fb8b320d2b401456e5bd8d494095a44b8620d0",
+ "src/inferrsi/data/skill_library/methods/calibrate-serving-simulation/metadata.json": "96771fb6fecf5fc7773bb7cf84038d8eaf436add1a0362dc26110c82037f8320",
"src/inferrsi/data/skill_library/methods/calibrate-serving-simulation/references/source-notes.md": "9696d3c0ef96b6611787446bd6f96d4d6255ca8d107105de8c1d9af83edb9fa5",
"src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
},
@@ -63,6 +63,26 @@
"readset": "references/source-notes.md"
}
],
+ "applicability": {
+ "scope": "Building an evidence-gated serving-configuration shortlist when a profiled simulator or cost model is proposed to reduce deployment experiments; extends performance-ceiling-analysis, optimization-search-policy and inference-benchmark-design without replacing their ceilings or measurement contract",
+ "prerequisites": "The packaged Vidur evidence and evidence-type notes are read; reading the skill does not authorize execution, and no validation check runs under it without separate execution authorization",
+ "input_contract": "Model, shape and dtype; actual kernel/runtime revision; graph mode; GPU, host and topology; request length and arrival distribution; total resources and cost basis; memory limits; latency definitions and percentile SLOs",
+ "produced_artifact": "A bounded decision memo with input identities, model assumptions, evidence type, exact metrics and denominators, unsupported dimensions, nondominated candidates with predicted ranges, joint constraint checks, the next discriminating trial and rollback/finalization requirements, keeping prediction and measurement columns apart; a shortlist for future measurement, not a measured result",
+ "failure_modes": [
+ "Learned runtime output is labeled a physical bound",
+ "A missing identity is filled with a paper default, or topology is substituted, instead of being marked unknown",
+ "A low-load check stands in for an SLO claim at the intended load regime and capacity boundary",
+ "A predictor that relies on summed KV volume hides batches with equal volume but unequal context distributions",
+ "Capacity binary search is applied without verifying its monotonic/stationary premise or treating overloaded samples explicitly",
+ "One predicted winner is reported when uncertainty exceeds rank separation"
+ ],
+ "refusal_point": "Stops predictive certification when profiles are missing, topology is substituted, error grows near capacity or host overhead is omitted; labels an ordering unresolved when uncertainty exceeds rank separation; asserts neither universal error below 9% nor current vLLM, asynchronous PP or NPU support from the historical simulator paper"
+ },
+ "recommendations": [
+ "Classify every estimate as a physical bound, measured envelope, Amdahl-style conditional impact or empirical prediction before comparing configurations.",
+ "Keep prefill and decode separate in the operator signatures, and record how sharding and memory planning enter the model.",
+ "Recalibrate after workload drift, kernel/graph/runtime changes, new hardware or topology, or mismatched residuals."
+ ],
"limitations": [
"Source-derived candidate; no local efficacy validation."
],
@@ -71,10 +91,25 @@
],
"consumption_mode": "packaged"
},
- "applicability": {},
+ "applicability": {
+ "scope": "Building an evidence-gated serving-configuration shortlist when a profiled simulator or cost model is proposed to reduce deployment experiments; extends performance-ceiling-analysis, optimization-search-policy and inference-benchmark-design without replacing their ceilings or measurement contract",
+ "prerequisites": "The packaged Vidur evidence and evidence-type notes are read; reading the skill does not authorize execution, and no validation check runs under it without separate execution authorization",
+ "input_contract": "Model, shape and dtype; actual kernel/runtime revision; graph mode; GPU, host and topology; request length and arrival distribution; total resources and cost basis; memory limits; latency definitions and percentile SLOs",
+ "produced_artifact": "A bounded decision memo with input identities, model assumptions, evidence type, exact metrics and denominators, unsupported dimensions, nondominated candidates with predicted ranges, joint constraint checks, the next discriminating trial and rollback/finalization requirements, keeping prediction and measurement columns apart; a shortlist for future measurement, not a measured result",
+ "failure_modes": [
+ "Learned runtime output is labeled a physical bound",
+ "A missing identity is filled with a paper default, or topology is substituted, instead of being marked unknown",
+ "A low-load check stands in for an SLO claim at the intended load regime and capacity boundary",
+ "A predictor that relies on summed KV volume hides batches with equal volume but unequal context distributions",
+ "Capacity binary search is applied without verifying its monotonic/stationary premise or treating overloaded samples explicitly",
+ "One predicted winner is reported when uncertainty exceeds rank separation"
+ ],
+ "refusal_point": "Stops predictive certification when profiles are missing, topology is substituted, error grows near capacity or host overhead is omitted; labels an ordering unresolved when uncertainty exceeds rank separation; asserts neither universal error below 9% nor current vLLM, asynchronous PP or NPU support from the historical simulator paper"
+ },
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Classify every estimate as a physical bound, measured envelope, Amdahl-style conditional impact or empirical prediction before comparing configurations.",
+ "Keep prefill and decode separate in the operator signatures, and record how sharding and memory planning enter the model.",
+ "Recalibrate after workload drift, kernel/graph/runtime changes, new hardware or topology, or mismatched residuals."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cgroup-resource-controls.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cgroup-resource-controls.json
index da420d71d..32c888f6b 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cgroup-resource-controls.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cgroup-resource-controls.json
@@ -25,8 +25,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/hardware/host/cgroup-resource-controls/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/hardware/host/cgroup-resource-controls/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/hardware/host/cgroup-resource-controls/SKILL.md",
"src/inferrsi/data/skill_library/hardware/host/cgroup-resource-controls/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/checkpoint-recovery-consistency.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/checkpoint-recovery-consistency.json
index a21ab3a02..ec86f4f46 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/checkpoint-recovery-consistency.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/checkpoint-recovery-consistency.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/methods/checkpoint-recovery-consistency/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/methods/checkpoint-recovery-consistency/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/methods/checkpoint-recovery-consistency/SKILL.md",
"src/inferrsi/data/skill_library/methods/checkpoint-recovery-consistency/metadata.json",
@@ -41,7 +41,7 @@
"checkout_ref": "checkout:checkpoint-recovery-consistency",
"source_files": {
"src/inferrsi/data/skill_library/methods/checkpoint-recovery-consistency/SKILL.md": "6b7db1eb2780dec27858cb7aa0a3c093c5b5dfa4c30261aebfeaa255213ff688",
- "src/inferrsi/data/skill_library/methods/checkpoint-recovery-consistency/metadata.json": "4dc8659806539b4784ea9348a01d4194ac70d0bb297a916f90ff6419efa5c328",
+ "src/inferrsi/data/skill_library/methods/checkpoint-recovery-consistency/metadata.json": "4185ff9f8b4c06d2a984753127780f29ceebf50d87b225a7697bfecccc2a4a60",
"src/inferrsi/data/skill_library/methods/checkpoint-recovery-consistency/references/dlrover-persistence-replicas.md": "055f4357f79026fcbcc766bbe77db610376ec353623f8898f5d14a30ee69e7bc",
"src/inferrsi/data/skill_library/methods/checkpoint-recovery-consistency/references/sources.md": "9425b1aab963a2c8d35f3db897bd15ee69699bf5b0d0c8425c3c8bba3e9a0233",
"src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
@@ -64,6 +64,28 @@
"external"
],
"empirical_validation": "not_run",
+ "applicability": {
+ "scope": "Assessing distributed checkpoint recovery through DLRover memory agreement, persistence completion and replica transport, and NVIDIA Resiliency Extension complete-iteration selection; distinguishes worker, host and replica loss and separates queued saves, publication, application-state validity and durable recovery. Save performance belongs to dcp-async-save-performance",
+ "prerequisites": "The pinned sources and limits are read; reading authorizes no launchers, failure injection or checkpoint loading, and restore evidence needs separate authorization",
+ "input_contract": "The declared failure domain; actual memory, storage and replica placement and helper lifetime; step or iteration identity, the expected shard/rank set and finalization result; the application's model, optimizer and caller-required RNG/scheduler/data-position state contract; the target environment's format and world-size/replication assumptions",
+ "produced_artifact": "An answer to each of the four acceptance questions (snapshot survival, cross-participant completeness, application-state completeness, restorability) at its own evidence level, a proposed per-generation recovery evidence record including errors and skipped saves, and a proposed verification split by worker, host, replica and persistent-storage loss; no recovery success, durability, NPU support or measured gain",
+ "failure_modes": [
+ "Equal steps or complete rank IDs are taken to verify tensor contents or whole application semantics",
+ "A skipped save, where not every participant acquired shared-memory access, is reported as a new recovery point",
+ "A returned step paired with an empty state dictionary is read as a successful restore",
+ "A local maximum filename or one clean shard substitutes for global coverage",
+ "The .dirty-then-rename publication is read as power-loss durability",
+ "An asynchronous save request is treated as a completed checkpoint, or queue concurrency is assumed unlimited",
+ "One failure domain is tested and recovery from all of them is claimed"
+ ],
+ "refusal_point": "Claims no recovery success, durability, NPU support or measured gain, does not upgrade allocation, metadata or a completed collective into proven recovery, and neither injects failures nor authorizes them"
+ },
+ "recommendations": [
+ "Answer the four acceptance questions separately; record each at its own evidence level.",
+ "Count replication as added availability only when copies occupy independent failure domains, and budget clean, dirty and replicated shards on each host.",
+ "Keep the one-outstanding asynchronous save restriction until the target queue version is checked.",
+ "Send save-throughput questions to dcp-async-save-performance; a faster return or higher write throughput does not establish recoverability."
+ ],
"limitations": [
"Source-derived candidate; no local efficacy validation.",
"Original typed source inventory is not normalized; retain source identities and limits in the procedure/readsets."
@@ -73,10 +95,27 @@
],
"consumption_mode": "packaged"
},
- "applicability": {},
+ "applicability": {
+ "scope": "Assessing distributed checkpoint recovery through DLRover memory agreement, persistence completion and replica transport, and NVIDIA Resiliency Extension complete-iteration selection; distinguishes worker, host and replica loss and separates queued saves, publication, application-state validity and durable recovery. Save performance belongs to dcp-async-save-performance",
+ "prerequisites": "The pinned sources and limits are read; reading authorizes no launchers, failure injection or checkpoint loading, and restore evidence needs separate authorization",
+ "input_contract": "The declared failure domain; actual memory, storage and replica placement and helper lifetime; step or iteration identity, the expected shard/rank set and finalization result; the application's model, optimizer and caller-required RNG/scheduler/data-position state contract; the target environment's format and world-size/replication assumptions",
+ "produced_artifact": "An answer to each of the four acceptance questions (snapshot survival, cross-participant completeness, application-state completeness, restorability) at its own evidence level, a proposed per-generation recovery evidence record including errors and skipped saves, and a proposed verification split by worker, host, replica and persistent-storage loss; no recovery success, durability, NPU support or measured gain",
+ "failure_modes": [
+ "Equal steps or complete rank IDs are taken to verify tensor contents or whole application semantics",
+ "A skipped save, where not every participant acquired shared-memory access, is reported as a new recovery point",
+ "A returned step paired with an empty state dictionary is read as a successful restore",
+ "A local maximum filename or one clean shard substitutes for global coverage",
+ "The .dirty-then-rename publication is read as power-loss durability",
+ "An asynchronous save request is treated as a completed checkpoint, or queue concurrency is assumed unlimited",
+ "One failure domain is tested and recovery from all of them is claimed"
+ ],
+ "refusal_point": "Claims no recovery success, durability, NPU support or measured gain, does not upgrade allocation, metadata or a completed collective into proven recovery, and neither injects failures nor authorizes them"
+ },
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Answer the four acceptance questions separately; record each at its own evidence level.",
+ "Count replication as added availability only when copies occupy independent failure domains, and budget clean, dirty and replicated shards on each host.",
+ "Keep the one-outstanding asynchronous save restriction until the target queue version is checked.",
+ "Send save-throughput questions to dcp-async-save-performance; a faster return or higher write throughput does not establish recoverability."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/choose-parallel-training-plan.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/choose-parallel-training-plan.json
index 7e43beb7d..4e33872bd 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/choose-parallel-training-plan.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/choose-parallel-training-plan.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/methods/choose-parallel-training-plan/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/methods/choose-parallel-training-plan/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/methods/choose-parallel-training-plan/SKILL.md",
"src/inferrsi/data/skill_library/methods/choose-parallel-training-plan/metadata.json",
@@ -34,7 +34,7 @@
"checkout_ref": "checkout:choose-parallel-training-plan",
"source_files": {
"src/inferrsi/data/skill_library/methods/choose-parallel-training-plan/SKILL.md": "25d04aed9782b2a8ff1186828becdc7fce93c4418039ceaef9b8dcc5ec9bcd24",
- "src/inferrsi/data/skill_library/methods/choose-parallel-training-plan/metadata.json": "c6acae57d70190cea0e18c71c79127962181e58169e011df75d8635b3db4daa7",
+ "src/inferrsi/data/skill_library/methods/choose-parallel-training-plan/metadata.json": "4990d0a2f1a269111a680f45a5d91221af37e3f8b3ce514fea84e97878e767cd",
"src/inferrsi/data/skill_library/methods/choose-parallel-training-plan/references/source-notes.md": "9f9184d46b19312138a6e805e439e43dc0594a0cc6b46675255a582dc946ca5b",
"src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
},
@@ -49,6 +49,26 @@
"origin": [
"external"
],
+ "applicability": {
+ "scope": "Comparing stage, device-mesh and operator-sharding candidates for static distributed training when heterogeneous stage memory or work, or communication cost, makes a uniform data/tensor/pipeline plan suspect; derived from historical Alpa, not a current compiler configuration guide",
+ "prerequisites": "The packaged source notes are read; the skill authorizes no compiler or solver install or run, parallelism reconfiguration, repository edit, hardware task or benchmark",
+ "input_contract": "Model operator graph and shapes; global batch, microbatch and accumulation; precision and optimizer; physical links and contention; candidate shard layouts; stage latency, workspace and in-flight activation memory; the objective and expected run length. A device count without topology is insufficient",
+ "produced_artifact": "The best supported candidate and a simpler fallback with a memory and critical-path ledger, planning cost and amortization, semantic invariants, counter-signals, missing evidence and a proposed matched comparison; a plan, not a reconfiguration",
+ "failure_modes": [
+ "Equal arithmetic work is taken to mean equal kernel time, carrying over Alpa's zero compute-cost simplification",
+ "Mean memory replaces the largest per-rank peak with multiple in-flight activation sets",
+ "Inter-stage transfer and microbatch count stay outside the objective, as in historical Alpa, without the gap being reported",
+ "Repeated slow-link sends to replicated destination tiles are not compared with one cross-mesh transfer plus local allgather",
+ "Inter-node Network traffic is merged with intra-node accelerator links",
+ "Static cost estimates are applied to dynamic shapes, asymmetric devices or changing communication contention"
+ ],
+ "refusal_point": "Rejects infeasible stage/mesh pairs, claims no global optimality outside the declared search space, and stops at the plan in paper-only mode"
+ },
+ "recommendations": [
+ "Keep a simple reference plan with the same training semantics and global batch as the comparison point.",
+ "Separate fixed-model scaling from model-size weak scaling before quoting a speedup; Alpa's 9.7x MoE ratio belongs to a specific weak-scaling workload, baseline and 25-Gbps cluster.",
+ "Weigh planning cost against the expected run length; a shorter job may not amortize it."
+ ],
"limitations": [
"Source-derived candidate; no local efficacy validation.",
"Original typed source inventory is not normalized; retain source identities and limits in the procedure/readsets."
@@ -58,10 +78,25 @@
],
"consumption_mode": "packaged"
},
- "applicability": {},
+ "applicability": {
+ "scope": "Comparing stage, device-mesh and operator-sharding candidates for static distributed training when heterogeneous stage memory or work, or communication cost, makes a uniform data/tensor/pipeline plan suspect; derived from historical Alpa, not a current compiler configuration guide",
+ "prerequisites": "The packaged source notes are read; the skill authorizes no compiler or solver install or run, parallelism reconfiguration, repository edit, hardware task or benchmark",
+ "input_contract": "Model operator graph and shapes; global batch, microbatch and accumulation; precision and optimizer; physical links and contention; candidate shard layouts; stage latency, workspace and in-flight activation memory; the objective and expected run length. A device count without topology is insufficient",
+ "produced_artifact": "The best supported candidate and a simpler fallback with a memory and critical-path ledger, planning cost and amortization, semantic invariants, counter-signals, missing evidence and a proposed matched comparison; a plan, not a reconfiguration",
+ "failure_modes": [
+ "Equal arithmetic work is taken to mean equal kernel time, carrying over Alpa's zero compute-cost simplification",
+ "Mean memory replaces the largest per-rank peak with multiple in-flight activation sets",
+ "Inter-stage transfer and microbatch count stay outside the objective, as in historical Alpa, without the gap being reported",
+ "Repeated slow-link sends to replicated destination tiles are not compared with one cross-mesh transfer plus local allgather",
+ "Inter-node Network traffic is merged with intra-node accelerator links",
+ "Static cost estimates are applied to dynamic shapes, asymmetric devices or changing communication contention"
+ ],
+ "refusal_point": "Rejects infeasible stage/mesh pairs, claims no global optimality outside the declared search space, and stops at the plan in paper-only mode"
+ },
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Keep a simple reference plan with the same training semantics and global batch as the comparison point.",
+ "Separate fixed-model scaling from model-size weak scaling before quoting a speedup; Alpa's 9.7x MoE ratio belongs to a specific weak-scaling workload, baseline and 25-Gbps cluster.",
+ "Weigh planning cost against the expected run length; a shorter job may not amortize it."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cite-check-downstream-pr.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cite-check-downstream-pr.json
index a7dffc283..0578cef02 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cite-check-downstream-pr.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cite-check-downstream-pr.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/workflows/cite-check-downstream-pr/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/workflows/cite-check-downstream-pr/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/skill-index.json",
"src/inferrsi/data/skill_library/workflows/cite-check-downstream-pr/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/claim-contract.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/claim-contract.json
index 2a622fe88..e31b42196 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/claim-contract.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/claim-contract.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/workflows/claim-contract/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/workflows/claim-contract/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/skill-index.json",
"src/inferrsi/data/skill_library/workflows/claim-contract/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/comfy-ui.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/comfy-ui.json
index b022ae568..619da8879 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/comfy-ui.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/comfy-ui.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/repositories/comfy-ui/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/repositories/comfy-ui/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/repositories/comfy-ui/LICENSE",
"src/inferrsi/data/skill_library/repositories/comfy-ui/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/competitor-scan.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/competitor-scan.json
index 50544030a..9b6470def 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/competitor-scan.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/competitor-scan.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/workflows/competitor-scan/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/workflows/competitor-scan/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/skill-index.json",
"src/inferrsi/data/skill_library/workflows/competitor-scan/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dcp-async-save-performance.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dcp-async-save-performance.json
index a891567b4..36fa0ec7b 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dcp-async-save-performance.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dcp-async-save-performance.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/SKILL.md",
"src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dense-performance-evidence.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dense-performance-evidence.json
index 3a5024c0d..841dfaa07 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dense-performance-evidence.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dense-performance-evidence.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/methods/dense-performance-evidence/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/methods/dense-performance-evidence/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/methods/dense-performance-evidence/SKILL.md",
"src/inferrsi/data/skill_library/methods/dense-performance-evidence/metadata.json",
@@ -38,7 +38,7 @@
"checkout_ref": "checkout:dense-performance-evidence",
"source_files": {
"src/inferrsi/data/skill_library/methods/dense-performance-evidence/SKILL.md": "53cc1867a13f54d1006d50517dea8848869ca50df3f0cd610b2424cc8d18c933",
- "src/inferrsi/data/skill_library/methods/dense-performance-evidence/metadata.json": "a0d1e3b17140295510a764045d1b222e19be3a3406e6d1e7dddaad49ff474490",
+ "src/inferrsi/data/skill_library/methods/dense-performance-evidence/metadata.json": "041780985dcaf76809d29dc0d4e242359b17b6bcfaadab2d05674fa4699a5427",
"src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
},
"skill_metadata": {
@@ -58,6 +58,26 @@
"external"
],
"empirical_validation": "not_run",
+ "applicability": {
+ "scope": "Deciding which performance evidence to collect next and which claim each evidence level supports, from correctness through microbenchmark, component profile, isolated scenario benchmark and multi-regime serving benchmark to a production A/B experiment; also structuring a measurement campaign",
+ "prerequisites": "Level 0 correctness evidence is collected before any performance measurement; a Level 5 experiment needs production infrastructure, traffic splitting and statistical power",
+ "input_contract": "The claim to be made or the question to answer, and the evidence already collected with its level, workload, hardware/software context and controls",
+ "produced_artifact": "The minimum evidence level each claim requires, the next evidence to collect under the escalation rule, and negative-result records with hypothesis, evidence level, the metric that did not improve, context and the reason if known; a classification of evidence, not a measurement",
+ "failure_modes": [
+ "An md5 match or a test pass is read as semantically correct output or as support for a performance claim",
+ "A microbenchmark improvement is reported as an inference throughput gain",
+ "Absolute kernel durations or single collective latencies are compared across devices or sessions",
+ "A win on one fixed workload or concurrency is generalized to production traffic",
+ "Worse metrics, per-task regressions or negative results are discarded, or a shared task budget is taken as equal total cost"
+ ],
+ "refusal_point": "Withholds any claim above the level its evidence reaches, leaves an unresolved source protocol or denominator mismatch unresolved, and lets neither upstream scores nor local packaging checks promote an InferRSI candidate"
+ },
+ "recommendations": [
+ "Collect Level 0 first and profile at Level 2 before implementing; escalate to Level 3 only after a promising Level 2 result, and to Level 4 only after a Level 3 win worth validating across regimes.",
+ "Collect a Level 1 microbenchmark only for an isolated-kernel question after Level 2 has confirmed that kernel is on the critical path.",
+ "For a Level 1 claim, pin input shape and dtype, exclude warmup, synchronize before timing, run at least 10 repetitions and report median and IQR; for a Level 3 claim, match hardware, software stack, model revision and inputs, run at least 3 repetitions and alternate A/B/A order.",
+ "Read the admission/drain occupancy case when interpreting finite service cohorts; phase-selected occupancy is not end-to-end gain."
+ ],
"limitations": [
"Source-derived candidate; no local efficacy validation.",
"Original typed source inventory is not normalized; retain source identities and limits in the procedure/readsets."
@@ -69,10 +89,25 @@
],
"consumption_mode": "packaged"
},
- "applicability": {},
+ "applicability": {
+ "scope": "Deciding which performance evidence to collect next and which claim each evidence level supports, from correctness through microbenchmark, component profile, isolated scenario benchmark and multi-regime serving benchmark to a production A/B experiment; also structuring a measurement campaign",
+ "prerequisites": "Level 0 correctness evidence is collected before any performance measurement; a Level 5 experiment needs production infrastructure, traffic splitting and statistical power",
+ "input_contract": "The claim to be made or the question to answer, and the evidence already collected with its level, workload, hardware/software context and controls",
+ "produced_artifact": "The minimum evidence level each claim requires, the next evidence to collect under the escalation rule, and negative-result records with hypothesis, evidence level, the metric that did not improve, context and the reason if known; a classification of evidence, not a measurement",
+ "failure_modes": [
+ "An md5 match or a test pass is read as semantically correct output or as support for a performance claim",
+ "A microbenchmark improvement is reported as an inference throughput gain",
+ "Absolute kernel durations or single collective latencies are compared across devices or sessions",
+ "A win on one fixed workload or concurrency is generalized to production traffic",
+ "Worse metrics, per-task regressions or negative results are discarded, or a shared task budget is taken as equal total cost"
+ ],
+ "refusal_point": "Withholds any claim above the level its evidence reaches, leaves an unresolved source protocol or denominator mismatch unresolved, and lets neither upstream scores nor local packaging checks promote an InferRSI candidate"
+ },
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Collect Level 0 first and profile at Level 2 before implementing; escalate to Level 3 only after a promising Level 2 result, and to Level 4 only after a Level 3 win worth validating across regimes.",
+ "Collect a Level 1 microbenchmark only for an isolated-kernel question after Level 2 has confirmed that kernel is on the critical path.",
+ "For a Level 1 claim, pin input shape and dtype, exclude warmup, synchronize before timing, run at least 10 repetitions and report median and IQR; for a Level 3 claim, match hardware, software stack, model revision and inputs, run at least 3 repetitions and alternate A/B/A order.",
+ "Read the admission/drain occupancy case when interpreting finite service cohorts; phase-selected occupancy is not end-to-end gain."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diagnose-vllm-admission-pressure.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diagnose-vllm-admission-pressure.json
index 4033ccc66..822ff37f7 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diagnose-vllm-admission-pressure.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diagnose-vllm-admission-pressure.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/SKILL.md",
"src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diffusers-invocation-state.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diffusers-invocation-state.json
index 0286c16cf..a1ca76b99 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diffusers-invocation-state.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diffusers-invocation-state.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/SKILL.md",
"src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/draft-acceptance-issue.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/draft-acceptance-issue.json
index 7c39bde78..4d293a971 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/draft-acceptance-issue.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/draft-acceptance-issue.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/workflows/draft-acceptance-issue/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/workflows/draft-acceptance-issue/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/skill-index.json",
"src/inferrsi/data/skill_library/workflows/draft-acceptance-issue/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dwarfstar.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dwarfstar.json
index e50958232..417292ed9 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dwarfstar.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dwarfstar.json
@@ -25,8 +25,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/repositories/dwarfstar/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/repositories/dwarfstar/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/repositories/dwarfstar/LICENSE",
"src/inferrsi/data/skill_library/repositories/dwarfstar/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/evaluate-elastic-serving-placement.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/evaluate-elastic-serving-placement.json
index b3929362e..5362c7516 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/evaluate-elastic-serving-placement.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/evaluate-elastic-serving-placement.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/methods/evaluate-elastic-serving-placement/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/methods/evaluate-elastic-serving-placement/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/methods/evaluate-elastic-serving-placement/SKILL.md",
"src/inferrsi/data/skill_library/methods/evaluate-elastic-serving-placement/metadata.json",
@@ -34,7 +34,7 @@
"checkout_ref": "checkout:evaluate-elastic-serving-placement",
"source_files": {
"src/inferrsi/data/skill_library/methods/evaluate-elastic-serving-placement/SKILL.md": "82d0099d3c33706f46c657d4fa7f76d3d486e019b36760d1de001921328a60cc",
- "src/inferrsi/data/skill_library/methods/evaluate-elastic-serving-placement/metadata.json": "37379a4d4a058ca46da2b20a920dc01be9ec11e557c32b2332702ec76027f4b5",
+ "src/inferrsi/data/skill_library/methods/evaluate-elastic-serving-placement/metadata.json": "091a3b95d1a8996b5ad917e7948efb4ed598e31c7163709377f8e9199cebfe26",
"src/inferrsi/data/skill_library/methods/evaluate-elastic-serving-placement/references/source-notes.md": "ea416831fd7d0874e06d0a76fb0c7196df53072b717673a0cdf635db6bfd917b",
"src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
},
@@ -50,6 +50,26 @@
"origin": [
"external"
],
+ "applicability": {
+ "scope": "Comparing migration of a running LLM request between compatible model replicas with changing its sequence-parallel execution group, including state-copy, reservation, recomputation, interruption and priority costs, for cross-instance memory imbalance or highly variable long-context serving; not connector wiring or live scaling",
+ "prerequisites": "The relevant Llumnix or LoongServe branch and its readset in the packaged source notes are read; moving state, altering admission or priority, scaling replicas, changing parallelism, running experiments or editing frameworks needs separate authorization",
+ "input_contract": "Exact model and cache identity and format; weight residency; request phase, context and generated length; source and destination KV bytes; physical free/reserved memory and virtual accounting; batch state; current parallel group; queue and service times; actual topology and transport for each path; the target SLO statistic and affected priority or tenant cohorts",
+ "produced_artifact": "The smallest justified branch with missing capability or evidence, expected saved work or wait, added state, traffic and replica cost, affected SLO and cohorts, counter-signals, the proposed next gate and a future verification plan; an evidence-based recommendation, not a validated deployment rule",
+ "failure_modes": [
+ "Model-weight repartitioning, pipeline configuration, instance migration and sequence sharding are treated as interchangeable uses of parallelism",
+ "Aggregate free bytes are taken to mean a single blocked request fits, conflating internal KV-block slack with cross-instance fragmentation",
+ "Llumnix's 20-30 ms measured downtime is substituted for total transfer time",
+ "Virtual headroom is counted as extra memory or as a tenant-fairness guarantee",
+ "A zero-overhead transition label is read as every group change being free, or a token sum hides length-sensitive attention work",
+ "Ongoing distributed-attention communication goes uncharged because existing KV does not migrate"
+ ],
+ "refusal_point": "Withholds the candidate when every destination is full, pre-copy catch-up is unproven, model or cache compatibility is unknown, or the measured work saved does not exceed exposed costs; current-code review and target-system tests remain a later authorization gate"
+ },
+ "recommendations": [
+ "Decide first whether the question is replica relocation, a sequence-parallel group change, or prefix reuse and phase pools, which evaluate-kv-cache-transfer-gates and choose-prefill-decode-schedule cover when available.",
+ "Name each transfer path separately; only inter-node transport is Network.",
+ "Keep a feasible static, no-migration baseline in any later comparison."
+ ],
"limitations": [
"Source-derived candidate; no local efficacy validation.",
"Original typed source inventory is not normalized; retain source identities and limits in the procedure/readsets."
@@ -59,10 +79,25 @@
],
"consumption_mode": "packaged"
},
- "applicability": {},
+ "applicability": {
+ "scope": "Comparing migration of a running LLM request between compatible model replicas with changing its sequence-parallel execution group, including state-copy, reservation, recomputation, interruption and priority costs, for cross-instance memory imbalance or highly variable long-context serving; not connector wiring or live scaling",
+ "prerequisites": "The relevant Llumnix or LoongServe branch and its readset in the packaged source notes are read; moving state, altering admission or priority, scaling replicas, changing parallelism, running experiments or editing frameworks needs separate authorization",
+ "input_contract": "Exact model and cache identity and format; weight residency; request phase, context and generated length; source and destination KV bytes; physical free/reserved memory and virtual accounting; batch state; current parallel group; queue and service times; actual topology and transport for each path; the target SLO statistic and affected priority or tenant cohorts",
+ "produced_artifact": "The smallest justified branch with missing capability or evidence, expected saved work or wait, added state, traffic and replica cost, affected SLO and cohorts, counter-signals, the proposed next gate and a future verification plan; an evidence-based recommendation, not a validated deployment rule",
+ "failure_modes": [
+ "Model-weight repartitioning, pipeline configuration, instance migration and sequence sharding are treated as interchangeable uses of parallelism",
+ "Aggregate free bytes are taken to mean a single blocked request fits, conflating internal KV-block slack with cross-instance fragmentation",
+ "Llumnix's 20-30 ms measured downtime is substituted for total transfer time",
+ "Virtual headroom is counted as extra memory or as a tenant-fairness guarantee",
+ "A zero-overhead transition label is read as every group change being free, or a token sum hides length-sensitive attention work",
+ "Ongoing distributed-attention communication goes uncharged because existing KV does not migrate"
+ ],
+ "refusal_point": "Withholds the candidate when every destination is full, pre-copy catch-up is unproven, model or cache compatibility is unknown, or the measured work saved does not exceed exposed costs; current-code review and target-system tests remain a later authorization gate"
+ },
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Decide first whether the question is replica relocation, a sequence-parallel group change, or prefix reuse and phase pools, which evaluate-kv-cache-transfer-gates and choose-prefill-decode-schedule cover when available.",
+ "Name each transfer path separately; only inter-node transport is Network.",
+ "Keep a feasible static, no-migration baseline in any later comparison."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/generation-cache-lifecycle.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/generation-cache-lifecycle.json
index 681117f60..9133cbcae 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/generation-cache-lifecycle.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/generation-cache-lifecycle.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/SKILL.md",
"src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-memory-management.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-memory-management.json
index 0e0eee119..8c1ad56be 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-memory-management.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-memory-management.json
@@ -30,8 +30,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/hardware/host/host-memory-management/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/hardware/host/host-memory-management/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/hardware/host/host-memory-management/SKILL.md",
"src/inferrsi/data/skill_library/hardware/host/host-memory-management/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-overhead-optimization.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-overhead-optimization.json
index 6ce57fba4..854e75629 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-overhead-optimization.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-overhead-optimization.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/hardware/host/host-overhead-optimization/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/hardware/host/host-overhead-optimization/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/hardware/host/host-overhead-optimization/SKILL.md",
"src/inferrsi/data/skill_library/hardware/host/host-overhead-optimization/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/husky.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/husky.json
index 05a8f448a..1d70d2187 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/husky.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/husky.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/repositories/husky/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/repositories/husky/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/repositories/husky/SKILL.md",
"src/inferrsi/data/skill_library/repositories/husky/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-benchmark-design.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-benchmark-design.json
index 415d03961..3f7cf357f 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-benchmark-design.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-benchmark-design.json
@@ -30,8 +30,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/methods/inference-benchmark-design/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/methods/inference-benchmark-design/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/methods/inference-benchmark-design/SKILL.md",
"src/inferrsi/data/skill_library/methods/inference-benchmark-design/metadata.json",
@@ -49,7 +49,7 @@
"checkout_ref": "checkout:inference-benchmark-design",
"source_files": {
"src/inferrsi/data/skill_library/methods/inference-benchmark-design/SKILL.md": "59b5656e7d989b3cee8d6f95cd43be03053b1fb11c3a82bb49ab3d402f7aacee",
- "src/inferrsi/data/skill_library/methods/inference-benchmark-design/metadata.json": "18310402a8ade03fd48ac687f79ba9d447871e90337de84c0e1bc15c0b827306",
+ "src/inferrsi/data/skill_library/methods/inference-benchmark-design/metadata.json": "d51b5b3600d13b38c2adbf6c6e149c7375c9f1db12bc078679cab67a32a3e9a7",
"src/inferrsi/data/skill_library/methods/inference-benchmark-design/references/benchmark-commands.md": "0898424e0bff3a6246e1eeeda6abd6c5bd974c047fe32f3bbd5ab9d31ce17cd3",
"src/inferrsi/data/skill_library/methods/inference-benchmark-design/references/task-benchmark-selection.md": "0d4bb080a754e26846e81680023db0dff5e6a4c96be4327c411c3b72ac4ef8e2",
"src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
@@ -200,6 +200,28 @@
}
],
"empirical_validation": "not_run",
+ "applicability": {
+ "scope": "Selecting task quality gates, reference parity checks and performance workloads for text, agentic coding, work agents, TTS, full-duplex audio/video, image and video inference, and designing goodput-based multi-regime comparisons that fit InferRSI's evidence trail",
+ "prerequisites": "The primary objective, timing boundary, quality gates, sample/failure accounting, required regimes, repetitions, seeds and decision rule are frozen before optimization; baseline and candidate share hardware, software stack, input distribution, benchmark configuration, warmup and measurement protocol, and on a shared machine also one time window and dedicated devices",
+ "input_contract": "The task family and deployment use case; candidate quality evaluators and reference parity; the declared SLO and the caller's policy on which regimes must improve or must not regress; for an audit, a supplied fixed-workload report with request-to-trial grouping, exclusion and failure counts and the declared aggregation rule",
+ "produced_artifact": "A frozen benchmark design with evaluators, reference parity, primary objective, timing boundary, quality gates, required regimes, repetition and seed rules and an explicit decision config, plus the evidence-tree layout a PASS claim must produce; for a supplied report, the computed threshold result and missing evidence. Candidate knowledge, not qualification of any model or backend",
+ "failure_modes": [
+ "Raw throughput is optimized in place of goodput, so requests served outside the SLO count toward a win",
+ "One concurrency level or one undisclosed aggregate score stands in for all required regimes",
+ "Recorded agent task scores or replay completion are read as current task success, or a VBench prompt workload as computed VBench quality",
+ "An upstream better/worse/inconclusive label is used as InferRSI's PASS/FAIL/INCONCLUSIVE verdict",
+ "Runs on the same hardware class, in different time windows or on shared devices are compared as if pinned",
+ "A quality pass is taken to certify that the timed work actually occurred",
+ "Best-seen results are reported as delivered without restarting and validating the final code/config artifact"
+ ],
+ "refusal_point": "Records missing coverage and high-variance results as INCONCLUSIVE rather than a win; a run without a frozen RunSpec, or whose validator changed after the plan was frozen, is INCONCLUSIVE regardless of metric values; sets no universal numeric gate or noise threshold"
+ },
+ "recommendations": [
+ "Freeze one primary objective, timing boundary, quality gates and sample/failure accounting before optimization.",
+ "Choose the required regimes from the deployment use case, report each separately, and record which must improve or must not regress in a versioned decision config.",
+ "Use the replay trace that matches the actual deployment pattern.",
+ "State the concurrency, trace, model, SKU and microbenchmark scope boundaries in any report that cites benchmark results."
+ ],
"limitations": [
"Source-derived candidate; no local efficacy validation.",
"Original typed source inventory is not normalized; retain source identities and limits in the procedure/readsets.",
@@ -216,10 +238,27 @@
],
"consumption_mode": "packaged"
},
- "applicability": {},
+ "applicability": {
+ "scope": "Selecting task quality gates, reference parity checks and performance workloads for text, agentic coding, work agents, TTS, full-duplex audio/video, image and video inference, and designing goodput-based multi-regime comparisons that fit InferRSI's evidence trail",
+ "prerequisites": "The primary objective, timing boundary, quality gates, sample/failure accounting, required regimes, repetitions, seeds and decision rule are frozen before optimization; baseline and candidate share hardware, software stack, input distribution, benchmark configuration, warmup and measurement protocol, and on a shared machine also one time window and dedicated devices",
+ "input_contract": "The task family and deployment use case; candidate quality evaluators and reference parity; the declared SLO and the caller's policy on which regimes must improve or must not regress; for an audit, a supplied fixed-workload report with request-to-trial grouping, exclusion and failure counts and the declared aggregation rule",
+ "produced_artifact": "A frozen benchmark design with evaluators, reference parity, primary objective, timing boundary, quality gates, required regimes, repetition and seed rules and an explicit decision config, plus the evidence-tree layout a PASS claim must produce; for a supplied report, the computed threshold result and missing evidence. Candidate knowledge, not qualification of any model or backend",
+ "failure_modes": [
+ "Raw throughput is optimized in place of goodput, so requests served outside the SLO count toward a win",
+ "One concurrency level or one undisclosed aggregate score stands in for all required regimes",
+ "Recorded agent task scores or replay completion are read as current task success, or a VBench prompt workload as computed VBench quality",
+ "An upstream better/worse/inconclusive label is used as InferRSI's PASS/FAIL/INCONCLUSIVE verdict",
+ "Runs on the same hardware class, in different time windows or on shared devices are compared as if pinned",
+ "A quality pass is taken to certify that the timed work actually occurred",
+ "Best-seen results are reported as delivered without restarting and validating the final code/config artifact"
+ ],
+ "refusal_point": "Records missing coverage and high-variance results as INCONCLUSIVE rather than a win; a run without a frozen RunSpec, or whose validator changed after the plan was frozen, is INCONCLUSIVE regardless of metric values; sets no universal numeric gate or noise threshold"
+ },
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Freeze one primary objective, timing boundary, quality gates and sample/failure accounting before optimization.",
+ "Choose the required regimes from the deployment use case, report each separately, and record which must improve or must not regress in a versioned decision config.",
+ "Use the replay trace that matches the actual deployment pattern.",
+ "State the concurrency, trace, model, SKU and microbenchmark scope boundaries in any report that cites benchmark results."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-optimization-casebook.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-optimization-casebook.json
index ab8a80244..3f8a4f44f 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-optimization-casebook.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-optimization-casebook.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/methods/inference-optimization-casebook/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/methods/inference-optimization-casebook/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/methods/inference-optimization-casebook/SKILL.md",
"src/inferrsi/data/skill_library/methods/inference-optimization-casebook/cases/admission-drain-occupancy.json",
@@ -54,7 +54,7 @@
"src/inferrsi/data/skill_library/methods/inference-optimization-casebook/cases/qwen-ascend-rejected-directions.md": "be396983812262515899a3e2ba4f01b23ff92e1b836881882c7580372b8ae552",
"src/inferrsi/data/skill_library/methods/inference-optimization-casebook/cases/qwen-b300-rejected-directions.json": "146ea1564a46ed03092c7dc5796b318fc05b04b1821105ee00eb7fa7fa1333e2",
"src/inferrsi/data/skill_library/methods/inference-optimization-casebook/cases/qwen-b300-rejected-directions.md": "cce2930f544a0f78f21797fba4d8ded298e4a81f76c34d7977b0e31d361bc738",
- "src/inferrsi/data/skill_library/methods/inference-optimization-casebook/metadata.json": "fb55067fb45970319787068a3206110641cceaf0ccbabd706be6d0b3d2afc2ed",
+ "src/inferrsi/data/skill_library/methods/inference-optimization-casebook/metadata.json": "963488658f327d2379183d70357eabf832d70a6a70db2aa67f548493f4088d97",
"src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
},
"skill_metadata": {
@@ -74,6 +74,27 @@
"external"
],
"empirical_validation": "not_run",
+ "applicability": {
+ "scope": "Structuring and querying a casebook of past optimization decisions as decision support, when recording optimization experiments or choosing the next optimization from prior art; the casebook is reference material and neither runs profiling nor makes decisions",
+ "prerequisites": "Whole-body text access to the case files for free-text search; applying a case first requires reading the nearest checked-in config and the model's deployment guide",
+ "input_contract": "For a query: natural search terms such as model name, knob, mechanism word or deployment feature, plus the current model and hardware. For a new case: mechanism, applies-when and does-not-apply conditions, procedure, expected effect, accuracy risk, verification, rollback and a real measurement trail with run ID, verdict, exact hardware SKU, model revision, workload and before/after result",
+ "produced_artifact": "Ranked matching cases read as precedents and diagnostic options, or a new or deprecated case record in the template; a precedent's procedure and expected direction of effect, not a guaranteed configuration or a measured result",
+ "failure_modes": [
+ "The query is pre-translated into canonical vocabulary or bounded by structured fields, so synonyms in the body text are missed",
+ "A case's specific values, such as batch size, tensor-parallel degree or kernel parameters, are copied instead of re-derived for the current model and hardware",
+ "The precedent's expected direction of effect stands in for measurement after the change is applied",
+ "A rejected direction is read as a permanent blacklist, or one discarded implementation as ruling out its mechanism",
+ "A confounded run, a missing experiment and a failed gate are recorded as the same evidence state",
+ "Cross-system cases are bootstrapped without re-verifying that the mechanism holds on the target hardware"
+ ],
+ "refusal_point": "Adds no case for an INCONCLUSIVE optimization-effect claim, a single-configuration result without plausible generalization, or a result without a real measurement trail, except a scoped diagnostic observation that keeps its verdict and configuration boundary; manufactures no PASS, and reading the records does not reopen a paused campaign"
+ },
+ "recommendations": [
+ "Search the full case bodies with free text first, then rank the hits by bottleneck class and signals.",
+ "Before retrying a rejected direction, state which condition changed or which uncertainty a bounded check resolves, run an unchanged control and choose the least expensive discriminating check.",
+ "Mark bootstrapped cases source_grounded or author_reported until a local measurement confirms or refutes them.",
+ "When a case's conditions no longer hold, mark it deprecated with the reason instead of deleting it."
+ ],
"limitations": [
"Source-derived candidate; no local efficacy validation.",
"Original typed source inventory is not normalized; retain source identities and limits in the procedure/readsets."
@@ -81,10 +102,26 @@
"provenance_readsets": [],
"consumption_mode": "packaged"
},
- "applicability": {},
+ "applicability": {
+ "scope": "Structuring and querying a casebook of past optimization decisions as decision support, when recording optimization experiments or choosing the next optimization from prior art; the casebook is reference material and neither runs profiling nor makes decisions",
+ "prerequisites": "Whole-body text access to the case files for free-text search; applying a case first requires reading the nearest checked-in config and the model's deployment guide",
+ "input_contract": "For a query: natural search terms such as model name, knob, mechanism word or deployment feature, plus the current model and hardware. For a new case: mechanism, applies-when and does-not-apply conditions, procedure, expected effect, accuracy risk, verification, rollback and a real measurement trail with run ID, verdict, exact hardware SKU, model revision, workload and before/after result",
+ "produced_artifact": "Ranked matching cases read as precedents and diagnostic options, or a new or deprecated case record in the template; a precedent's procedure and expected direction of effect, not a guaranteed configuration or a measured result",
+ "failure_modes": [
+ "The query is pre-translated into canonical vocabulary or bounded by structured fields, so synonyms in the body text are missed",
+ "A case's specific values, such as batch size, tensor-parallel degree or kernel parameters, are copied instead of re-derived for the current model and hardware",
+ "The precedent's expected direction of effect stands in for measurement after the change is applied",
+ "A rejected direction is read as a permanent blacklist, or one discarded implementation as ruling out its mechanism",
+ "A confounded run, a missing experiment and a failed gate are recorded as the same evidence state",
+ "Cross-system cases are bootstrapped without re-verifying that the mechanism holds on the target hardware"
+ ],
+ "refusal_point": "Adds no case for an INCONCLUSIVE optimization-effect claim, a single-configuration result without plausible generalization, or a result without a real measurement trail, except a scoped diagnostic observation that keeps its verdict and configuration boundary; manufactures no PASS, and reading the records does not reopen a paused campaign"
+ },
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Search the full case bodies with free text first, then rank the hits by bottleneck class and signals.",
+ "Before retrying a rejected direction, state which condition changed or which uncertainty a bounded check resolves, run an unchanged control and choose the least expensive discriminating check.",
+ "Mark bootstrapped cases source_grounded or author_reported until a local measurement confirms or refutes them.",
+ "When a case's conditions no longer hold, mark it deprecated with the reason instead of deleting it."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kernel-writing-guide.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kernel-writing-guide.json
index 8e8626e6a..c4999c702 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kernel-writing-guide.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kernel-writing-guide.json
@@ -39,8 +39,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/methods/kernel-writing-guide/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/methods/kernel-writing-guide/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/methods/kernel-writing-guide/SKILL.md",
"src/inferrsi/data/skill_library/methods/kernel-writing-guide/cases/causal-conv1d-dispatch-state.md",
@@ -79,7 +79,7 @@
"src/inferrsi/data/skill_library/methods/kernel-writing-guide/cases/vllm-gdn-causal-conv-contract.md": "2f3dcee66e2dfe7a98bee21a36b892538998de5de3caf1bc63257137f9984236",
"src/inferrsi/data/skill_library/methods/kernel-writing-guide/cases/vllm-gdn-source-contract.json": "73f1a2d8fa89ce05aead2091a05a285b13cbdc314d05452c35c6a2ad15ffeff7",
"src/inferrsi/data/skill_library/methods/kernel-writing-guide/cases/vllm-gdn-source-contract.md": "7f9ebe8de7e55d3db2a119b6a56e15c37448fd7d4e022a47122a143206d31687",
- "src/inferrsi/data/skill_library/methods/kernel-writing-guide/metadata.json": "c86f93f4eacde789e5438327d85b1de0bfe623fb87acad49306e47b7e8ef1261",
+ "src/inferrsi/data/skill_library/methods/kernel-writing-guide/metadata.json": "766aa688d573c790789e2836f952315e91c4e06806fca3d04523a8a748bf9c32",
"src/inferrsi/data/skill_library/methods/kernel-writing-guide/references/akg-affine-tail-guards.md": "113deb2db10b171db3425bb3f1792bee763706b794c8cc2e222b4cc0d0114118",
"src/inferrsi/data/skill_library/methods/kernel-writing-guide/references/causal-conv1d-contract.md": "b3a6245cf5e9deeba0dffd4a15ddcf97da6a4315f4d3131921779438832663ad",
"src/inferrsi/data/skill_library/methods/kernel-writing-guide/references/hf-qwen35-gdn-binding.md": "848d3c99b47ef711e8820bbc4f6bcbd1db7211197811c9d855201025448d806b",
@@ -251,6 +251,37 @@
}
],
"empirical_validation": "not_run",
+ "applicability": {
+ "scope": "Deciding whether to write a custom inference kernel, routing a kernel pattern to Triton, CuTe DSL, CUDA C++, AscendC or TileIR, applying framework-independent correctness rules, and verifying the executing kernel against a reference before benchmarking; covers fused elementwise chains, reductions, tiled GEMM, attention variants, custom collectives, causal convolution and stateful serving-operator qualification",
+ "prerequisites": "The applicable operator, caller/state and layout contracts are read before substitution, including the causal-conv1d contract and HF Qwen3.5 GDN binding; the CuTe DSL environment uses Linux x86_64, Python 3.10–3.13 and Ampere+ (SM80+) with nvidia-cutlass-dsl installed and no NPU target; the CANN programming guide for the specific SoC is consulted for AscendC",
+ "input_contract": "The candidate operation or operation chain, its shapes, dtypes, quantization and layout requirements, and the target execution phase; the target accelerator and framework constraints; for serving operators, the concrete caller, effective phase and backend, state-row mapping, cache or buffer lifetime and in-place-write behavior; a reference implementation with the same signature, a callable kernel wrapper and fresh CUDA or NPU test tensors; verification shapes and dtypes plus profiler output for the target metric",
+ "produced_artifact": "A write-or-decline decision for the candidate operation with the profiled bottleneck, fusable pair or required dtype, shape or quantization variant unsupported by the standard implementation that justifies it, a framework choice routed by pattern and accelerator, a correctness checklist covering fp32 transcendental round-trips, sync-free wrappers, C division semantics, pre-allocated outputs, fp32 matmul accumulators and masked boundary loads and stores, a starting block or tile size with its autotune and warmup cost noted, a kernel shipping kernel_fn, reference_fn and get_inputs, a reference-parity result across the verification shapes and dtypes, and a post-verification profiler reading on the target metric; for an existing causal-convolution or stateful serving operator, a qualification record naming the caller, phase, backend, state rows and in-place writes, with source selection kept apart from executed dispatch; a routing and correctness decision, not a complete implementation or a measured speedup",
+ "failure_modes": [
+ "A kernel is written for a standalone elementwise op, a standalone matmul or standard attention where the vendor library already wins",
+ "Host-side scheduling cost is treated as a kernel problem instead of being routed to host-overhead-profiling",
+ "Fusion is justified by removed Python call overhead rather than by the DRAM traffic the shared data no longer makes",
+ "A transcendental runs on fp16/bf16 inputs without the fp32 cast round-trip, so the error stays silent through benchmarking",
+ "tl.dot accumulates in the input dtype because Triton is assumed to promote fp16/bf16 to fp32 automatically",
+ "Negative offsets are computed assuming Python floor division while Triton or AscendC rounds toward zero",
+ "An unmasked load or store is accepted because the benchmark shapes happen to divide the block size",
+ "A kernel is benchmarked before it matches its reference, or a shape-boundary, accumulator-dtype or precision failure is read as a tuning problem",
+ "A verification pass is reported as a win although the profiler metric did not move, so the kernel was never on the critical path",
+ "Autotune compile cost is measured as steady-state serving latency instead of being pre-warmed or cached",
+ "A CuTe DSL pattern is routed to an NPU target, or a CUDA tail-guard, attention primitive or SM-matched example is assumed to carry to Ascend",
+ "A causal-conv1d or GDN operator is substituted on build success, import success or matching names without resolving dtype, strides, packed boundaries, state rows and cursors",
+ "An in-tree vLLM wrapper, a GDN prefill flag or a cache-free classification adapter is read as establishing the executed convolution path or generation-state support"
+ ],
+ "refusal_point": "Declines to start a kernel until a profiler puts the target on the critical path near its hardware ceiling, a concrete fusable pair is identified, or the required dtype, shape or quantization variant is unsupported by the standard implementation, and sends host-bound cases elsewhere; refuses to benchmark an unverified kernel and withholds a gain whose profiler metric did not move; keeps operator source selection separate from executed dispatch, numerical parity and hardware qualification, so build or import success qualifies nothing; treats AscendC and TileIR content as source-inferred pending CANN and SoC verification with no local NPU run, treats the AKG Affine FileCheck expectations as IR structure rather than executed CUDA correctness, and reads author-reported overlap and FP8 cases as counterexamples to unconditional tuning rules; claims no Ascend support from NVIDIA primitives",
+ "hardware": "CUDA Ampere+ and Ascend NPU have separate framework routes for each kernel pattern; CuTe DSL requires Linux x86_64, Python 3.10-3.13 and SM80+ with nvidia-cutlass-dsl and is unavailable on NPU; profiling uses ncu on CUDA and msprof on Ascend; the AscendC and TileIR guidance is source-inferred with no local NPU verification and requires validation against the specific CANN version and SoC"
+ },
+ "recommendations": [
+ "Confirm with nsys/ncu or msprof that the target kernel sits on the critical path near its hardware ceiling, name the adjacent operations whose shared data would otherwise round-trip through DRAM, or identify a required dtype, shape or quantization variant unsupported by the standard implementation, before writing any kernel.",
+ "Ship every kernel with kernel_fn, reference_fn and get_inputs, pass reference parity across the verification shapes and dtypes first, and profile afterwards to confirm the target metric actually moved.",
+ "Cast fp16/bf16 to fp32 and back for any math beyond +, -, * and /, declare tl.dot accumulators as tl.float32, and mask every load and store whose shape does not divide the block size.",
+ "Start Triton at BLOCK_SIZE 1024 on Hopper and 512 on Ampere using powers of two, take GEMM tiles from the source skeleton configs (128/256/64 for Hopper, 64/128/32 for smaller cards) and autotune from there.",
+ "On Ascend, reach for TileIR before raw AscendC, size tiles to fit the AI Core Unified Buffer and check cube-unit alignment against the CANN guide for the specific SoC.",
+ "Match a CuTe DSL example to the GPU architecture (SM80 Ampere, SM90 Hopper, SM100 Blackwell) and study its structure instead of copying it, and pre-warm or cache autotuned kernels at startup so first-run compile cost stays out of serving latency."
+ ],
"limitations": [
"Source-derived candidate; no local efficacy validation.",
"Original typed source inventory is not normalized; retain source identities and limits in the procedure/readsets.",
@@ -270,10 +301,36 @@
],
"consumption_mode": "packaged"
},
- "applicability": {},
+ "applicability": {
+ "scope": "Deciding whether to write a custom inference kernel, routing a kernel pattern to Triton, CuTe DSL, CUDA C++, AscendC or TileIR, applying framework-independent correctness rules, and verifying the executing kernel against a reference before benchmarking; covers fused elementwise chains, reductions, tiled GEMM, attention variants, custom collectives, causal convolution and stateful serving-operator qualification",
+ "prerequisites": "The applicable operator, caller/state and layout contracts are read before substitution, including the causal-conv1d contract and HF Qwen3.5 GDN binding; the CuTe DSL environment uses Linux x86_64, Python 3.10–3.13 and Ampere+ (SM80+) with nvidia-cutlass-dsl installed and no NPU target; the CANN programming guide for the specific SoC is consulted for AscendC",
+ "input_contract": "The candidate operation or operation chain, its shapes, dtypes, quantization and layout requirements, and the target execution phase; the target accelerator and framework constraints; for serving operators, the concrete caller, effective phase and backend, state-row mapping, cache or buffer lifetime and in-place-write behavior; a reference implementation with the same signature, a callable kernel wrapper and fresh CUDA or NPU test tensors; verification shapes and dtypes plus profiler output for the target metric",
+ "produced_artifact": "A write-or-decline decision for the candidate operation with the profiled bottleneck, fusable pair or required dtype, shape or quantization variant unsupported by the standard implementation that justifies it, a framework choice routed by pattern and accelerator, a correctness checklist covering fp32 transcendental round-trips, sync-free wrappers, C division semantics, pre-allocated outputs, fp32 matmul accumulators and masked boundary loads and stores, a starting block or tile size with its autotune and warmup cost noted, a kernel shipping kernel_fn, reference_fn and get_inputs, a reference-parity result across the verification shapes and dtypes, and a post-verification profiler reading on the target metric; for an existing causal-convolution or stateful serving operator, a qualification record naming the caller, phase, backend, state rows and in-place writes, with source selection kept apart from executed dispatch; a routing and correctness decision, not a complete implementation or a measured speedup",
+ "failure_modes": [
+ "A kernel is written for a standalone elementwise op, a standalone matmul or standard attention where the vendor library already wins",
+ "Host-side scheduling cost is treated as a kernel problem instead of being routed to host-overhead-profiling",
+ "Fusion is justified by removed Python call overhead rather than by the DRAM traffic the shared data no longer makes",
+ "A transcendental runs on fp16/bf16 inputs without the fp32 cast round-trip, so the error stays silent through benchmarking",
+ "tl.dot accumulates in the input dtype because Triton is assumed to promote fp16/bf16 to fp32 automatically",
+ "Negative offsets are computed assuming Python floor division while Triton or AscendC rounds toward zero",
+ "An unmasked load or store is accepted because the benchmark shapes happen to divide the block size",
+ "A kernel is benchmarked before it matches its reference, or a shape-boundary, accumulator-dtype or precision failure is read as a tuning problem",
+ "A verification pass is reported as a win although the profiler metric did not move, so the kernel was never on the critical path",
+ "Autotune compile cost is measured as steady-state serving latency instead of being pre-warmed or cached",
+ "A CuTe DSL pattern is routed to an NPU target, or a CUDA tail-guard, attention primitive or SM-matched example is assumed to carry to Ascend",
+ "A causal-conv1d or GDN operator is substituted on build success, import success or matching names without resolving dtype, strides, packed boundaries, state rows and cursors",
+ "An in-tree vLLM wrapper, a GDN prefill flag or a cache-free classification adapter is read as establishing the executed convolution path or generation-state support"
+ ],
+ "refusal_point": "Declines to start a kernel until a profiler puts the target on the critical path near its hardware ceiling, a concrete fusable pair is identified, or the required dtype, shape or quantization variant is unsupported by the standard implementation, and sends host-bound cases elsewhere; refuses to benchmark an unverified kernel and withholds a gain whose profiler metric did not move; keeps operator source selection separate from executed dispatch, numerical parity and hardware qualification, so build or import success qualifies nothing; treats AscendC and TileIR content as source-inferred pending CANN and SoC verification with no local NPU run, treats the AKG Affine FileCheck expectations as IR structure rather than executed CUDA correctness, and reads author-reported overlap and FP8 cases as counterexamples to unconditional tuning rules; claims no Ascend support from NVIDIA primitives",
+ "hardware": "CUDA Ampere+ and Ascend NPU have separate framework routes for each kernel pattern; CuTe DSL requires Linux x86_64, Python 3.10-3.13 and SM80+ with nvidia-cutlass-dsl and is unavailable on NPU; profiling uses ncu on CUDA and msprof on Ascend; the AscendC and TileIR guidance is source-inferred with no local NPU verification and requires validation against the specific CANN version and SoC"
+ },
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Confirm with nsys/ncu or msprof that the target kernel sits on the critical path near its hardware ceiling, name the adjacent operations whose shared data would otherwise round-trip through DRAM, or identify a required dtype, shape or quantization variant unsupported by the standard implementation, before writing any kernel.",
+ "Ship every kernel with kernel_fn, reference_fn and get_inputs, pass reference parity across the verification shapes and dtypes first, and profile afterwards to confirm the target metric actually moved.",
+ "Cast fp16/bf16 to fp32 and back for any math beyond +, -, * and /, declare tl.dot accumulators as tl.float32, and mask every load and store whose shape does not divide the block size.",
+ "Start Triton at BLOCK_SIZE 1024 on Hopper and 512 on Ampere using powers of two, take GEMM tiles from the source skeleton configs (128/256/64 for Hopper, 64/128/32 for smaller cards) and autotune from there.",
+ "On Ascend, reach for TileIR before raw AscendC, size tiles to fit the AI Core Unified Buffer and check cube-unit alignment against the CANN guide for the specific SoC.",
+ "Match a CuTe DSL example to the GPU architecture (SM80 Ampere, SM90 Hopper, SM100 Blackwell) and study its structure instead of copying it, and pre-warm or cache autotuned kernels at startup so first-run compile cost stays out of serving latency."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kv-tier-management.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kv-tier-management.json
index f09aca89f..9ffb79691 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kv-tier-management.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kv-tier-management.json
@@ -24,8 +24,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/repositories/lmcache/kv-tier-management/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/repositories/lmcache/kv-tier-management/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/repositories/lmcache/kv-tier-management/SKILL.md",
"src/inferrsi/data/skill_library/repositories/lmcache/kv-tier-management/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lazy-evaluation-stream-lifetimes.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lazy-evaluation-stream-lifetimes.json
index baf15822e..37c1ec38c 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lazy-evaluation-stream-lifetimes.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lazy-evaluation-stream-lifetimes.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/SKILL.md",
"src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lily.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lily.json
index 7bad127d3..46b15dd95 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lily.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lily.json
@@ -25,8 +25,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/repositories/lily/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/repositories/lily/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/repositories/lily/LICENSE",
"src/inferrsi/data/skill_library/repositories/lily/NOTICE.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lithos-metal.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lithos-metal.json
index 73a0d5000..2fbc1e211 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lithos-metal.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lithos-metal.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/repositories/lithos-metal/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/repositories/lithos-metal/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/repositories/lithos-metal/LICENSE",
"src/inferrsi/data/skill_library/repositories/lithos-metal/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/local-dream.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/local-dream.json
index 7d3f5719f..2c1afedd7 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/local-dream.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/local-dream.json
@@ -29,8 +29,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/repositories/local-dream/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/repositories/local-dream/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/repositories/local-dream/SKILL.md",
"src/inferrsi/data/skill_library/repositories/local-dream/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mistral-rs.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mistral-rs.json
index 1a3eafe8e..97b18bf21 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mistral-rs.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mistral-rs.json
@@ -26,8 +26,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/repositories/mistral-rs/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/repositories/mistral-rs/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/repositories/mistral-rs/LICENSE",
"src/inferrsi/data/skill_library/repositories/mistral-rs/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlc-llm.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlc-llm.json
index aa435a5c8..065eeab0d 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlc-llm.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlc-llm.json
@@ -27,8 +27,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/repositories/mlc-llm/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/repositories/mlc-llm/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/repositories/mlc-llm/LICENSE",
"src/inferrsi/data/skill_library/repositories/mlc-llm/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlx-swift-lm.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlx-swift-lm.json
index b6d2effd3..33298fa72 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlx-swift-lm.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlx-swift-lm.json
@@ -27,8 +27,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/repositories/mlx-swift-lm/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/repositories/mlx-swift-lm/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/repositories/mlx-swift-lm/LICENSE",
"src/inferrsi/data/skill_library/repositories/mlx-swift-lm/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/npu-skill-factory.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/npu-skill-factory.json
index 646bd2108..995c075fc 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/npu-skill-factory.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/npu-skill-factory.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/workflows/npu-skill-factory/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/workflows/npu-skill-factory/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/skill-index.json",
"src/inferrsi/data/skill_library/workflows/npu-skill-factory/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/optimization-search-policy.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/optimization-search-policy.json
index 4ad246ad3..38c73f365 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/optimization-search-policy.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/optimization-search-policy.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/methods/optimization-search-policy/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/methods/optimization-search-policy/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/methods/optimization-search-policy/SKILL.md",
"src/inferrsi/data/skill_library/methods/optimization-search-policy/cases/sarathi-chunk-budget.json",
@@ -50,7 +50,7 @@
"src/inferrsi/data/skill_library/methods/optimization-search-policy/SKILL.md": "0ba4ade1648764a7422f564abe2d4fc9f9a25b5943a6c70c81ce17b8e7fb48d9",
"src/inferrsi/data/skill_library/methods/optimization-search-policy/cases/sarathi-chunk-budget.json": "f8448973cf1c9e506179c02c5e55bd804ea039b7276cf08071952f6c0be656ad",
"src/inferrsi/data/skill_library/methods/optimization-search-policy/cases/sarathi-chunk-budget.md": "2c2a8147001d24d1abc8672f15cf0ca759ae461b734413c8dcb7497634bb6821",
- "src/inferrsi/data/skill_library/methods/optimization-search-policy/metadata.json": "d5aaaa4511c9287e1f3d1f07a25c36320a68cb35a5066778cbd3c2bbb5203c6a",
+ "src/inferrsi/data/skill_library/methods/optimization-search-policy/metadata.json": "6dc0c3b8ce13cff285ad99da4441f2fb6a5931247e7634e657275dd915854e02",
"src/inferrsi/data/skill_library/methods/optimization-search-policy/references/distserve-phase-placement/README.md": "56e16efce447dfd4990a024ff1d6f7cb428f8e3e1f7b32a7e4d05f57139b836a",
"src/inferrsi/data/skill_library/methods/optimization-search-policy/references/distserve-phase-placement/source-record.json": "21cc9562f1c19d6b57957c01b53e0adf8700d05ffaaa7764afe07c4da1284de1",
"src/inferrsi/data/skill_library/methods/optimization-search-policy/references/papers-20261015/INTEGRATION_NOTES.md": "157e74ba180426b279b548a729769f152bf18f9a78ffca4ebfcaf26b7d0c5292",
@@ -78,25 +78,27 @@
"external"
],
"empirical_validation": "not_run",
- "sources": [
- {
- "type": "paper",
- "id": "usenix:osdi24:zhong-yinmin",
- "version": "OSDI 24 proceedings, July 2024, pages 193-210",
- "url": "https://www.usenix.org/conference/osdi24/presentation/zhong-yinmin",
- "sections": [
- "3.3",
- "4.1-4.2",
- "6.1-6.5",
- "7",
- "Appendices A-C"
- ],
- "readset": "references/distserve-phase-placement/source-record.json"
- },
- {
- "type": "unknown",
- "scope": "Pre-existing sources remain in their original procedure/readsets; this amendment does not normalize that inventory"
- }
+ "applicability": {
+ "scope": "Selecting, ordering and stopping optimization experiments: candidate generation, exploration versus exploitation, mechanism diversity, negative experiment memory, cost-aware escalation, budget allocation and stopping criteria, drawn from vllm-evolve, the TRT-LLM casebook and SkillOpt",
+ "prerequisites": "The useful work, quality contract, allowed approximations, timing boundary and resource budget are frozen before search, and a complete restartable incumbent is preserved; reading the procedures authorizes no execution, framework change, new resource spending or publication",
+ "input_contract": "Profiling data, existing config, prior experiment and negative records and the casebook for the current bottleneck class, model and hardware; each target stage's fraction of E2E time; the target, locally measured uncertainty and the authorized budget",
+ "produced_artifact": "An ordered experiment plan: one candidate change per round with an explicit hypothesis and its motivating evidence, a bounded shortlist of distinct mechanisms with assumptions, uncertainty, expected impact and evaluation cost, negative records turned into search constraints, and a stop-or-switch decision; when authorized rounds run, a recorded verdict per round with best-seen and delivered outcomes kept apart",
+ "failure_modes": [
+ "The evaluation protocol, measurement setup or baseline changes in the same round as the candidate",
+ "Benchmark budget is spent on a candidate that fails correctness",
+ "A matching keyword in a rejection record is used as a causal diagnosis or automatic veto, or one failed implementation excludes a whole solution family",
+ "Expensive work starts on a stage without computing its Amdahl contribution, so a large local speedup yields little E2E gain",
+ "One mechanism family keeps being sampled only because budget remains, without re-profiling after the limiting stage moves",
+ "Roofline proximity or a utilization threshold is used as a universal stopping rule",
+ "Screening is allowed to skip the frozen acceptance regimes or required service-level evidence"
+ ],
+ "refusal_point": "Discards a correctness-breaking, worse or inconclusive change instead of integrating it, preserves no-change when candidates cannot be separated, does not count a shortcut outside the authorized semantics as an optimization, and does not let a successful probe qualify a broken final state"
+ },
+ "recommendations": [
+ "Before generating a candidate, consult the casebook, the negative record and the target stage's Amdahl contribution.",
+ "Record every worse or inconclusive result with hypothesis, metric, context, evidence level and failure type, and derive a search constraint from it.",
+ "When about three experiments on one hypothesis return inconclusive from measurement variance, fix the protocol or move to a different hypothesis.",
+ "Before reporting, restore and verify the best fully valid incumbent through its canonical launcher."
],
"limitations": [
"Source-derived candidate; no local efficacy validation.",
@@ -129,12 +131,49 @@
"../performance-ceiling-analysis/SKILL.md",
"../../repositories/vllm/audit-vllm-kv-lifetimes/SKILL.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "sources": [
+ {
+ "type": "paper",
+ "id": "usenix:osdi24:zhong-yinmin",
+ "version": "OSDI 24 proceedings, July 2024, pages 193-210",
+ "url": "https://www.usenix.org/conference/osdi24/presentation/zhong-yinmin",
+ "sections": [
+ "3.3",
+ "4.1-4.2",
+ "6.1-6.5",
+ "7",
+ "Appendices A-C"
+ ],
+ "readset": "references/distserve-phase-placement/source-record.json"
+ },
+ {
+ "type": "unknown",
+ "scope": "Pre-existing sources remain in their original procedure/readsets; this amendment does not normalize that inventory"
+ }
+ ]
+ },
+ "applicability": {
+ "scope": "Selecting, ordering and stopping optimization experiments: candidate generation, exploration versus exploitation, mechanism diversity, negative experiment memory, cost-aware escalation, budget allocation and stopping criteria, drawn from vllm-evolve, the TRT-LLM casebook and SkillOpt",
+ "prerequisites": "The useful work, quality contract, allowed approximations, timing boundary and resource budget are frozen before search, and a complete restartable incumbent is preserved; reading the procedures authorizes no execution, framework change, new resource spending or publication",
+ "input_contract": "Profiling data, existing config, prior experiment and negative records and the casebook for the current bottleneck class, model and hardware; each target stage's fraction of E2E time; the target, locally measured uncertainty and the authorized budget",
+ "produced_artifact": "An ordered experiment plan: one candidate change per round with an explicit hypothesis and its motivating evidence, a bounded shortlist of distinct mechanisms with assumptions, uncertainty, expected impact and evaluation cost, negative records turned into search constraints, and a stop-or-switch decision; when authorized rounds run, a recorded verdict per round with best-seen and delivered outcomes kept apart",
+ "failure_modes": [
+ "The evaluation protocol, measurement setup or baseline changes in the same round as the candidate",
+ "Benchmark budget is spent on a candidate that fails correctness",
+ "A matching keyword in a rejection record is used as a causal diagnosis or automatic veto, or one failed implementation excludes a whole solution family",
+ "Expensive work starts on a stage without computing its Amdahl contribution, so a large local speedup yields little E2E gain",
+ "One mechanism family keeps being sampled only because budget remains, without re-profiling after the limiting stage moves",
+ "Roofline proximity or a utilization threshold is used as a universal stopping rule",
+ "Screening is allowed to skip the frozen acceptance regimes or required service-level evidence"
+ ],
+ "refusal_point": "Discards a correctness-breaking, worse or inconclusive change instead of integrating it, preserves no-change when candidates cannot be separated, does not count a shortcut outside the authorized semantics as an optimization, and does not let a successful probe qualify a broken final state"
},
- "applicability": {},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Before generating a candidate, consult the casebook, the negative record and the target stage's Amdahl contribution.",
+ "Record every worse or inconclusive result with hypothesis, metric, context, evidence level and failure type, and derive a search constraint from it.",
+ "When about three experiments on one hypothesis return inconclusive from measurement variance, fix the protocol or move to a different hypothesis.",
+ "Before reporting, restore and verify the best fully valid incumbent through its canonical launcher."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/os-process-placement.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/os-process-placement.json
index 2e1a0ab3a..9b9fe366f 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/os-process-placement.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/os-process-placement.json
@@ -24,8 +24,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/hardware/host/os-process-placement/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/hardware/host/os-process-placement/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/hardware/host/os-process-placement/SKILL.md",
"src/inferrsi/data/skill_library/hardware/host/os-process-placement/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/performance-ceiling-analysis.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/performance-ceiling-analysis.json
index bb08f01fb..d74d01244 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/performance-ceiling-analysis.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/performance-ceiling-analysis.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/SKILL.md",
"src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/plan-activation-residency.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/plan-activation-residency.json
index 4dcb0dc4b..77fead2ce 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/plan-activation-residency.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/plan-activation-residency.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/methods/plan-activation-residency/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/methods/plan-activation-residency/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/methods/plan-activation-residency/SKILL.md",
"src/inferrsi/data/skill_library/methods/plan-activation-residency/metadata.json",
@@ -34,7 +34,7 @@
"checkout_ref": "checkout:plan-activation-residency",
"source_files": {
"src/inferrsi/data/skill_library/methods/plan-activation-residency/SKILL.md": "542d8a1e6963f5019986e1fbd79d3a247a045a51487c09313026087ac457c15e",
- "src/inferrsi/data/skill_library/methods/plan-activation-residency/metadata.json": "c6acae57d70190cea0e18c71c79127962181e58169e011df75d8635b3db4daa7",
+ "src/inferrsi/data/skill_library/methods/plan-activation-residency/metadata.json": "0289c62f93d087cd6a4ba4140487925c4b3258a9db1f7004ff1dfb22fd6d4582",
"src/inferrsi/data/skill_library/methods/plan-activation-residency/references/source-notes.md": "5ff52655cfd06fcf970ed2a7d4d4fe1c31115b102af2d41497de0c74848ed2b3",
"src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
},
@@ -49,6 +49,27 @@
"origin": [
"external"
],
+ "applicability": {
+ "scope": "Proposing training activation and gradient memory reductions when pressure appears to come from live tensors or allocator-reserved memory, by distinguishing lifetime reordering, address packing and selective rematerialization; covers activation checkpointing for recomputation, not persistent recovery checkpoints",
+ "prerequisites": "The packaged evidence and limits are read before citing paper results; installing or running a solver or framework, modifying allocation or checkpoint policies, reproducing workloads or provisioning hardware needs separate authorization",
+ "input_contract": "The actual forward/backward/update DAG, shapes and precision, tensor sizes and consumer lifetimes; resident versus reserved peak bytes; operator, workspace and temporary-output peaks; batch and accumulation policy; optimizer and collective barriers; representative compute and transfer costs; whether the goal is fit, step time or time to a fixed training objective",
+ "produced_artifact": "A preferred candidate, simpler fallback, resource ledger, critical-path prediction, assumptions, counter-signals and unresolved-evidence list, with source claims kept apart from proposed decisions, plus a proposed matched comparison; a reviewable proposal, not an implementation",
+ "failure_modes": [
+ "Fragmentation and live-byte excess are not distinguished at the same phase",
+ "OLLA address packing is read as tensor placement across CPU/SSD tiers",
+ "Reordering moves a release or update past a dependency such as shared weights, accumulation, gradient clipping, optimizer state or collectives",
+ "A static packing plan ignores alignment, aliasing, workspace, asynchronous overlap or shape variability",
+ "Added recomputation consumes the compute window that previously hid transfers",
+ "Checkmate's rounding is taken to guarantee the memory cap, or its 10% allowance is used as a portable safety margin",
+ "A solver timeout is read as proof of infeasibility"
+ ],
+ "refusal_point": "Returns the missing classification instead of a preferred checkpoint interval when only total OOM memory is known, and escalates when a legal dependency graph, memory margin or correctness model is unavailable"
+ },
+ "recommendations": [
+ "Where ordering is legal, try earlier last-use and gradient release before adding computation or transfer.",
+ "Check dependency validity and memory feasibility separately for an approximate schedule.",
+ "Account for planning cost over the expected number of repeated steps."
+ ],
"limitations": [
"Source-derived candidate; no local efficacy validation.",
"Original typed source inventory is not normalized; retain source identities and limits in the procedure/readsets."
@@ -58,10 +79,26 @@
],
"consumption_mode": "packaged"
},
- "applicability": {},
+ "applicability": {
+ "scope": "Proposing training activation and gradient memory reductions when pressure appears to come from live tensors or allocator-reserved memory, by distinguishing lifetime reordering, address packing and selective rematerialization; covers activation checkpointing for recomputation, not persistent recovery checkpoints",
+ "prerequisites": "The packaged evidence and limits are read before citing paper results; installing or running a solver or framework, modifying allocation or checkpoint policies, reproducing workloads or provisioning hardware needs separate authorization",
+ "input_contract": "The actual forward/backward/update DAG, shapes and precision, tensor sizes and consumer lifetimes; resident versus reserved peak bytes; operator, workspace and temporary-output peaks; batch and accumulation policy; optimizer and collective barriers; representative compute and transfer costs; whether the goal is fit, step time or time to a fixed training objective",
+ "produced_artifact": "A preferred candidate, simpler fallback, resource ledger, critical-path prediction, assumptions, counter-signals and unresolved-evidence list, with source claims kept apart from proposed decisions, plus a proposed matched comparison; a reviewable proposal, not an implementation",
+ "failure_modes": [
+ "Fragmentation and live-byte excess are not distinguished at the same phase",
+ "OLLA address packing is read as tensor placement across CPU/SSD tiers",
+ "Reordering moves a release or update past a dependency such as shared weights, accumulation, gradient clipping, optimizer state or collectives",
+ "A static packing plan ignores alignment, aliasing, workspace, asynchronous overlap or shape variability",
+ "Added recomputation consumes the compute window that previously hid transfers",
+ "Checkmate's rounding is taken to guarantee the memory cap, or its 10% allowance is used as a portable safety margin",
+ "A solver timeout is read as proof of infeasibility"
+ ],
+ "refusal_point": "Returns the missing classification instead of a preferred checkpoint interval when only total OOM memory is known, and escalates when a legal dependency graph, memory margin or correctness model is unavailable"
+ },
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Where ordering is legal, try earlier last-use and gradient release before adding computation or transfer.",
+ "Check dependency validity and memory feasibility separately for an approximate schedule.",
+ "Account for planning cost over the expected number of repeated steps."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/project-maya.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/project-maya.json
index 4f2588d8f..cb38b28d4 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/project-maya.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/project-maya.json
@@ -25,8 +25,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/repositories/project-maya/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/repositories/project-maya/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/repositories/project-maya/LICENSE",
"src/inferrsi/data/skill_library/repositories/project-maya/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/repository-evolution-analysis.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/repository-evolution-analysis.json
index 3f66326e2..3edca3450 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/repository-evolution-analysis.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/repository-evolution-analysis.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/methods/repository-evolution-analysis/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/methods/repository-evolution-analysis/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/methods/repository-evolution-analysis/SKILL.md",
"src/inferrsi/data/skill_library/methods/repository-evolution-analysis/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/rsi-evidence.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/rsi-evidence.json
index 29b03048f..8d542abf9 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/rsi-evidence.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/rsi-evidence.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/workflows/rsi-evidence/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/workflows/rsi-evidence/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/skill-index.json",
"src/inferrsi/data/skill_library/workflows/rsi-evidence/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/skill-evolution-playbook.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/skill-evolution-playbook.json
index 2929f3f3b..b1b494df5 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/skill-evolution-playbook.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/skill-evolution-playbook.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/methods/skill-evolution-playbook/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/methods/skill-evolution-playbook/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/methods/skill-evolution-playbook/SKILL.md",
"src/inferrsi/data/skill_library/methods/skill-evolution-playbook/metadata.json",
@@ -38,7 +38,7 @@
"checkout_ref": "checkout:skill-evolution-playbook",
"source_files": {
"src/inferrsi/data/skill_library/methods/skill-evolution-playbook/SKILL.md": "9cd37f21fe0d862069e4798edca8a04d59f8b840e55d521901594a63dddd9ba3",
- "src/inferrsi/data/skill_library/methods/skill-evolution-playbook/metadata.json": "92b36338b9257e2e41b95e4370694d7de52ed84bc7971056b3bfced07b0993d8",
+ "src/inferrsi/data/skill_library/methods/skill-evolution-playbook/metadata.json": "d3e790614e09d2d57acb24bfe8d2c2a524cc36bd667bd3867c557554bcbb605e",
"src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
},
"skill_metadata": {
@@ -58,6 +58,28 @@
"external"
],
"empirical_validation": "not_run",
+ "applicability": {
+ "scope": "Design choices for evolving agent skills from execution traces, drawn from Trace2Skill and SkillOpt: when a trajectory should update a skill, how conflicting lessons are handled, how negative evidence is preserved, how large an edit should be and when skills split or merge; for designing or operating a skill evolution pipeline",
+ "prerequisites": "A final-test partition kept outside edit selection and the incumbent skill retained for rollback; changing a skill's scope first needs explicit review of its identity and consumers",
+ "input_contract": "Execution trajectories or trace IDs with the diagnosed causal failure and the successful behavior an amendment must preserve; the incumbent skill and its revision lineage; held-out evaluation results for candidate edits",
+ "produced_artifact": "A scoped case or amendment proposal, or a consolidated candidate edit, with rejection records and the versioned revision lineage; a candidate for independently evaluated acceptance, not a replacement decision or an efficacy claim",
+ "failure_modes": [
+ "An anecdote, or agreement among analysts, is taken as efficacy",
+ "Consolidation output is treated as a replacement decision instead of a candidate awaiting independent acceptance",
+ "A tie or uncertain result is reported as improvement",
+ "A conflict is resolved by voting for whichever outcome appeared more often in the pool",
+ "A rejection is applied to every future use of an idea instead of the comparison it constrains",
+ "Paper-specific section counts or length budgets are used as universal defaults, or a skill is split on length alone",
+ "A stronger task solver is assumed to be a stronger improver, or replay rankings and private inner scores are read as outer generalization"
+ ],
+ "refusal_point": "Records a conflict that conditioning cannot resolve, where the same context produces different outcomes across runs, as INCONCLUSIVE and flags it for human review instead of choosing a lesson; claims no universal minimum trace count, paper-to-code equivalence or local efficacy"
+ },
+ "recommendations": [
+ "Condition a conflicting lesson on its context (when X, do A; when Y, avoid A) before choosing one side.",
+ "Store each rejection record with its evaluation evidence, outside the deployed task procedure, and give it a retirement condition.",
+ "Bound the target passage and intended behavior before editing, inspect the actual diff and preserve rollback state.",
+ "Prefer an amendment when the existing skill owns the workflow; create a new candidate only for a genuinely distinct capability."
+ ],
"limitations": [
"Source-derived candidate; no local efficacy validation.",
"Original typed source inventory is not normalized; retain source identities and limits in the procedure/readsets."
@@ -65,10 +87,27 @@
"provenance_readsets": [],
"consumption_mode": "packaged"
},
- "applicability": {},
+ "applicability": {
+ "scope": "Design choices for evolving agent skills from execution traces, drawn from Trace2Skill and SkillOpt: when a trajectory should update a skill, how conflicting lessons are handled, how negative evidence is preserved, how large an edit should be and when skills split or merge; for designing or operating a skill evolution pipeline",
+ "prerequisites": "A final-test partition kept outside edit selection and the incumbent skill retained for rollback; changing a skill's scope first needs explicit review of its identity and consumers",
+ "input_contract": "Execution trajectories or trace IDs with the diagnosed causal failure and the successful behavior an amendment must preserve; the incumbent skill and its revision lineage; held-out evaluation results for candidate edits",
+ "produced_artifact": "A scoped case or amendment proposal, or a consolidated candidate edit, with rejection records and the versioned revision lineage; a candidate for independently evaluated acceptance, not a replacement decision or an efficacy claim",
+ "failure_modes": [
+ "An anecdote, or agreement among analysts, is taken as efficacy",
+ "Consolidation output is treated as a replacement decision instead of a candidate awaiting independent acceptance",
+ "A tie or uncertain result is reported as improvement",
+ "A conflict is resolved by voting for whichever outcome appeared more often in the pool",
+ "A rejection is applied to every future use of an idea instead of the comparison it constrains",
+ "Paper-specific section counts or length budgets are used as universal defaults, or a skill is split on length alone",
+ "A stronger task solver is assumed to be a stronger improver, or replay rankings and private inner scores are read as outer generalization"
+ ],
+ "refusal_point": "Records a conflict that conditioning cannot resolve, where the same context produces different outcomes across runs, as INCONCLUSIVE and flags it for human review instead of choosing a lesson; claims no universal minimum trace count, paper-to-code equivalence or local efficacy"
+ },
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Condition a conflicting lesson on its context (when X, do A; when Y, avoid A) before choosing one side.",
+ "Store each rejection record with its evaluation evidence, outside the deployed task procedure, and give it a retirement condition.",
+ "Bound the target passage and intended behavior before editing, inspect the actual diff and preserve rollback state.",
+ "Prefer an amendment when the existing skill owns the workflow; create a new candidate only for a genuinely distinct capability."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/splash.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/splash.json
index 2c920a600..7ce820099 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/splash.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/splash.json
@@ -25,8 +25,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/repositories/splash/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/repositories/splash/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/repositories/splash/LICENSE",
"src/inferrsi/data/skill_library/repositories/splash/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/storage-io-measurement.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/storage-io-measurement.json
index 430810584..66c9defc1 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/storage-io-measurement.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/storage-io-measurement.json
@@ -24,8 +24,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/hardware/host/storage-io-measurement/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/hardware/host/storage-io-measurement/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/hardware/host/storage-io-measurement/SKILL.md",
"src/inferrsi/data/skill_library/hardware/host/storage-io-measurement/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-input-pipeline.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-input-pipeline.json
index be9f0ab60..644f8c2a1 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-input-pipeline.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-input-pipeline.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/methods/training-input-pipeline/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/methods/training-input-pipeline/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/methods/training-input-pipeline/SKILL.md",
"src/inferrsi/data/skill_library/methods/training-input-pipeline/metadata.json",
@@ -41,7 +41,7 @@
"checkout_ref": "checkout:training-input-pipeline",
"source_files": {
"src/inferrsi/data/skill_library/methods/training-input-pipeline/SKILL.md": "64bd2c6b795d966045d1d3cca55976d3d49b08b06df7c7e06a70c2824cf55b7e",
- "src/inferrsi/data/skill_library/methods/training-input-pipeline/metadata.json": "80659099fce6a6ec05e9084b75452c877a5812e30f60a78f31ef8c7efcdc1bac",
+ "src/inferrsi/data/skill_library/methods/training-input-pipeline/metadata.json": "debad47896f600c7a1e19e477e34026c975dfde320a786dca1ddd02030a5feb1",
"src/inferrsi/data/skill_library/methods/training-input-pipeline/references/sources.md": "60e0f3e88655d8e1d5eb4797f39ca4370deda01614f265b8d2ea4559d8ac020f",
"src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
},
@@ -64,6 +64,28 @@
"external"
],
"empirical_validation": "not_run",
+ "applicability": {
+ "scope": "Diagnosing training input stalls and choosing bounded queue, shard-cache and cache-ownership changes from DALI, Mosaic Streaming and FFCV, while preserving execution-mode gates, sample ordering and per-host memory budgets and separating remote downloads, local storage reads and host/device transfers",
+ "prerequisites": "The pinned readset and its license constraints are read; reading authorizes no commands, cache manipulation or experiment",
+ "input_contract": "Dataset format, sample-size distribution, workers and ranks per host, ordering/shuffle, local-cache state and executor mode; a per-host memory budget that includes checkpoint staging and runtime offload; timings that separate remote fetch, decompression/decoding, transformation, batch readiness and device transfer",
+ "produced_artifact": "One proposed queue, shard-cache or cache-ownership change with its source-backed rule and boundary, the evidence proposed to confirm it, and a verification contract for a separately authorized comparison; no input-throughput, model-quality or hardware result",
+ "failure_modes": [
+ "Low accelerator utilization is taken to identify which producer stage is late",
+ "Missing legacy DALI statistics are read as zero use, or the execution mode is switched silently to obtain them, which changes the comparison",
+ "DALI run() returning is read as output-ready timing instead of the outputs() synchronization boundary",
+ "Mosaic predownload is treated as a byte cap or request count, or per-worker sample targets are multiplied into an exact network-byte prediction",
+ "Look-ahead is increased when the local cache cannot retain the working set, so evicted shards are downloaded again",
+ "cache_limit is read as a bound on total process RAM, or the constructor minimum as an optimum or a guarantee against thrashing",
+ "An FFCV memory map is taken as proof that its pages are resident"
+ ],
+ "refusal_point": "Rejects changes that violate sample coverage or the stated reproducibility contract even if batches arrive faster, does not report queue depth alone as a performance result, and draws no NPU acceleration claim from the CUDA-oriented DALI source"
+ },
+ "recommendations": [
+ "Record dataset format, sample sizes, workers per host, ordering, cache state and executor mode before proposing one change.",
+ "Increase Mosaic look-ahead only when readiness starvation is observed and cache headroom exists; otherwise address shard and working-set residency first.",
+ "Check the exact DALI configuration surface before use, because the constructor and deserialization paths use different queue keys.",
+ "In a later comparison, separate warm-cache throughput from cold retrieval and initialization, and report step time with input wait, host/HBM peaks and actual I/O."
+ ],
"limitations": [
"Source-derived candidate; no local efficacy validation.",
"Original typed source inventory is not normalized; retain source identities and limits in the procedure/readsets."
@@ -73,10 +95,27 @@
],
"consumption_mode": "packaged"
},
- "applicability": {},
+ "applicability": {
+ "scope": "Diagnosing training input stalls and choosing bounded queue, shard-cache and cache-ownership changes from DALI, Mosaic Streaming and FFCV, while preserving execution-mode gates, sample ordering and per-host memory budgets and separating remote downloads, local storage reads and host/device transfers",
+ "prerequisites": "The pinned readset and its license constraints are read; reading authorizes no commands, cache manipulation or experiment",
+ "input_contract": "Dataset format, sample-size distribution, workers and ranks per host, ordering/shuffle, local-cache state and executor mode; a per-host memory budget that includes checkpoint staging and runtime offload; timings that separate remote fetch, decompression/decoding, transformation, batch readiness and device transfer",
+ "produced_artifact": "One proposed queue, shard-cache or cache-ownership change with its source-backed rule and boundary, the evidence proposed to confirm it, and a verification contract for a separately authorized comparison; no input-throughput, model-quality or hardware result",
+ "failure_modes": [
+ "Low accelerator utilization is taken to identify which producer stage is late",
+ "Missing legacy DALI statistics are read as zero use, or the execution mode is switched silently to obtain them, which changes the comparison",
+ "DALI run() returning is read as output-ready timing instead of the outputs() synchronization boundary",
+ "Mosaic predownload is treated as a byte cap or request count, or per-worker sample targets are multiplied into an exact network-byte prediction",
+ "Look-ahead is increased when the local cache cannot retain the working set, so evicted shards are downloaded again",
+ "cache_limit is read as a bound on total process RAM, or the constructor minimum as an optimum or a guarantee against thrashing",
+ "An FFCV memory map is taken as proof that its pages are resident"
+ ],
+ "refusal_point": "Rejects changes that violate sample coverage or the stated reproducibility contract even if batches arrive faster, does not report queue depth alone as a performance result, and draws no NPU acceleration claim from the CUDA-oriented DALI source"
+ },
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Record dataset format, sample sizes, workers per host, ordering, cache state and executor mode before proposing one change.",
+ "Increase Mosaic look-ahead only when readiness starvation is observed and cache headroom exists; otherwise address shard and working-set residency first.",
+ "Check the exact DALI configuration surface before use, because the constructor and deserialization paths use different queue keys.",
+ "In a later comparison, separate warm-cache throughput from cold retrieval and initialization, and report step time with input wait, host/HBM peaks and actual I/O."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-resource-budget.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-resource-budget.json
index e3482f0f0..efe943ba2 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-resource-budget.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-resource-budget.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/methods/training-resource-budget/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/methods/training-resource-budget/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/methods/training-resource-budget/SKILL.md",
"src/inferrsi/data/skill_library/methods/training-resource-budget/metadata.json",
@@ -49,7 +49,7 @@
"checkout_ref": "checkout:training-resource-budget",
"source_files": {
"src/inferrsi/data/skill_library/methods/training-resource-budget/SKILL.md": "d3805c5f605faae663ad5a674a56c1e1e5fdd211cd58833c008e557cfd7e4c37",
- "src/inferrsi/data/skill_library/methods/training-resource-budget/metadata.json": "adb0473945920794e07f62e19230683e6f8b2a0f07f91d41a451374f919d343b",
+ "src/inferrsi/data/skill_library/methods/training-resource-budget/metadata.json": "9ae014b0ca5ca5d32e31e05b52e7436ad1c197fdce8d731fd4d9c79965fd4f2d",
"src/inferrsi/data/skill_library/methods/training-resource-budget/references/mindspeed-historical.md": "ad65c27f278c3d0ca3599a7630c02d3e529477a05b71b7703bd2d75f0d8fdd2c",
"src/inferrsi/data/skill_library/methods/training-resource-budget/references/mindspeed-source-readset.md": "4462d69ed6a941a03639c756968be3ce54f947e51f6ca618f4fb1cd43b643b35",
"src/inferrsi/data/skill_library/methods/training-resource-budget/references/papers-20261015/INTEGRATION_NOTES.md": "c55c6ad2131ef788866411dcc7056f221420ebff30c95c6bc919b09a4818cb73",
@@ -80,6 +80,28 @@
"external"
],
"empirical_validation": "not_run",
+ "applicability": {
+ "scope": "Choosing training offload and resource-budget changes from pinned DeepSpeed configuration, attributed expert diagnosis and a historical MindSpeed case, balancing accelerator capacity, CPU work, pinned RAM and local storage while keeping host/device transfer exposure distinct from collective network bottlenecks",
+ "prerequisites": "The source boundaries are read; garbage-collection or affinity changes, privileged memlock changes and delayed parameter updates each need separate approval, and the skill authorizes no newest-wheel installation, scheduler-priority change, benchmark or failure injection; historical MindSpeed reuse requires compatible allocator symbols, tracked swappable storage, sufficient host capacity and verified device/NUMA/CPU mapping, not merely the presence of npu-smi and lspci",
+ "input_contract": "State attributed to parameters, optimizer state, gradients or activations with its current memory tier, lifetime and next consumer; whether the failure is allocation capacity or throughput; a per-host budget for input buffers, checkpoint snapshots, offloaded state, OS cache and CPU workers with the actual ranks sharing the host; the host/device link recorded apart from collective traffic; the exact source version and enabled execution mode",
+ "produced_artifact": "A candidate decision record per proposal: the observed limiting interval and owning state or resource, source version and execution mode, capacity released and resources consumed with units and uncertainty, overlap readiness and mutation boundaries with expected contention, and the fixed-workload step and correctness evidence required for later validation",
+ "failure_modes": [
+ "A change that lets the model fit is presented as a speedup, although it can slow each step",
+ "A tensor copied to CPU is taken as an InfiniBand/RoCE bottleneck, or CPU/NVMe offload is counted as network evidence",
+ "Parameter max_in_cpu is read as a byte or total-host-memory cap instead of an element count",
+ "NVMe optimizer placement is assumed to remove CPU optimizer demand",
+ "The parameter buffer-pool count is transplanted to the optimizer, or all optimizers are assumed to hold identical states",
+ "Configuration presence or schema documentation is taken as evidence that a knob or overlap actually ran",
+ "Ordinary overlap is taken as permission for delayed, stale parameter updates"
+ ],
+ "refusal_point": "Claims no hardware validation, measured gain or current NPU compatibility, keeps source support and extracted coverage separate from measured improvement, and does not reuse the MindSpeed headline gain; historical source observations do not establish current-stack compatibility, whole-model correctness or measured benefit"
+ },
+ "recommendations": [
+ "Attribute the state and separate a capacity failure from a throughput bottleneck before choosing an offload surface.",
+ "Verify the consumer of buffer_size before calculating bytes or tuning it, and measure the executing path before claiming a knob ran.",
+ "Compare the whole step's critical path, not host capacity alone.",
+ "Use plan-activation-residency, choose-parallel-training-plan and assess-split-backward-schedule for live-range, parallel-plan and backward-ordering questions."
+ ],
"limitations": [
"Source-derived candidate; no local efficacy validation.",
"Original typed source inventory is not normalized; retain source identities and limits in the procedure/readsets."
@@ -97,10 +119,27 @@
],
"consumption_mode": "packaged"
},
- "applicability": {},
+ "applicability": {
+ "scope": "Choosing training offload and resource-budget changes from pinned DeepSpeed configuration, attributed expert diagnosis and a historical MindSpeed case, balancing accelerator capacity, CPU work, pinned RAM and local storage while keeping host/device transfer exposure distinct from collective network bottlenecks",
+ "prerequisites": "The source boundaries are read; garbage-collection or affinity changes, privileged memlock changes and delayed parameter updates each need separate approval, and the skill authorizes no newest-wheel installation, scheduler-priority change, benchmark or failure injection; historical MindSpeed reuse requires compatible allocator symbols, tracked swappable storage, sufficient host capacity and verified device/NUMA/CPU mapping, not merely the presence of npu-smi and lspci",
+ "input_contract": "State attributed to parameters, optimizer state, gradients or activations with its current memory tier, lifetime and next consumer; whether the failure is allocation capacity or throughput; a per-host budget for input buffers, checkpoint snapshots, offloaded state, OS cache and CPU workers with the actual ranks sharing the host; the host/device link recorded apart from collective traffic; the exact source version and enabled execution mode",
+ "produced_artifact": "A candidate decision record per proposal: the observed limiting interval and owning state or resource, source version and execution mode, capacity released and resources consumed with units and uncertainty, overlap readiness and mutation boundaries with expected contention, and the fixed-workload step and correctness evidence required for later validation",
+ "failure_modes": [
+ "A change that lets the model fit is presented as a speedup, although it can slow each step",
+ "A tensor copied to CPU is taken as an InfiniBand/RoCE bottleneck, or CPU/NVMe offload is counted as network evidence",
+ "Parameter max_in_cpu is read as a byte or total-host-memory cap instead of an element count",
+ "NVMe optimizer placement is assumed to remove CPU optimizer demand",
+ "The parameter buffer-pool count is transplanted to the optimizer, or all optimizers are assumed to hold identical states",
+ "Configuration presence or schema documentation is taken as evidence that a knob or overlap actually ran",
+ "Ordinary overlap is taken as permission for delayed, stale parameter updates"
+ ],
+ "refusal_point": "Claims no hardware validation, measured gain or current NPU compatibility, keeps source support and extracted coverage separate from measured improvement, and does not reuse the MindSpeed headline gain; historical source observations do not establish current-stack compatibility, whole-model correctness or measured benefit"
+ },
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Attribute the state and separate a capacity failure from a throughput bottleneck before choosing an offload surface.",
+ "Verify the consumer of buffer_size before calculating bytes or tuning it, and measure the executing path before claiming a knob ran.",
+ "Compare the whole step's critical path, not host capacity alone.",
+ "Use plan-activation-residency, choose-parallel-training-plan and assess-split-backward-schedule for live-range, parallel-plan and backward-ordering questions."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/video-diffusion-optimization.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/video-diffusion-optimization.json
index 24af5a0d3..6df1833ba 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/video-diffusion-optimization.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/video-diffusion-optimization.json
@@ -26,8 +26,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/methods/video-diffusion-optimization/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/methods/video-diffusion-optimization/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/methods/video-diffusion-optimization/SKILL.md",
"src/inferrsi/data/skill_library/methods/video-diffusion-optimization/assets/campaign.template.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/vllm-metal.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/vllm-metal.json
index c6b525a1b..ecb493542 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/vllm-metal.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/vllm-metal.json
@@ -26,8 +26,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/repositories/vllm-metal/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/repositories/vllm-metal/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/repositories/vllm-metal/LICENSE",
"src/inferrsi/data/skill_library/repositories/vllm-metal/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/weight-loading-optimization.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/weight-loading-optimization.json
index 3e3110f54..d101d12cd 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/weight-loading-optimization.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/weight-loading-optimization.json
@@ -29,8 +29,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/methods/weight-loading-optimization/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/methods/weight-loading-optimization/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/methods/weight-loading-optimization/SKILL.md",
"src/inferrsi/data/skill_library/methods/weight-loading-optimization/metadata.json",
@@ -49,7 +49,7 @@
"checkout_ref": "checkout:weight-loading-optimization",
"source_files": {
"src/inferrsi/data/skill_library/methods/weight-loading-optimization/SKILL.md": "a1bf0e2ec066d5f30225c6cc5fb603f830f48eb5e2aee3fc3d084ed9d8aebe2f",
- "src/inferrsi/data/skill_library/methods/weight-loading-optimization/metadata.json": "6df47ea943e42a11807c54869774957ac2383266120caad730f788207403581c",
+ "src/inferrsi/data/skill_library/methods/weight-loading-optimization/metadata.json": "ebcc5425d7b6409e5c509122e9dc76bed1caab09d73daf013d96086d7faaba2b",
"src/inferrsi/data/skill_library/methods/weight-loading-optimization/references/fastsafetensors-transfer-readset.md": "5a39f14b102b0cee304541e011aa31e44e6b13527d1bee9987c3f3b66c064d3d",
"src/inferrsi/data/skill_library/methods/weight-loading-optimization/references/flare-preparation-boundary.md": "1d7922785a196bdbafdc9791fe79428465a5acc0292ca8f99446f8c764b35454",
"src/inferrsi/data/skill_library/methods/weight-loading-optimization/references/llama-loading-readset.md": "6c640f28ee3f0253a17b81461c865cb9f947367002cb75bf72550c4572d02e97",
@@ -136,6 +136,32 @@
}
],
"empirical_validation": "not_run",
+ "applicability": {
+ "scope": "Diagnosing the serving-level weight load path from storage to device and choosing changes to it when startup or loading latency dominates: locating the cost by source, format, placement and duplication, accounting once-per-process loading and warmup apart from per-request cost, verifying sharded downloads, restricting device-side prefetch to profiled memory-bound hot spots and ordering CPU-tier offload; also covers source-scoped disk-loading modes, pinned staging, synchronous checkpoint save semantics, fastsafetensors broadcast/scatter/coalesced delivery and the separation of model preparation from serving. Request-time host overhead belongs to host-overhead-profiling, DCP asynchronous save performance to dcp-async-save-performance and distributed recovery validity to checkpoint-recovery-consistency",
+ "prerequisites": "Flag names and API entry paths are re-verified against the pinned source revisions, and torch_npu.npu_prefetch details against the target CANN version, before they are relied on; card state is checked before launch and process placement fixed before measuring, through ascend-device-health and os-process-placement; the Diffusers enable_model_cpu_offload path requires accelerate",
+ "input_contract": "The current load path's byte source, file format, host-first or direct device placement and duplicated copies across processes or instances; startup timings for loading and warmup with byte and access counts per segment, and the expected warmup hit criteria; the safetensors index weight_map and the downloaded shard set; a profile showing memory-bound hot spots and the adjacent compute or communication segments; free device memory and whether an OOM occurs during construction or inference; for cross-rank delivery, the file-to-loader-group-rank map, tensor names, shapes, dtypes and call order, the required replicated or sliced output, the model's expected shard layout and the collective backend",
+ "produced_artifact": "The located load-path cost on the four dimensions with its matching lever, a startup ledger that keeps fixed overhead apart from per-request cost, a structural shard-completeness result naming the missing shards, a prefetch proposal behind an enable_prefetch switch defaulting to False or a recorded decision not to prefetch, an offload candidate order, the recorded loading mode, a distributed delivery choice with its proposed verification, and a preparation/serving record of serialized and rebuilt state, live resource owners and ABI/prepared-contract versions; proposed changes and verification targets, not a measured startup or delivery gain",
+ "failure_modes": [
+ "Loading and warmup are amortized into per-request latency, or a host segment without timing points is treated as costing nothing",
+ "Warmup is accepted on total time while mismatched warmup shapes have warmed a different set of cache entries",
+ "In-flight .incomplete shards are re-downloaded as corrupt, or gap-free shard numbering is accepted without matching the index weight_map",
+ "Prefetch is forced without a profiled memory-bound hot spot or a non-memory-bound dependency window",
+ "Resolution or CFG tier-downs are applied to a construction-phase OOM, or offload is adopted when compute already hides the transfer",
+ "A free-memory probe runs on an mmap-backed load, or a run whose first mlock failure latched with only a warning is treated as fully resident",
+ "Disk counters that include cache and pipes are read as physical writes, or a run without a page-cache drop is reported as a cold start",
+ "The ceil-block scatter partition is used without checking it against the model's expected shard layout, or ranks run different orders or plans so their collectives mismatch",
+ "A faster H2D copy, local clone or rank-local filtered read is counted as a cross-rank communication gain"
+ ],
+ "refusal_point": "Withholds a load-path win until the verification checklist holds and keeps the distributed-delivery checks as verification targets rather than results; does not force prefetch without a safe dependency window and disables it when neither a smaller size nor a moved position removes a regression; records unsupported save behaviors as unknown; rejects mismatched execution ABI or prepared-contract versions instead of editing their tags; and implies no Ascend support from the llama.cpp or fastsafetensors mechanisms",
+ "hardware": "NPU for the device-side prefetch API (torch_npu.npu_prefetch, checked against the target CANN version) and the Diffusers offload playbook; the prefetch and fastsafetensors methodology transfers while the fastsafetensors and llama.cpp APIs do not, and cross-rank completion follows each collective backend's own wait contract"
+ },
+ "recommendations": [
+ "Report preparation/startup/readiness, warmups and measured work as separate numbers, with load latency split into init, header, io and get phases, the cold-cache protocol stated and the effective loading mode identical across compared runs.",
+ "Place each prefetch behind an enable_prefetch switch defaulting to False, start max_size at 50% of the theoretical size, verify a single operator first and step up to 0.7 and then 1.0 only after verified wins.",
+ "Try offload candidates in order of change surface: batch size, resolution, CFG/feature flags, CPU or layered offload, tensor-parallel degree, then activation checkpointing; re-estimate the un-overlapped transfer share after adopting each tier.",
+ "Record whole-set atomic publication, asynchronous checkpointing, multi-rank save coordination and crash durability as unknown for the fastsafetensors saver instead of assuming them.",
+ "Before adopting coalesced delivery, compare it disabled and enabled with identical checkpoint, rank mapping, backend, topology, dtype, cache state, chunk/queue budgets and consumer retention, then measure complete startup under the same accounting protocol."
+ ],
"limitations": [
"Source-derived candidate; no local efficacy validation.",
"Normalized provenance covers the existing llama.cpp loading slice at 3e390af18de1131dda27913c00511725b4274c8a and the fastsafetensors distributed-delivery slice at 219e2561a41ede2d5d8eba611c248bb0effbf580; other inherited claims remain unnormalized with original identities and limits retained.",
@@ -150,10 +176,31 @@
],
"consumption_mode": "packaged"
},
- "applicability": {},
+ "applicability": {
+ "scope": "Diagnosing the serving-level weight load path from storage to device and choosing changes to it when startup or loading latency dominates: locating the cost by source, format, placement and duplication, accounting once-per-process loading and warmup apart from per-request cost, verifying sharded downloads, restricting device-side prefetch to profiled memory-bound hot spots and ordering CPU-tier offload; also covers source-scoped disk-loading modes, pinned staging, synchronous checkpoint save semantics, fastsafetensors broadcast/scatter/coalesced delivery and the separation of model preparation from serving. Request-time host overhead belongs to host-overhead-profiling, DCP asynchronous save performance to dcp-async-save-performance and distributed recovery validity to checkpoint-recovery-consistency",
+ "prerequisites": "Flag names and API entry paths are re-verified against the pinned source revisions, and torch_npu.npu_prefetch details against the target CANN version, before they are relied on; card state is checked before launch and process placement fixed before measuring, through ascend-device-health and os-process-placement; the Diffusers enable_model_cpu_offload path requires accelerate",
+ "input_contract": "The current load path's byte source, file format, host-first or direct device placement and duplicated copies across processes or instances; startup timings for loading and warmup with byte and access counts per segment, and the expected warmup hit criteria; the safetensors index weight_map and the downloaded shard set; a profile showing memory-bound hot spots and the adjacent compute or communication segments; free device memory and whether an OOM occurs during construction or inference; for cross-rank delivery, the file-to-loader-group-rank map, tensor names, shapes, dtypes and call order, the required replicated or sliced output, the model's expected shard layout and the collective backend",
+ "produced_artifact": "The located load-path cost on the four dimensions with its matching lever, a startup ledger that keeps fixed overhead apart from per-request cost, a structural shard-completeness result naming the missing shards, a prefetch proposal behind an enable_prefetch switch defaulting to False or a recorded decision not to prefetch, an offload candidate order, the recorded loading mode, a distributed delivery choice with its proposed verification, and a preparation/serving record of serialized and rebuilt state, live resource owners and ABI/prepared-contract versions; proposed changes and verification targets, not a measured startup or delivery gain",
+ "failure_modes": [
+ "Loading and warmup are amortized into per-request latency, or a host segment without timing points is treated as costing nothing",
+ "Warmup is accepted on total time while mismatched warmup shapes have warmed a different set of cache entries",
+ "In-flight .incomplete shards are re-downloaded as corrupt, or gap-free shard numbering is accepted without matching the index weight_map",
+ "Prefetch is forced without a profiled memory-bound hot spot or a non-memory-bound dependency window",
+ "Resolution or CFG tier-downs are applied to a construction-phase OOM, or offload is adopted when compute already hides the transfer",
+ "A free-memory probe runs on an mmap-backed load, or a run whose first mlock failure latched with only a warning is treated as fully resident",
+ "Disk counters that include cache and pipes are read as physical writes, or a run without a page-cache drop is reported as a cold start",
+ "The ceil-block scatter partition is used without checking it against the model's expected shard layout, or ranks run different orders or plans so their collectives mismatch",
+ "A faster H2D copy, local clone or rank-local filtered read is counted as a cross-rank communication gain"
+ ],
+ "refusal_point": "Withholds a load-path win until the verification checklist holds and keeps the distributed-delivery checks as verification targets rather than results; does not force prefetch without a safe dependency window and disables it when neither a smaller size nor a moved position removes a regression; records unsupported save behaviors as unknown; rejects mismatched execution ABI or prepared-contract versions instead of editing their tags; and implies no Ascend support from the llama.cpp or fastsafetensors mechanisms",
+ "hardware": "NPU for the device-side prefetch API (torch_npu.npu_prefetch, checked against the target CANN version) and the Diffusers offload playbook; the prefetch and fastsafetensors methodology transfers while the fastsafetensors and llama.cpp APIs do not, and cross-rank completion follows each collective backend's own wait contract"
+ },
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Report preparation/startup/readiness, warmups and measured work as separate numbers, with load latency split into init, header, io and get phases, the cold-cache protocol stated and the effective loading mode identical across compared runs.",
+ "Place each prefetch behind an enable_prefetch switch defaulting to False, start max_size at 50% of the theoretical size, verify a single operator first and step up to 0.7 and then 1.0 only after verified wins.",
+ "Try offload candidates in order of change surface: batch size, resolution, CFG/feature flags, CPU or layered offload, tensor-parallel degree, then activation checkpointing; re-estimate the un-overlapped transfer share after adopting each tier.",
+ "Record whole-set atomic publication, asynchronous checkpointing, multi-rank save coordination and crash durability as unknown for the fastsafetensors saver instead of assuming them.",
+ "Before adopting coalesced delivery, compare it disabled and enabled with identical checkpoint, rank mapping, backend, topology, dtype, cache state, chunk/queue budgets and consumer retention, then measure complete startup under the same accounting protocol."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-afd-p2p-exchanges.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-afd-p2p-exchanges.json
index b60cefdb7..14c7effb8 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-afd-p2p-exchanges.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-afd-p2p-exchanges.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/SKILL.md",
"src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-flare-session-lifecycle.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-flare-session-lifecycle.json
index 69b4d7bb3..000544f88 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-flare-session-lifecycle.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-flare-session-lifecycle.json
@@ -29,8 +29,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/LICENSE",
"src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/diagnose-attention-bottlenecks.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/diagnose-attention-bottlenecks.json
index f8aca5f03..4d70595fd 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/diagnose-attention-bottlenecks.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/diagnose-attention-bottlenecks.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/hardware/cuda/diagnose-attention-bottlenecks/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/hardware/cuda/diagnose-attention-bottlenecks/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/hardware/cuda/diagnose-attention-bottlenecks/SKILL.md",
"src/inferrsi/data/skill_library/hardware/cuda/diagnose-attention-bottlenecks/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/host-overhead-profiling.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/host-overhead-profiling.json
index cf0b05595..ab9a3acbb 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/host-overhead-profiling.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/host-overhead-profiling.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/hardware/cuda/host-overhead-profiling/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/hardware/cuda/host-overhead-profiling/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/hardware/cuda/host-overhead-profiling/SKILL.md",
"src/inferrsi/data/skill_library/hardware/cuda/host-overhead-profiling/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/reason-hopper-kernels.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/reason-hopper-kernels.json
index 731bc6296..f4108a23c 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/reason-hopper-kernels.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/reason-hopper-kernels.json
@@ -36,8 +36,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/a7f23dfef4770e8542f8390709b029fc934928e4/src/inferrsi/data/skill_library/hardware/cuda/reason-hopper-kernels/SKILL.md",
- "commit": "a7f23dfef4770e8542f8390709b029fc934928e4",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/hardware/cuda/reason-hopper-kernels/SKILL.md",
+ "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
"paths": [
"src/inferrsi/data/skill_library/hardware/cuda/reason-hopper-kernels/SKILL.md",
"src/inferrsi/data/skill_library/hardware/cuda/reason-hopper-kernels/metadata.json",
From a452bf829a3fbe831d54e599e3236fcedd18c5ae Mon Sep 17 00:00:00 2001
From: Dong1017 <55723162+Dong1017@users.noreply.github.com>
Date: Mon, 2 Nov 2026 11:17:04 +0800
Subject: [PATCH 09/11] docs: regenerate Batch 4 discovery projections after
main merge
Bind discovery records to combined source snapshot 78002cb43fdea1956876880ab251f9f1cd9a367f.
Agent-assisted-by: Codex / gpt-6-astra [effort=xhigh] (author)
Signed-off-by: Dong1017 <55723162+Dong1017@users.noreply.github.com>
---
docs/skill-coverage.md | 6 +-
.../repo_packs/inferrsi/knowledge/README.md | 4 +-
.../ascend-aclgraph-capture-replay.json | 4 +-
.../ascend/skills/ascend-device-health.json | 4 +-
.../skills/ascend-distributed-parallel.json | 4 +-
.../ascend/skills/ascend-graph-mode.json | 4 +-
.../skills/ascend-model-adaptation.json | 4 +-
.../skills/ascend-operator-profiling.json | 4 +-
.../ascend/skills/ascend-operator-tuning.json | 4 +-
.../skills/ascend-optimization-casebook.json | 4 +-
.../skills/ascend-performance-analysis.json | 4 +-
.../ascend-video-diffusion-optimization.json | 4 +-
.../ascend/skills/ascendc-tiling-design.json | 4 +-
.../skills/cann-attribution-overlap.json | 4 +-
.../torchnpu-device-memory-lifecycle.json | 4 +-
...torchnpu-pinned-host-memory-lifecycle.json | 4 +-
.../skills/assess-serving-preemption.json | 4 +-
.../assess-split-backward-schedule.json | 4 +-
.../skills/audit-fixed-workload-claim.json | 4 +-
.../audit-h3-metal-fasth3-semantics.json | 4 +-
.../skills/audit-h3-runtime-residency.json | 4 +-
.../skills/audit-llama-runtime-state.json | 4 +-
.../audit-recurrent-kv-finalization.json | 4 +-
.../common/skills/audit-sd-cpp-residency.json | 4 +-
.../audit-sglang-diffusion-lifetimes.json | 4 +-
.../skills/audit-vllm-kv-lifetimes.json | 4 +-
.../bound-windowed-attention-replay.json | 4 +-
.../budget-speculative-verification.json | 4 +-
.../skills/calibrate-serving-simulation.json | 4 +-
.../skills/cgroup-resource-controls.json | 4 +-
.../checkpoint-recovery-consistency.json | 4 +-
.../skills/choose-parallel-training-plan.json | 4 +-
.../skills/cite-check-downstream-pr.json | 4 +-
.../common/skills/claim-contract.json | 4 +-
.../knowledge/common/skills/comfy-ui.json | 4 +-
.../common/skills/competitor-scan.json | 4 +-
.../skills/dcp-async-save-performance.json | 4 +-
.../skills/dense-performance-evidence.json | 4 +-
.../diagnose-vllm-admission-pressure.json | 4 +-
.../skills/diffusers-invocation-state.json | 4 +-
.../common/skills/draft-acceptance-issue.json | 4 +-
.../knowledge/common/skills/dwarfstar.json | 4 +-
.../evaluate-elastic-serving-placement.json | 4 +-
.../skills/generation-cache-lifecycle.json | 4 +-
.../common/skills/host-memory-management.json | 4 +-
.../skills/host-overhead-optimization.json | 4 +-
.../knowledge/common/skills/husky.json | 4 +-
.../skills/inference-benchmark-design.json | 104 +++++++++++++++---
.../inference-optimization-casebook.json | 4 +-
.../common/skills/kernel-writing-guide.json | 4 +-
.../common/skills/kv-tier-management.json | 4 +-
.../lazy-evaluation-stream-lifetimes.json | 4 +-
.../knowledge/common/skills/lily.json | 4 +-
.../knowledge/common/skills/lithos-metal.json | 4 +-
.../knowledge/common/skills/local-dream.json | 4 +-
.../knowledge/common/skills/mistral-rs.json | 4 +-
.../knowledge/common/skills/mlc-llm.json | 4 +-
.../knowledge/common/skills/mlx-swift-lm.json | 4 +-
.../common/skills/npu-skill-factory.json | 4 +-
.../skills/optimization-search-policy.json | 4 +-
.../common/skills/os-process-placement.json | 4 +-
.../skills/performance-ceiling-analysis.json | 33 ++++--
.../skills/plan-activation-residency.json | 4 +-
.../knowledge/common/skills/project-maya.json | 4 +-
.../skills/repository-evolution-analysis.json | 4 +-
.../knowledge/common/skills/rsi-evidence.json | 4 +-
.../skills/skill-evolution-playbook.json | 4 +-
.../knowledge/common/skills/splash.json | 4 +-
.../common/skills/storage-io-measurement.json | 4 +-
.../skills/training-input-pipeline.json | 4 +-
.../skills/training-resource-budget.json | 4 +-
.../skills/video-diffusion-optimization.json | 4 +-
.../knowledge/common/skills/vllm-metal.json | 4 +-
.../skills/weight-loading-optimization.json | 4 +-
.../cuda/skills/audit-afd-p2p-exchanges.json | 4 +-
.../skills/audit-flare-session-lifecycle.json | 4 +-
.../diagnose-attention-bottlenecks.json | 4 +-
.../cuda/skills/host-overhead-profiling.json | 4 +-
.../cuda/skills/reason-hopper-kernels.json | 4 +-
79 files changed, 267 insertions(+), 180 deletions(-)
diff --git a/docs/skill-coverage.md b/docs/skill-coverage.md
index 9f8223d37..853eb321c 100644
--- a/docs/skill-coverage.md
+++ b/docs/skill-coverage.md
@@ -4,7 +4,7 @@ Generated from the registry, the skill sidecars and the two maintained
matrices. **Do not edit by hand**; regenerate after any registry, skill,
matrix or frozen-plan change.
-- Input digest: `sha256:37e7219c0b159ecbdf593d1bf903ce83a7de56d618dd22d0feecd83841ea4ecd`
+- Input digest: `sha256:946ec9585c7e530125131a12ef50666c490fb436c4a2c60c0a638ea7c88bcdf5`
- Regenerate: `python tools/build_skill_coverage.py`
- Verify without writing: `python tools/build_skill_coverage.py --check`
@@ -126,11 +126,11 @@ optimization is fully covered or that any procedure has measured benefit.
| [choose-parallel-training-plan](../src/inferrsi/data/skill_library/methods/choose-parallel-training-plan/SKILL.md) | common | Compare stage, device-mesh and operator-sharding candidates for static distributed training while accounting for ... | Comparing stage, device-mesh and operator-sharding candidates for static distributed ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator · Diagnose / Configure / Optimize / Verify
4 Network · Diagnose / Configure / Optimize / Verify | 1 · Systems sources (Alpa — [2201.12023v3](https://arxiv.org/abs/2201.12023v3)) | candidate · not_run · 1 readset(s) | no-eval |
| [dense-performance-evidence](../src/inferrsi/data/skill_library/methods/dense-performance-evidence/SKILL.md) | common | Taxonomy of performance evidence an optimization agent needs, ordered by collection cost and diagnostic power. Maps ... | Deciding which performance evidence to collect next and which claim each evidence level ... | failure_modes, input_contract, produced_artifact, refusal_point
4 advice | 1 Accelerator · Observe / Profile / Verify
2 RAM/Host · Verify
4 Network · Verify | 1 · Systems sources (NVIDIA/TensorRT-LLM @ f33e0e5fd)
2 · Methodology and corpus references (JiusiServe/vllm-evolve @ 43cc82c)
2 · Methodology and corpus references (microsoft/SkillOpt @ 343db22) | candidate · not_run · 0 readset(s) | — |
| [evaluate-elastic-serving-placement](../src/inferrsi/data/skill_library/methods/evaluate-elastic-serving-placement/SKILL.md) | common | Compare migrating an LLM request between model replicas with changing its sequence-parallel execution group, including ... | Comparing migration of a running LLM request between compatible model replicas with ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator · Diagnose / Configure / Optimize / Verify
2 RAM/Host · Diagnose / Configure / Optimize / Verify
4 Network · Diagnose / Configure / Optimize / Verify | 1 · Systems sources (Llumnix — [2406.03243v1](https://arxiv.org/abs/2406.03243v1)) | candidate · not_run · 1 readset(s) | no-eval |
-| [inference-benchmark-design](../src/inferrsi/data/skill_library/methods/inference-benchmark-design/SKILL.md) | common | Select task quality gates, reference parity checks and performance workloads for text, agentic coding, work agents ... | Selecting task quality gates, reference parity checks and performance workloads for text ... | failure_modes, input_contract, produced_artifact, refusal_point
4 advice | — | 2 · Methodology and corpus references (JiusiServe/vllm-evolve @ 43cc82c)
Task-specific benchmark selection (issue #60) ([Issue #60 report](https://github.com/InferMatrixLab/InferRSI/issues/6 ...) | candidate · not_run · 2 readset(s) | no-SearchSpace-layer |
+| [inference-benchmark-design](../src/inferrsi/data/skill_library/methods/inference-benchmark-design/SKILL.md) | common | Select task quality gates, reference parity checks and performance workloads for text, agentic coding, work agents ... | Selecting task quality gates, reference parity checks and performance workloads for text ... | failure_modes, input_contract, produced_artifact, refusal_point
4 advice | — | 2 · Methodology and corpus references (JiusiServe/vllm-evolve @ 43cc82c)
Task-specific benchmark selection (issue #60) ([Issue #60 report](https://github.com/InferMatrixLab/InferRSI/issues/6 ...)
Task-specific benchmark selection (issue #60) ([NVIDIA/NeMo-Speech.cpp @ 20999cc9](https://github.com/NVIDIA/NeMo-Spe ...) | candidate · not_run · 5 readset(s) | no-SearchSpace-layer |
| [inference-optimization-casebook](../src/inferrsi/data/skill_library/methods/inference-optimization-casebook/SKILL.md) | common | How to structure and use a casebook of past optimization decisions as a decision-support tool. Extracts the casebook ... | Structuring and querying a casebook of past optimization decisions as decision support ... | failure_modes, input_contract, produced_artifact, refusal_point
4 advice | 1 Accelerator · Optimize | 1 · Systems sources (NVIDIA/TensorRT-LLM @ f33e0e5fd) | candidate · not_run · 0 readset(s) | no-eval |
| [kernel-writing-guide](../src/inferrsi/data/skill_library/methods/kernel-writing-guide/SKILL.md) | common | Decision guide for writing custom inference kernels: when to write one, which framework to choose (Triton, CuTe DSL ... | Deciding whether to write a custom inference kernel, routing a kernel pattern to Triton ... | failure_modes, input_contract, produced_artifact, refusal_point
6 advice | 1 Accelerator · Optimize / Diagnose / Configure / Verify | 1 · Systems sources (Dao-AILab/causal-conv1d @ `cd81f0413cad2fc1e6f17e785ac39f59aae690cd` ...)
1 · Systems sources (NVIDIA/TensorRT-LLM @ f33e0e5fd)
1 · Systems sources (huggingface/transformers @ `0dad7b822255a0ae261ec45ae937371e859ffd1a` ...)
1 · Systems sources (mindspore-ai/akg @ `5aa15f3fe4d856eb60cd3b27447e04f5f1fb4dff` ...)
1 · Systems sources (sgl-project/sglang @ `16176278d0180f1a1a578e53e6eef432577876c2` ...)
1 · Systems sources (vllm-project/vllm @ `537a79f0fc5ba564d2bef43d54915b2f80f9f77d` (root ...) | candidate · not_run · 7 readset(s) | no-eval |
| [optimization-search-policy](../src/inferrsi/data/skill_library/methods/optimization-search-policy/SKILL.md) | common | Policy for how an optimization agent should select, order, and stop optimization experiments. Covers candidate ... | Selecting, ordering and stopping optimization experiments: candidate generation ... | failure_modes, input_contract, produced_artifact, refusal_point
4 advice | 1 Accelerator · Optimize | 2 · Methodology and corpus references (JiusiServe/vllm-evolve @ 43cc82c)
2 · Methodology and corpus references (RRSI — arXiv 2609.24972v3)
2 · Methodology and corpus references (arXiv 2608.01867 (CRISP))
2 · Methodology and corpus references (microsoft/SkillOpt @ 343db22) | candidate · not_run · 9 readset(s) | no-eval |
-| [performance-ceiling-analysis](../src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/SKILL.md) | common | Estimate optimization headroom before modifying an inference system. Produces a three-level ceiling model (analytical ... | inference serving systems (LLM, diffusion, audio) | — | 1 Accelerator · Diagnose / Verify
5 OS · Configure | 1 · Systems sources (Mooncake — FAST '25 proceedings, pp. 155-170)
1 · Systems sources (NVIDIA/TensorRT-LLM @ f33e0e5fd)
1 · Systems sources (ebugger/Empirical-Roofline-Toolkit @ 0771fa9)
1 · Systems sources (hahnyuan/LLM-Viewer @ 1893e4b) | candidate · not_run · 17 readset(s) | no-eval |
+| [performance-ceiling-analysis](../src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/SKILL.md) | common | Estimate optimization headroom before modifying an inference system. Produces a three-level ceiling model (analytical ... | inference serving systems (LLM, diffusion, audio) | — | 1 Accelerator · Diagnose / Verify
5 OS · Configure | 1 · Systems sources (FlexGen — [ICML 2023 / PMLR 202:31094–31116](https://proceedings.mlr.p ...)
1 · Systems sources (Mooncake — FAST '25 proceedings, pp. 155-170)
1 · Systems sources (NVIDIA/TensorRT-LLM @ f33e0e5fd)
1 · Systems sources (ebugger/Empirical-Roofline-Toolkit @ 0771fa9)
1 · Systems sources (hahnyuan/LLM-Viewer @ 1893e4b) | candidate · not_run · 18 readset(s) | no-eval |
| [plan-activation-residency](../src/inferrsi/data/skill_library/methods/plan-activation-residency/SKILL.md) | common | Propose training activation and gradient memory reductions by distinguishing lifetime reordering, address packing and ... | Proposing training activation and gradient memory reductions when pressure appears to ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator · Diagnose / Optimize / Verify | 1 · Systems sources (Checkmate — [1910.02653v3](https://arxiv.org/abs/1910.02653v3)) | candidate · not_run · 1 readset(s) | no-eval |
| [repository-evolution-analysis](../src/inferrsi/data/skill_library/methods/repository-evolution-analysis/SKILL.md) | common | Analyze a repository's commit history to build a capability-evolution model: what the system can do, how it got there ... | any software repository with a recoverable git history | — | — | 2 · Methodology and corpus references (LearningToCommit/LearningToCommit @ 7e5a1d7)
2 · Methodology and corpus references (yetone/magpie @ 77800032) | candidate · not_run · 0 readset(s) | no-SearchSpace-layer, no-eval |
| [skill-evolution-playbook](../src/inferrsi/data/skill_library/methods/skill-evolution-playbook/SKILL.md) | common | Practical design choices for evolving agent skills from execution traces: when one trajectory should update a skill ... | Design choices for evolving agent skills from execution traces, drawn from Trace2Skill ... | failure_modes, input_contract, produced_artifact, refusal_point
4 advice | — | 2 · Methodology and corpus references (Qwen-Applications/Trace2Skill @ 3d0b52a)
2 · Methodology and corpus references (microsoft/SkillOpt @ 343db22) | candidate · not_run · 0 readset(s) | no-SearchSpace-layer, no-eval |
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/README.md b/src/inferrsi/data/repo_packs/inferrsi/knowledge/README.md
index 0645548d5..daaa46804 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/README.md
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/README.md
@@ -15,7 +15,7 @@ These are knowledge abstracts (`--kind knowledge`), with namespaced package
IDs distinct from explicitly selected `checkout_ref` skills. Ascend/CUDA
procedures use their corresponding backend; shared methods use `common`.
-Source snapshot: [daa33a4ba588298ff76e3e077d6c894d7750d9bc](https://github.com/InferMatrixLab/InferRSI/tree/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library).
+Source snapshot: [78002cb43fdea1956876880ab251f9f1cd9a367f](https://github.com/InferMatrixLab/InferRSI/tree/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library).
```bash
inferrsi knowledge search 'profiling' --repo-type inferrsi --limit 50
@@ -29,7 +29,7 @@ contain the current input paths and bytes (Git checkout filters are respected).
```bash
python tools/build_skill_index.py --source-revision
-python tools/build_skill_index.py --source-revision daa33a4ba588298ff76e3e077d6c894d7750d9bc --check
+python tools/build_skill_index.py --source-revision 78002cb43fdea1956876880ab251f9f1cd9a367f --check
```
`--check` detects missing, stale and unexpected generated records without writing.
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-aclgraph-capture-replay.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-aclgraph-capture-replay.json
index ca56dfe40..391ded6cd 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-aclgraph-capture-replay.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-aclgraph-capture-replay.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/hardware/ascend/ascend-aclgraph-capture-replay/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/hardware/ascend/ascend-aclgraph-capture-replay/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-aclgraph-capture-replay/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-aclgraph-capture-replay/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-device-health.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-device-health.json
index 7395d5531..eef063cf4 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-device-health.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-device-health.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/hardware/ascend/ascend-device-health/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/hardware/ascend/ascend-device-health/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-device-health/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-device-health/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-distributed-parallel.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-distributed-parallel.json
index cb6c8c4fd..527153054 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-distributed-parallel.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-distributed-parallel.json
@@ -27,8 +27,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/hardware/ascend/ascend-distributed-parallel/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/hardware/ascend/ascend-distributed-parallel/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-distributed-parallel/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-distributed-parallel/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-graph-mode.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-graph-mode.json
index 836b58e05..41d95e6f2 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-graph-mode.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-graph-mode.json
@@ -26,8 +26,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/hardware/ascend/ascend-graph-mode/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/hardware/ascend/ascend-graph-mode/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-graph-mode/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-graph-mode/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-model-adaptation.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-model-adaptation.json
index e55deb1d0..8414cfc82 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-model-adaptation.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-model-adaptation.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/hardware/ascend/ascend-model-adaptation/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/hardware/ascend/ascend-model-adaptation/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-model-adaptation/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-model-adaptation/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-profiling.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-profiling.json
index 936e75bb5..603c07ca8 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-profiling.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-profiling.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-profiling/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-profiling/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-profiling/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-profiling/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-tuning.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-tuning.json
index 9b9ac9da6..35921208e 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-tuning.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-tuning.json
@@ -24,8 +24,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-tuning/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-tuning/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-tuning/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-tuning/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-optimization-casebook.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-optimization-casebook.json
index 436cc7e64..654424c77 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-optimization-casebook.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-optimization-casebook.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/hardware/ascend/ascend-optimization-casebook/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/hardware/ascend/ascend-optimization-casebook/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-optimization-casebook/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-optimization-casebook/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-performance-analysis.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-performance-analysis.json
index 532c02bae..fc8237e50 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-performance-analysis.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-performance-analysis.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/hardware/ascend/ascend-performance-analysis/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/hardware/ascend/ascend-performance-analysis/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-performance-analysis/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-performance-analysis/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-video-diffusion-optimization.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-video-diffusion-optimization.json
index d18b828b2..572a93c5a 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-video-diffusion-optimization.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-video-diffusion-optimization.json
@@ -26,8 +26,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/hardware/ascend/ascend-video-diffusion-optimization/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/hardware/ascend/ascend-video-diffusion-optimization/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-video-diffusion-optimization/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-video-diffusion-optimization/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascendc-tiling-design.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascendc-tiling-design.json
index 549d7438f..bbc49aec6 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascendc-tiling-design.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascendc-tiling-design.json
@@ -24,8 +24,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/hardware/ascend/ascendc-tiling-design/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/hardware/ascend/ascendc-tiling-design/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascendc-tiling-design/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascendc-tiling-design/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/cann-attribution-overlap.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/cann-attribution-overlap.json
index 56e78c7fa..939f791d3 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/cann-attribution-overlap.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/cann-attribution-overlap.json
@@ -24,8 +24,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/hardware/ascend/cann-attribution-overlap/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/hardware/ascend/cann-attribution-overlap/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/cann-attribution-overlap/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/cann-attribution-overlap/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-device-memory-lifecycle.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-device-memory-lifecycle.json
index 4946c2645..2c44b362d 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-device-memory-lifecycle.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-device-memory-lifecycle.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/hardware/ascend/torchnpu-device-memory-lifecycle/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/hardware/ascend/torchnpu-device-memory-lifecycle/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/torchnpu-device-memory-lifecycle/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/torchnpu-device-memory-lifecycle/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-pinned-host-memory-lifecycle.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-pinned-host-memory-lifecycle.json
index bb0c2b4cc..f574ae8ab 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-pinned-host-memory-lifecycle.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-pinned-host-memory-lifecycle.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/hardware/ascend/torchnpu-pinned-host-memory-lifecycle/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/hardware/ascend/torchnpu-pinned-host-memory-lifecycle/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/torchnpu-pinned-host-memory-lifecycle/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/torchnpu-pinned-host-memory-lifecycle/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-serving-preemption.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-serving-preemption.json
index 7776378c8..91d99b710 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-serving-preemption.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-serving-preemption.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/methods/assess-serving-preemption/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/methods/assess-serving-preemption/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/methods/assess-serving-preemption/SKILL.md",
"src/inferrsi/data/skill_library/methods/assess-serving-preemption/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-split-backward-schedule.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-split-backward-schedule.json
index b0aa98710..3f809f979 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-split-backward-schedule.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-split-backward-schedule.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/methods/assess-split-backward-schedule/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/methods/assess-split-backward-schedule/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/methods/assess-split-backward-schedule/SKILL.md",
"src/inferrsi/data/skill_library/methods/assess-split-backward-schedule/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-fixed-workload-claim.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-fixed-workload-claim.json
index fbcb6c5f5..46bd61d02 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-fixed-workload-claim.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-fixed-workload-claim.json
@@ -20,8 +20,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/methods/audit-fixed-workload-claim/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/methods/audit-fixed-workload-claim/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/methods/audit-fixed-workload-claim/SKILL.md",
"src/inferrsi/data/skill_library/methods/audit-fixed-workload-claim/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-metal-fasth3-semantics.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-metal-fasth3-semantics.json
index 5499c5247..1941bc103 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-metal-fasth3-semantics.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-metal-fasth3-semantics.json
@@ -29,8 +29,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/LICENSE",
"src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-runtime-residency.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-runtime-residency.json
index d47eaa5a7..3cab95bbb 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-runtime-residency.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-runtime-residency.json
@@ -30,8 +30,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/LICENSE",
"src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-llama-runtime-state.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-llama-runtime-state.json
index 88e489b35..3f3701e61 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-llama-runtime-state.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-llama-runtime-state.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/SKILL.md",
"src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-recurrent-kv-finalization.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-recurrent-kv-finalization.json
index e2e13984b..1c2fa168b 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-recurrent-kv-finalization.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-recurrent-kv-finalization.json
@@ -26,8 +26,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/LICENSE",
"src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sd-cpp-residency.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sd-cpp-residency.json
index 99ae7ca7f..4e73951f0 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sd-cpp-residency.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sd-cpp-residency.json
@@ -25,8 +25,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/LICENSE",
"src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sglang-diffusion-lifetimes.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sglang-diffusion-lifetimes.json
index 4f0e4108d..ba11f6422 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sglang-diffusion-lifetimes.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sglang-diffusion-lifetimes.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/LICENSE",
"src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-vllm-kv-lifetimes.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-vllm-kv-lifetimes.json
index f7d867115..af929ae7e 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-vllm-kv-lifetimes.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-vllm-kv-lifetimes.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/SKILL.md",
"src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/bound-windowed-attention-replay.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/bound-windowed-attention-replay.json
index 828a9fefb..439cf4f58 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/bound-windowed-attention-replay.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/bound-windowed-attention-replay.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/SKILL.md",
"src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/budget-speculative-verification.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/budget-speculative-verification.json
index 61a5d12ac..06a71583a 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/budget-speculative-verification.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/budget-speculative-verification.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/SKILL.md",
"src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/calibrate-serving-simulation.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/calibrate-serving-simulation.json
index f17baeb43..cb09587b8 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/calibrate-serving-simulation.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/calibrate-serving-simulation.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/methods/calibrate-serving-simulation/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/methods/calibrate-serving-simulation/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/methods/calibrate-serving-simulation/SKILL.md",
"src/inferrsi/data/skill_library/methods/calibrate-serving-simulation/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cgroup-resource-controls.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cgroup-resource-controls.json
index 32c888f6b..b26d1e4b5 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cgroup-resource-controls.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cgroup-resource-controls.json
@@ -25,8 +25,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/hardware/host/cgroup-resource-controls/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/hardware/host/cgroup-resource-controls/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/hardware/host/cgroup-resource-controls/SKILL.md",
"src/inferrsi/data/skill_library/hardware/host/cgroup-resource-controls/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/checkpoint-recovery-consistency.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/checkpoint-recovery-consistency.json
index ec86f4f46..dd85bc623 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/checkpoint-recovery-consistency.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/checkpoint-recovery-consistency.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/methods/checkpoint-recovery-consistency/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/methods/checkpoint-recovery-consistency/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/methods/checkpoint-recovery-consistency/SKILL.md",
"src/inferrsi/data/skill_library/methods/checkpoint-recovery-consistency/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/choose-parallel-training-plan.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/choose-parallel-training-plan.json
index 4e33872bd..7c2f13e5d 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/choose-parallel-training-plan.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/choose-parallel-training-plan.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/methods/choose-parallel-training-plan/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/methods/choose-parallel-training-plan/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/methods/choose-parallel-training-plan/SKILL.md",
"src/inferrsi/data/skill_library/methods/choose-parallel-training-plan/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cite-check-downstream-pr.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cite-check-downstream-pr.json
index 0578cef02..98628f65a 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cite-check-downstream-pr.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cite-check-downstream-pr.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/workflows/cite-check-downstream-pr/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/workflows/cite-check-downstream-pr/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/skill-index.json",
"src/inferrsi/data/skill_library/workflows/cite-check-downstream-pr/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/claim-contract.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/claim-contract.json
index e31b42196..96e9330f7 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/claim-contract.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/claim-contract.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/workflows/claim-contract/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/workflows/claim-contract/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/skill-index.json",
"src/inferrsi/data/skill_library/workflows/claim-contract/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/comfy-ui.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/comfy-ui.json
index 619da8879..d40ae76ab 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/comfy-ui.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/comfy-ui.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/repositories/comfy-ui/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/repositories/comfy-ui/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/repositories/comfy-ui/LICENSE",
"src/inferrsi/data/skill_library/repositories/comfy-ui/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/competitor-scan.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/competitor-scan.json
index 9b6470def..42280bcf2 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/competitor-scan.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/competitor-scan.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/workflows/competitor-scan/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/workflows/competitor-scan/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/skill-index.json",
"src/inferrsi/data/skill_library/workflows/competitor-scan/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dcp-async-save-performance.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dcp-async-save-performance.json
index 36fa0ec7b..f7cc2da85 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dcp-async-save-performance.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dcp-async-save-performance.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/SKILL.md",
"src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dense-performance-evidence.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dense-performance-evidence.json
index 841dfaa07..e6edc9c66 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dense-performance-evidence.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dense-performance-evidence.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/methods/dense-performance-evidence/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/methods/dense-performance-evidence/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/methods/dense-performance-evidence/SKILL.md",
"src/inferrsi/data/skill_library/methods/dense-performance-evidence/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diagnose-vllm-admission-pressure.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diagnose-vllm-admission-pressure.json
index 822ff37f7..b7bac3770 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diagnose-vllm-admission-pressure.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diagnose-vllm-admission-pressure.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/SKILL.md",
"src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diffusers-invocation-state.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diffusers-invocation-state.json
index a1ca76b99..b88df15c5 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diffusers-invocation-state.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diffusers-invocation-state.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/SKILL.md",
"src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/draft-acceptance-issue.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/draft-acceptance-issue.json
index 4d293a971..eca015bea 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/draft-acceptance-issue.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/draft-acceptance-issue.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/workflows/draft-acceptance-issue/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/workflows/draft-acceptance-issue/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/skill-index.json",
"src/inferrsi/data/skill_library/workflows/draft-acceptance-issue/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dwarfstar.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dwarfstar.json
index 417292ed9..7495c14b1 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dwarfstar.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dwarfstar.json
@@ -25,8 +25,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/repositories/dwarfstar/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/repositories/dwarfstar/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/repositories/dwarfstar/LICENSE",
"src/inferrsi/data/skill_library/repositories/dwarfstar/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/evaluate-elastic-serving-placement.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/evaluate-elastic-serving-placement.json
index 5362c7516..2b69cb795 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/evaluate-elastic-serving-placement.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/evaluate-elastic-serving-placement.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/methods/evaluate-elastic-serving-placement/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/methods/evaluate-elastic-serving-placement/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/methods/evaluate-elastic-serving-placement/SKILL.md",
"src/inferrsi/data/skill_library/methods/evaluate-elastic-serving-placement/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/generation-cache-lifecycle.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/generation-cache-lifecycle.json
index 9133cbcae..3f8d133e7 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/generation-cache-lifecycle.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/generation-cache-lifecycle.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/SKILL.md",
"src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-memory-management.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-memory-management.json
index 8c1ad56be..48329956a 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-memory-management.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-memory-management.json
@@ -30,8 +30,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/hardware/host/host-memory-management/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/hardware/host/host-memory-management/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/hardware/host/host-memory-management/SKILL.md",
"src/inferrsi/data/skill_library/hardware/host/host-memory-management/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-overhead-optimization.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-overhead-optimization.json
index 854e75629..137cf6178 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-overhead-optimization.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-overhead-optimization.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/hardware/host/host-overhead-optimization/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/hardware/host/host-overhead-optimization/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/hardware/host/host-overhead-optimization/SKILL.md",
"src/inferrsi/data/skill_library/hardware/host/host-overhead-optimization/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/husky.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/husky.json
index 1d70d2187..fb9e22206 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/husky.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/husky.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/repositories/husky/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/repositories/husky/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/repositories/husky/SKILL.md",
"src/inferrsi/data/skill_library/repositories/husky/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-benchmark-design.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-benchmark-design.json
index 3f7cf357f..00314c24f 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-benchmark-design.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-benchmark-design.json
@@ -8,7 +8,7 @@
"component": "skills",
"scenario": "skills.discovery",
"status": "imported",
- "summary": "Select task quality gates, reference parity checks and performance workloads for text, agentic coding, work agents, TTS, full-duplex audio/video, image and video inference. Covers AIPerf AgentX/Exgentic replay, vLLM/SGLang benchmark boundaries, SLO-aware metrics and InferRSI integration.",
+ "summary": "Select task quality gates, reference parity checks and performance workloads for text, agentic coding, work agents, cache-aware ASR, TTS, full-duplex audio/video, image and video inference. Covers AIPerf AgentX/Exgentic replay, vLLM/SGLang benchmark boundaries, SLO-aware metrics and InferRSI integration.",
"tags": [
"benchmark",
"evaluation",
@@ -22,7 +22,9 @@
"tts",
"full-duplex",
"image-generation",
- "video-generation"
+ "video-generation",
+ "asr",
+ "stream-boundaries"
],
"origin": [
"external"
@@ -30,16 +32,19 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/methods/inference-benchmark-design/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/methods/inference-benchmark-design/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/methods/inference-benchmark-design/SKILL.md",
"src/inferrsi/data/skill_library/methods/inference-benchmark-design/metadata.json",
"src/inferrsi/data/skill_library/methods/inference-benchmark-design/references/benchmark-commands.md",
+ "src/inferrsi/data/skill_library/methods/inference-benchmark-design/references/cache-aware-asr-stream-boundaries.md",
+ "src/inferrsi/data/skill_library/methods/inference-benchmark-design/references/justgoal-evaluation-contracts.md",
+ "src/inferrsi/data/skill_library/methods/inference-benchmark-design/references/justgoal-source.json",
"src/inferrsi/data/skill_library/methods/inference-benchmark-design/references/task-benchmark-selection.md",
"src/inferrsi/data/skill_library/skill-index.json"
],
- "content_sha256": "59b5656e7d989b3cee8d6f95cd43be03053b1fb11c3a82bb49ab3d402f7aacee"
+ "content_sha256": "a977e81fb83e3e476f66dc97133f1a60a35983bdfaec538ad5063defda41d4b7"
}
],
"index_type": "packaged-skill",
@@ -48,15 +53,18 @@
},
"checkout_ref": "checkout:inference-benchmark-design",
"source_files": {
- "src/inferrsi/data/skill_library/methods/inference-benchmark-design/SKILL.md": "59b5656e7d989b3cee8d6f95cd43be03053b1fb11c3a82bb49ab3d402f7aacee",
- "src/inferrsi/data/skill_library/methods/inference-benchmark-design/metadata.json": "d51b5b3600d13b38c2adbf6c6e149c7375c9f1db12bc078679cab67a32a3e9a7",
+ "src/inferrsi/data/skill_library/methods/inference-benchmark-design/SKILL.md": "a977e81fb83e3e476f66dc97133f1a60a35983bdfaec538ad5063defda41d4b7",
+ "src/inferrsi/data/skill_library/methods/inference-benchmark-design/metadata.json": "3570ee36df8037c97c4692afe39316b62c0c91c28b2ea29ab9b72c5ae6740bb6",
"src/inferrsi/data/skill_library/methods/inference-benchmark-design/references/benchmark-commands.md": "0898424e0bff3a6246e1eeeda6abd6c5bd974c047fe32f3bbd5ab9d31ce17cd3",
+ "src/inferrsi/data/skill_library/methods/inference-benchmark-design/references/cache-aware-asr-stream-boundaries.md": "4e0ea9c1bf6b096a49a91ad5266e90b8a456f4f5aca15eba8ac270a53c2309b5",
+ "src/inferrsi/data/skill_library/methods/inference-benchmark-design/references/justgoal-evaluation-contracts.md": "d25df795fd301e8a1bb436e5f8122d0bac2441c43104e59a3b310fd5a64f2f3e",
+ "src/inferrsi/data/skill_library/methods/inference-benchmark-design/references/justgoal-source.json": "60f2784384ea1a3b08a2ecafb60207d8e3707247dc7bc177b95b463d81f1cc16",
"src/inferrsi/data/skill_library/methods/inference-benchmark-design/references/task-benchmark-selection.md": "0d4bb080a754e26846e81680023db0dff5e6a4c96be4327c411c3b72ac4ef8e2",
"src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
},
"skill_metadata": {
"name": "inference-benchmark-design",
- "description": "Select task quality gates, reference parity checks and performance workloads for text, agentic coding, work agents, TTS, full-duplex audio/video, image and video inference. Covers AIPerf AgentX/Exgentic replay, vLLM/SGLang benchmark boundaries, SLO-aware metrics and InferRSI integration.",
+ "description": "Select task quality gates, reference parity checks and performance workloads for text, agentic coding, work agents, cache-aware ASR, TTS, full-duplex audio/video, image and video inference. Covers AIPerf AgentX/Exgentic replay, vLLM/SGLang benchmark boundaries, SLO-aware metrics and InferRSI integration.",
"tags": [
"benchmark",
"evaluation",
@@ -70,7 +78,9 @@
"tts",
"full-duplex",
"image-generation",
- "video-generation"
+ "video-generation",
+ "asr",
+ "stream-boundaries"
],
"status": "candidate",
"profile": "local-skill-v1",
@@ -197,12 +207,61 @@
"src/inferrsi/model/protocols/metrics.py"
],
"readset": "references/task-benchmark-selection.md"
+ },
+ {
+ "type": "git",
+ "url": "https://github.com/NVIDIA/NeMo-Speech.cpp",
+ "commit": "20999cc921d2b14aa58708f61756a8b2e787ba20",
+ "paths": [
+ "BENCHMARK.md",
+ "LICENSE",
+ "NOTICE",
+ "THIRD_PARTY_NOTICES.md",
+ "src/asr/encoder/cache_aware_encoder.cpp",
+ "src/asr/encoder/cache_aware_encoder.h",
+ "src/asr/encoder/fastconformer.h",
+ "src/asr/recognizer.cpp",
+ "src/asr/runner.cpp",
+ "src/asr/runner.h",
+ "src/asr/vad/vad_endpointer.cpp",
+ "src/asr/vad/vad_endpointer.h",
+ "tests/cpp/asr/CMakeLists.txt",
+ "tests/cpp/asr/test_cache_stream_geometry.cpp"
+ ],
+ "readset": "references/cache-aware-asr-stream-boundaries.md"
+ },
+ {
+ "type": "git",
+ "url": "https://github.com/OpenBMB/JustGoal",
+ "commit": "f052fc0b28fc7f10b2ef1ae7aa4682aabbbe868d",
+ "paths": [
+ "LICENSE",
+ "NOTICE",
+ "README.md",
+ "goal_full_shape/fable-5.1/dsa_decode_fwd/dsa_sparse_attention_h16_ckv512_kpe64_topk2048_ps64/dsa_decode.cu",
+ "goal_full_shape/fable-5.1/dsa_decode_fwd/dsa_sparse_attention_h16_ckv512_kpe64_topk2048_ps64/main.py",
+ "goal_single_shape/fable-5.1/dsa_decode_fwd/t8_topk2048/dsa_kernel.cu",
+ "goal_single_shape/fable-5.1/dsa_decode_fwd/t8_topk2048/dsa_kernel.py",
+ "kda05_reproduction_full_shape/opus-4.8/fa4_bwd/attention_bwd_causal_h16_d128/cute_fabwd_bshd_v5c_fusedzero/helpers/NOT_INCLUDED.md"
+ ],
+ "readset": "references/justgoal-source.json",
+ "license": "Apache-2.0 repository default; NOTICE exceptions and excluded proprietary helper retained",
+ "scope": "Selected kernel entry/workspace/stream contracts and distribution boundary; source-only, no execution"
+ },
+ {
+ "type": "document",
+ "url": "https://github.com/OpenBMB/JustGoal/blob/f052fc0b28fc7f10b2ef1ae7aa4682aabbbe868d/paper.pdf",
+ "version": "Repository-paired PDF at f052fc0b28fc7f10b2ef1ae7aa4682aabbbe868d",
+ "sha256": "6bfa1a2728546def4c48e15cf55f31d92a543264787318f1522236c80a0ae8c6",
+ "readset": "references/justgoal-source.json",
+ "license": "UNKNOWN",
+ "scope": "Complete 21-page paper and appendices: controlled comparison, independent correctness, repair/coverage and timing evidence; author-reported, no PDF/figure redistribution"
}
],
"empirical_validation": "not_run",
"applicability": {
- "scope": "Selecting task quality gates, reference parity checks and performance workloads for text, agentic coding, work agents, TTS, full-duplex audio/video, image and video inference, and designing goodput-based multi-regime comparisons that fit InferRSI's evidence trail",
- "prerequisites": "The primary objective, timing boundary, quality gates, sample/failure accounting, required regimes, repetitions, seeds and decision rule are frozen before optimization; baseline and candidate share hardware, software stack, input distribution, benchmark configuration, warmup and measurement protocol, and on a shared machine also one time window and dedicated devices",
+ "scope": "Selecting task quality gates, reference parity checks and performance workloads for text, agentic coding, work agents, cache-aware ASR, TTS, full-duplex audio/video, image and video inference, and designing goodput-based multi-regime comparisons that fit InferRSI's evidence trail",
+ "prerequisites": "The primary objective, timing boundary, quality gates, sample/failure accounting, required regimes, repetitions, seeds and decision rule are frozen before optimization; baseline and candidate share hardware, software stack, input distribution, benchmark configuration, warmup and measurement protocol, and on a shared machine also one time window and dedicated devices; cache-aware ASR also requires the selected model/config and first/steady geometry, decoded clock and reset/EOU/EOF boundaries to be established",
"input_contract": "The task family and deployment use case; candidate quality evaluators and reference parity; the declared SLO and the caller's policy on which regimes must improve or must not regress; for an audit, a supplied fixed-workload report with request-to-trial grouping, exclusion and failure counts and the declared aggregation rule",
"produced_artifact": "A frozen benchmark design with evaluators, reference parity, primary objective, timing boundary, quality gates, required regimes, repetition and seed rules and an explicit decision config, plus the evidence-tree layout a PASS claim must produce; for a supplied report, the computed threshold result and missing evidence. Candidate knowledge, not qualification of any model or backend",
"failure_modes": [
@@ -212,7 +271,8 @@
"An upstream better/worse/inconclusive label is used as InferRSI's PASS/FAIL/INCONCLUSIVE verdict",
"Runs on the same hardware class, in different time windows or on shared devices are compared as if pinned",
"A quality pass is taken to certify that the timed work actually occurred",
- "Best-seen results are reported as delivered without restarting and validating the final code/config artifact"
+ "Best-seen results are reported as delivered without restarting and validating the final code/config artifact",
+ "A cache-aware ASR EOU is treated as EOF or full reset, nominal chunk duration as the decoded mel frontier, or shape-only geometry PASS as a native model-state test"
],
"refusal_point": "Records missing coverage and high-variance results as INCONCLUSIVE rather than a win; a run without a frozen RunSpec, or whose validator changed after the plan was frozen, is INCONCLUSIVE regardless of metric values; sets no universal numeric gate or noise threshold"
},
@@ -226,11 +286,16 @@
"Source-derived candidate; no local efficacy validation.",
"Original typed source inventory is not normalized; retain source identities and limits in the procedure/readsets.",
"Task benchmark knowledge is distilled from issue #60 with selected pinned-source rechecks; no all-task runtime qualification or optimization benefit is established.",
- "Benchmark feasibility and package-consumption checks do not promote this candidate or replace task quality, repeated A/B and service qualification."
+ "Benchmark feasibility and package-consumption checks do not promote this candidate or replace task quality, repeated A/B and service qualification.",
+ "The pinned NeMo-Speech.cpp ASR reference covers selected source/test definitions and author-reported benchmark boundaries; native model/device, decoder internals, quality, long-audio/load and service behavior remain not_run.",
+ "JustGoal contributes source-only controlled-evaluation and kernel-acceptance guidance; its repaired/retested timings, released-pass counts and protocol differences are author-reported, not locally reproduced or a complete campaign runner."
],
"provenance_readsets": [
"references/task-benchmark-selection.md",
- "references/benchmark-commands.md"
+ "references/benchmark-commands.md",
+ "references/cache-aware-asr-stream-boundaries.md",
+ "references/justgoal-evaluation-contracts.md",
+ "references/justgoal-source.json"
],
"references": [
"../inference-optimization-casebook/cases/admission-drain-occupancy.md",
@@ -239,8 +304,8 @@
"consumption_mode": "packaged"
},
"applicability": {
- "scope": "Selecting task quality gates, reference parity checks and performance workloads for text, agentic coding, work agents, TTS, full-duplex audio/video, image and video inference, and designing goodput-based multi-regime comparisons that fit InferRSI's evidence trail",
- "prerequisites": "The primary objective, timing boundary, quality gates, sample/failure accounting, required regimes, repetitions, seeds and decision rule are frozen before optimization; baseline and candidate share hardware, software stack, input distribution, benchmark configuration, warmup and measurement protocol, and on a shared machine also one time window and dedicated devices",
+ "scope": "Selecting task quality gates, reference parity checks and performance workloads for text, agentic coding, work agents, cache-aware ASR, TTS, full-duplex audio/video, image and video inference, and designing goodput-based multi-regime comparisons that fit InferRSI's evidence trail",
+ "prerequisites": "The primary objective, timing boundary, quality gates, sample/failure accounting, required regimes, repetitions, seeds and decision rule are frozen before optimization; baseline and candidate share hardware, software stack, input distribution, benchmark configuration, warmup and measurement protocol, and on a shared machine also one time window and dedicated devices; cache-aware ASR also requires the selected model/config and first/steady geometry, decoded clock and reset/EOU/EOF boundaries to be established",
"input_contract": "The task family and deployment use case; candidate quality evaluators and reference parity; the declared SLO and the caller's policy on which regimes must improve or must not regress; for an audit, a supplied fixed-workload report with request-to-trial grouping, exclusion and failure counts and the declared aggregation rule",
"produced_artifact": "A frozen benchmark design with evaluators, reference parity, primary objective, timing boundary, quality gates, required regimes, repetition and seed rules and an explicit decision config, plus the evidence-tree layout a PASS claim must produce; for a supplied report, the computed threshold result and missing evidence. Candidate knowledge, not qualification of any model or backend",
"failure_modes": [
@@ -250,7 +315,8 @@
"An upstream better/worse/inconclusive label is used as InferRSI's PASS/FAIL/INCONCLUSIVE verdict",
"Runs on the same hardware class, in different time windows or on shared devices are compared as if pinned",
"A quality pass is taken to certify that the timed work actually occurred",
- "Best-seen results are reported as delivered without restarting and validating the final code/config artifact"
+ "Best-seen results are reported as delivered without restarting and validating the final code/config artifact",
+ "A cache-aware ASR EOU is treated as EOF or full reset, nominal chunk duration as the decoded mel frontier, or shape-only geometry PASS as a native model-state test"
],
"refusal_point": "Records missing coverage and high-variance results as INCONCLUSIVE rather than a win; a run without a frozen RunSpec, or whose validator changed after the plan was frozen, is INCONCLUSIVE regardless of metric values; sets no universal numeric gate or noise threshold"
},
@@ -268,7 +334,9 @@
"Source-derived candidate; no local efficacy validation.",
"Original typed source inventory is not normalized; retain source identities and limits in the procedure/readsets.",
"Task benchmark knowledge is distilled from issue #60 with selected pinned-source rechecks; no all-task runtime qualification or optimization benefit is established.",
- "Benchmark feasibility and package-consumption checks do not promote this candidate or replace task quality, repeated A/B and service qualification."
+ "Benchmark feasibility and package-consumption checks do not promote this candidate or replace task quality, repeated A/B and service qualification.",
+ "The pinned NeMo-Speech.cpp ASR reference covers selected source/test definitions and author-reported benchmark boundaries; native model/device, decoder internals, quality, long-audio/load and service behavior remain not_run.",
+ "JustGoal contributes source-only controlled-evaluation and kernel-acceptance guidance; its repaired/retested timings, released-pass counts and protocol differences are author-reported, not locally reproduced or a complete campaign runner."
],
"verification": {
"locally_verified": false,
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-optimization-casebook.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-optimization-casebook.json
index 3f8a4f44f..4cf356416 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-optimization-casebook.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-optimization-casebook.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/methods/inference-optimization-casebook/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/methods/inference-optimization-casebook/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/methods/inference-optimization-casebook/SKILL.md",
"src/inferrsi/data/skill_library/methods/inference-optimization-casebook/cases/admission-drain-occupancy.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kernel-writing-guide.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kernel-writing-guide.json
index c4999c702..ad463bd9b 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kernel-writing-guide.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kernel-writing-guide.json
@@ -39,8 +39,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/methods/kernel-writing-guide/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/methods/kernel-writing-guide/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/methods/kernel-writing-guide/SKILL.md",
"src/inferrsi/data/skill_library/methods/kernel-writing-guide/cases/causal-conv1d-dispatch-state.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kv-tier-management.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kv-tier-management.json
index 9ffb79691..3e8c398cf 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kv-tier-management.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kv-tier-management.json
@@ -24,8 +24,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/repositories/lmcache/kv-tier-management/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/repositories/lmcache/kv-tier-management/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/repositories/lmcache/kv-tier-management/SKILL.md",
"src/inferrsi/data/skill_library/repositories/lmcache/kv-tier-management/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lazy-evaluation-stream-lifetimes.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lazy-evaluation-stream-lifetimes.json
index 37c1ec38c..4c97a5bb2 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lazy-evaluation-stream-lifetimes.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lazy-evaluation-stream-lifetimes.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/SKILL.md",
"src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lily.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lily.json
index 46b15dd95..e7875efa9 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lily.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lily.json
@@ -25,8 +25,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/repositories/lily/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/repositories/lily/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/repositories/lily/LICENSE",
"src/inferrsi/data/skill_library/repositories/lily/NOTICE.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lithos-metal.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lithos-metal.json
index 2fbc1e211..d3aa5fa09 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lithos-metal.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lithos-metal.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/repositories/lithos-metal/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/repositories/lithos-metal/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/repositories/lithos-metal/LICENSE",
"src/inferrsi/data/skill_library/repositories/lithos-metal/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/local-dream.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/local-dream.json
index 2c1afedd7..5d641cc45 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/local-dream.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/local-dream.json
@@ -29,8 +29,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/repositories/local-dream/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/repositories/local-dream/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/repositories/local-dream/SKILL.md",
"src/inferrsi/data/skill_library/repositories/local-dream/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mistral-rs.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mistral-rs.json
index 97b18bf21..55fb63635 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mistral-rs.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mistral-rs.json
@@ -26,8 +26,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/repositories/mistral-rs/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/repositories/mistral-rs/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/repositories/mistral-rs/LICENSE",
"src/inferrsi/data/skill_library/repositories/mistral-rs/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlc-llm.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlc-llm.json
index 065eeab0d..00a38685c 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlc-llm.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlc-llm.json
@@ -27,8 +27,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/repositories/mlc-llm/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/repositories/mlc-llm/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/repositories/mlc-llm/LICENSE",
"src/inferrsi/data/skill_library/repositories/mlc-llm/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlx-swift-lm.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlx-swift-lm.json
index 33298fa72..1dced88de 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlx-swift-lm.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlx-swift-lm.json
@@ -27,8 +27,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/repositories/mlx-swift-lm/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/repositories/mlx-swift-lm/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/repositories/mlx-swift-lm/LICENSE",
"src/inferrsi/data/skill_library/repositories/mlx-swift-lm/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/npu-skill-factory.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/npu-skill-factory.json
index 995c075fc..4d6ec218d 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/npu-skill-factory.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/npu-skill-factory.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/workflows/npu-skill-factory/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/workflows/npu-skill-factory/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/skill-index.json",
"src/inferrsi/data/skill_library/workflows/npu-skill-factory/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/optimization-search-policy.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/optimization-search-policy.json
index 38c73f365..13a8349da 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/optimization-search-policy.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/optimization-search-policy.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/methods/optimization-search-policy/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/methods/optimization-search-policy/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/methods/optimization-search-policy/SKILL.md",
"src/inferrsi/data/skill_library/methods/optimization-search-policy/cases/sarathi-chunk-budget.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/os-process-placement.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/os-process-placement.json
index 9b9fe366f..d91899f66 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/os-process-placement.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/os-process-placement.json
@@ -24,8 +24,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/hardware/host/os-process-placement/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/hardware/host/os-process-placement/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/hardware/host/os-process-placement/SKILL.md",
"src/inferrsi/data/skill_library/hardware/host/os-process-placement/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/performance-ceiling-analysis.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/performance-ceiling-analysis.json
index d74d01244..251b7694e 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/performance-ceiling-analysis.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/performance-ceiling-analysis.json
@@ -22,11 +22,12 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/SKILL.md",
"src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/metadata.json",
+ "src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/references/flexgen-offload-feasibility.md",
"src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/references/mooncake-prefix-reuse-feasibility.md",
"src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/references/papers-20261015/INTEGRATION_NOTES.md",
"src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/references/papers-20261015/MECHANISMS.md",
@@ -49,7 +50,7 @@
"src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/scripts/roofline.py",
"src/inferrsi/data/skill_library/skill-index.json"
],
- "content_sha256": "4810761a3ab1e5e6e000060d9ad140a0e8892937486f2ab384d50868c22b0f11"
+ "content_sha256": "0203dfabd2a02d2ab8f8696b922dcc3df86b3a811352a5bc6cea57aee68e869c"
}
],
"index_type": "packaged-skill",
@@ -58,8 +59,9 @@
},
"checkout_ref": "checkout:performance-ceiling-analysis",
"source_files": {
- "src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/SKILL.md": "4810761a3ab1e5e6e000060d9ad140a0e8892937486f2ab384d50868c22b0f11",
- "src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/metadata.json": "3fcd2126846cf42ee77b1def39fc05ebfc54cf7428e0a6844b0ade1acf9f47ea",
+ "src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/SKILL.md": "0203dfabd2a02d2ab8f8696b922dcc3df86b3a811352a5bc6cea57aee68e869c",
+ "src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/metadata.json": "70cd45ab665d4a83544a55fb48f2c83f195715a39af1e8e238ab574403150212",
+ "src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/references/flexgen-offload-feasibility.md": "984cc41c4412eafa6d20e109dea8927ad321a668776bc86bbffda0a94dfa9cc7",
"src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/references/mooncake-prefix-reuse-feasibility.md": "e64b62fded4490708414a898f540b0319c54d2b205fd5d060acaab419734d100",
"src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/references/papers-20261015/INTEGRATION_NOTES.md": "083e462e0a204dce599362d46712fcfb392d053c5cda56216e22e28d4b8fa0fa",
"src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/references/papers-20261015/MECHANISMS.md": "742c249082ceca230326f4407a40130fe624553b796ae5a157a8a4c15a7e1398",
@@ -140,6 +142,20 @@
],
"readset": "references/mooncake-prefix-reuse-feasibility.md",
"license": "UNKNOWN"
+ },
+ {
+ "type": "paper",
+ "id": "pmlr-v202-sheng23a",
+ "version": "ICML 2023 official ten-author edition, PMLR 202:31094-31116",
+ "url": "https://proceedings.mlr.press/v202/sheng23a/sheng23a.pdf",
+ "content_sha256": "7cd9673c37640011d38ef4b8595f5e1e09a85b1f803a3734f19c97ca66c75b9e",
+ "sections": [
+ "1 and 3; 4.1-4.3, Figures 1-3, Algorithm 1, Equation 1",
+ "4.4 (scope boundary only); 6 and 6.1, Tables 2-3 (denominators and ablations)",
+ "Appendix A.3 selected cost/peak-memory formulas, PDF pages 15-17; A.2 ending only (theorem boundary)"
+ ],
+ "readset": "references/flexgen-offload-feasibility.md",
+ "license": "CC-BY-4.0"
}
],
"verification": {
@@ -167,9 +183,11 @@
"empirical_validation": "not_run",
"limitations": [
"Source-derived candidate; no local efficacy validation.",
- "Mooncake's prefix-reuse condition is an analytical approximation; current cache validity, placement, effective rates, overlap and SLO outcomes remain unverified."
+ "Mooncake's prefix-reuse condition is an analytical approximation; current cache validity, placement, effective rates, overlap and SLO outcomes remain unverified.",
+ "FlexGen's cross-tier feasibility comparison assumes fitted rates and perfect overlap; relaxed placements may OOM, the printed bls wording remains unresolved, and current compatibility, latency and efficacy are unverified."
],
"provenance_readsets": [
+ "references/flexgen-offload-feasibility.md",
"references/mooncake-prefix-reuse-feasibility.md",
"references/papers-20261015/cases/decision-cases.md",
"references/papers-20261015/govern-optimization-experiments/procedure.md",
@@ -205,7 +223,8 @@
"Reading or indexing does not authorize execution, validate efficacy, promote a candidate or grant a license.",
"The backend categorizes the procedure's hardware dependency; it does not certify compatibility, and skills.discovery is a discovery scenario.",
"Source-derived candidate; no local efficacy validation.",
- "Mooncake's prefix-reuse condition is an analytical approximation; current cache validity, placement, effective rates, overlap and SLO outcomes remain unverified."
+ "Mooncake's prefix-reuse condition is an analytical approximation; current cache validity, placement, effective rates, overlap and SLO outcomes remain unverified.",
+ "FlexGen's cross-tier feasibility comparison assumes fitted rates and perfect overlap; relaxed placements may OOM, the printed bls wording remains unresolved, and current compatibility, latency and efficacy are unverified."
],
"verification": {
"locally_verified": false,
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/plan-activation-residency.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/plan-activation-residency.json
index 77fead2ce..a99ef44cb 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/plan-activation-residency.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/plan-activation-residency.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/methods/plan-activation-residency/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/methods/plan-activation-residency/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/methods/plan-activation-residency/SKILL.md",
"src/inferrsi/data/skill_library/methods/plan-activation-residency/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/project-maya.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/project-maya.json
index cb38b28d4..85f6a9fe5 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/project-maya.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/project-maya.json
@@ -25,8 +25,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/repositories/project-maya/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/repositories/project-maya/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/repositories/project-maya/LICENSE",
"src/inferrsi/data/skill_library/repositories/project-maya/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/repository-evolution-analysis.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/repository-evolution-analysis.json
index 3edca3450..f1811d204 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/repository-evolution-analysis.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/repository-evolution-analysis.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/methods/repository-evolution-analysis/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/methods/repository-evolution-analysis/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/methods/repository-evolution-analysis/SKILL.md",
"src/inferrsi/data/skill_library/methods/repository-evolution-analysis/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/rsi-evidence.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/rsi-evidence.json
index 8d542abf9..bcdcc3afd 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/rsi-evidence.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/rsi-evidence.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/workflows/rsi-evidence/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/workflows/rsi-evidence/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/skill-index.json",
"src/inferrsi/data/skill_library/workflows/rsi-evidence/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/skill-evolution-playbook.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/skill-evolution-playbook.json
index b1b494df5..2289e3110 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/skill-evolution-playbook.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/skill-evolution-playbook.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/methods/skill-evolution-playbook/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/methods/skill-evolution-playbook/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/methods/skill-evolution-playbook/SKILL.md",
"src/inferrsi/data/skill_library/methods/skill-evolution-playbook/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/splash.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/splash.json
index 7ce820099..8e8de52f3 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/splash.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/splash.json
@@ -25,8 +25,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/repositories/splash/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/repositories/splash/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/repositories/splash/LICENSE",
"src/inferrsi/data/skill_library/repositories/splash/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/storage-io-measurement.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/storage-io-measurement.json
index 66c9defc1..8bc9231e4 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/storage-io-measurement.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/storage-io-measurement.json
@@ -24,8 +24,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/hardware/host/storage-io-measurement/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/hardware/host/storage-io-measurement/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/hardware/host/storage-io-measurement/SKILL.md",
"src/inferrsi/data/skill_library/hardware/host/storage-io-measurement/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-input-pipeline.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-input-pipeline.json
index 644f8c2a1..ce75694ad 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-input-pipeline.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-input-pipeline.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/methods/training-input-pipeline/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/methods/training-input-pipeline/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/methods/training-input-pipeline/SKILL.md",
"src/inferrsi/data/skill_library/methods/training-input-pipeline/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-resource-budget.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-resource-budget.json
index efe943ba2..4dd3878e3 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-resource-budget.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-resource-budget.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/methods/training-resource-budget/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/methods/training-resource-budget/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/methods/training-resource-budget/SKILL.md",
"src/inferrsi/data/skill_library/methods/training-resource-budget/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/video-diffusion-optimization.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/video-diffusion-optimization.json
index 6df1833ba..bb3e3aa4d 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/video-diffusion-optimization.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/video-diffusion-optimization.json
@@ -26,8 +26,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/methods/video-diffusion-optimization/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/methods/video-diffusion-optimization/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/methods/video-diffusion-optimization/SKILL.md",
"src/inferrsi/data/skill_library/methods/video-diffusion-optimization/assets/campaign.template.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/vllm-metal.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/vllm-metal.json
index ecb493542..3776d52d5 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/vllm-metal.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/vllm-metal.json
@@ -26,8 +26,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/repositories/vllm-metal/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/repositories/vllm-metal/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/repositories/vllm-metal/LICENSE",
"src/inferrsi/data/skill_library/repositories/vllm-metal/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/weight-loading-optimization.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/weight-loading-optimization.json
index d101d12cd..9e0124b10 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/weight-loading-optimization.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/weight-loading-optimization.json
@@ -29,8 +29,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/methods/weight-loading-optimization/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/methods/weight-loading-optimization/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/methods/weight-loading-optimization/SKILL.md",
"src/inferrsi/data/skill_library/methods/weight-loading-optimization/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-afd-p2p-exchanges.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-afd-p2p-exchanges.json
index 14c7effb8..1f7e75553 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-afd-p2p-exchanges.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-afd-p2p-exchanges.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/SKILL.md",
"src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-flare-session-lifecycle.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-flare-session-lifecycle.json
index 000544f88..f0443368f 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-flare-session-lifecycle.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-flare-session-lifecycle.json
@@ -29,8 +29,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/LICENSE",
"src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/diagnose-attention-bottlenecks.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/diagnose-attention-bottlenecks.json
index 4d70595fd..280f92791 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/diagnose-attention-bottlenecks.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/diagnose-attention-bottlenecks.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/hardware/cuda/diagnose-attention-bottlenecks/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/hardware/cuda/diagnose-attention-bottlenecks/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/hardware/cuda/diagnose-attention-bottlenecks/SKILL.md",
"src/inferrsi/data/skill_library/hardware/cuda/diagnose-attention-bottlenecks/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/host-overhead-profiling.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/host-overhead-profiling.json
index ab9a3acbb..48930d1e2 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/host-overhead-profiling.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/host-overhead-profiling.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/hardware/cuda/host-overhead-profiling/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/hardware/cuda/host-overhead-profiling/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/hardware/cuda/host-overhead-profiling/SKILL.md",
"src/inferrsi/data/skill_library/hardware/cuda/host-overhead-profiling/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/reason-hopper-kernels.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/reason-hopper-kernels.json
index f4108a23c..10dba6944 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/reason-hopper-kernels.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/reason-hopper-kernels.json
@@ -36,8 +36,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/daa33a4ba588298ff76e3e077d6c894d7750d9bc/src/inferrsi/data/skill_library/hardware/cuda/reason-hopper-kernels/SKILL.md",
- "commit": "daa33a4ba588298ff76e3e077d6c894d7750d9bc",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/hardware/cuda/reason-hopper-kernels/SKILL.md",
+ "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
"paths": [
"src/inferrsi/data/skill_library/hardware/cuda/reason-hopper-kernels/SKILL.md",
"src/inferrsi/data/skill_library/hardware/cuda/reason-hopper-kernels/metadata.json",
From b6091aa98b2ad674c6d4b634f1a864cea11e1ce8 Mon Sep 17 00:00:00 2001
From: Dong1017 <55723162+Dong1017@users.noreply.github.com>
Date: Tue, 3 Nov 2026 12:26:15 +0800
Subject: [PATCH 10/11] docs: refresh discovery after framework graph
integration
Bind all projections to combined source snapshot ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90.
Agent-assisted-by: Codex / gpt-6-astra [effort=xhigh] (author)
Signed-off-by: Dong1017 <55723162+Dong1017@users.noreply.github.com>
---
docs/skill-coverage.md | 48 +++++---
.../repo_packs/inferrsi/knowledge/README.md | 8 +-
.../ascend-aclgraph-capture-replay.json | 6 +-
.../ascend/skills/ascend-device-health.json | 6 +-
.../skills/ascend-distributed-parallel.json | 6 +-
.../ascend/skills/ascend-graph-mode.json | 6 +-
.../skills/ascend-model-adaptation.json | 6 +-
.../skills/ascend-operator-profiling.json | 6 +-
.../ascend/skills/ascend-operator-tuning.json | 6 +-
.../skills/ascend-optimization-casebook.json | 6 +-
.../skills/ascend-performance-analysis.json | 6 +-
.../ascend-video-diffusion-optimization.json | 6 +-
.../ascend/skills/ascendc-tiling-design.json | 6 +-
.../skills/cann-attribution-overlap.json | 6 +-
.../knowledge/ascend/skills/torch-npu.json | 4 +-
.../torchnpu-device-memory-lifecycle.json | 6 +-
...torchnpu-pinned-host-memory-lifecycle.json | 6 +-
.../skills/assess-serving-preemption.json | 6 +-
.../assess-split-backward-schedule.json | 6 +-
.../skills/audit-fixed-workload-claim.json | 6 +-
.../audit-h3-metal-fasth3-semantics.json | 6 +-
.../skills/audit-h3-runtime-residency.json | 6 +-
.../skills/audit-llama-runtime-state.json | 6 +-
.../audit-recurrent-kv-finalization.json | 6 +-
.../common/skills/audit-sd-cpp-residency.json | 6 +-
.../audit-sglang-diffusion-lifetimes.json | 6 +-
.../skills/audit-vllm-kv-lifetimes.json | 6 +-
.../bound-windowed-attention-replay.json | 6 +-
.../budget-speculative-verification.json | 6 +-
.../skills/calibrate-serving-simulation.json | 6 +-
.../skills/cgroup-resource-controls.json | 6 +-
.../checkpoint-recovery-consistency.json | 6 +-
.../skills/choose-parallel-training-plan.json | 6 +-
.../skills/cite-check-downstream-pr.json | 6 +-
.../common/skills/claim-contract.json | 6 +-
.../knowledge/common/skills/comfy-ui.json | 6 +-
.../common/skills/competitor-scan.json | 6 +-
.../skills/dcp-async-save-performance.json | 6 +-
.../skills/dense-performance-evidence.json | 6 +-
.../diagnose-vllm-admission-pressure.json | 6 +-
.../skills/diffusers-invocation-state.json | 6 +-
.../common/skills/draft-acceptance-issue.json | 6 +-
.../knowledge/common/skills/dwarfstar.json | 6 +-
.../evaluate-elastic-serving-placement.json | 6 +-
.../skills/generation-cache-lifecycle.json | 6 +-
.../common/skills/host-memory-management.json | 6 +-
.../skills/host-overhead-optimization.json | 6 +-
.../knowledge/common/skills/husky.json | 6 +-
.../skills/inference-benchmark-design.json | 112 ++++++++++++++----
.../inference-optimization-casebook.json | 6 +-
.../common/skills/kernel-writing-guide.json | 6 +-
.../common/skills/kv-tier-management.json | 6 +-
.../lazy-evaluation-stream-lifetimes.json | 6 +-
.../knowledge/common/skills/lily.json | 6 +-
.../knowledge/common/skills/lithos-metal.json | 6 +-
.../knowledge/common/skills/local-dream.json | 6 +-
.../knowledge/common/skills/mistral-rs.json | 6 +-
.../knowledge/common/skills/mlc-llm.json | 6 +-
.../knowledge/common/skills/mlx-swift-lm.json | 6 +-
.../common/skills/npu-skill-factory.json | 6 +-
.../skills/optimization-search-policy.json | 6 +-
.../common/skills/os-process-placement.json | 6 +-
.../skills/performance-ceiling-analysis.json | 6 +-
.../skills/plan-activation-residency.json | 6 +-
.../knowledge/common/skills/project-maya.json | 6 +-
.../knowledge/common/skills/pytorch.json | 4 +-
.../skills/repository-evolution-analysis.json | 6 +-
.../knowledge/common/skills/rsi-evidence.json | 6 +-
.../skills/skill-evolution-playbook.json | 6 +-
.../knowledge/common/skills/splash.json | 6 +-
.../common/skills/storage-io-measurement.json | 6 +-
.../skills/training-input-pipeline.json | 6 +-
.../skills/training-resource-budget.json | 6 +-
.../skills/video-diffusion-optimization.json | 6 +-
.../knowledge/common/skills/vllm-metal.json | 6 +-
.../skills/weight-loading-optimization.json | 6 +-
.../cuda/skills/audit-afd-p2p-exchanges.json | 6 +-
.../skills/audit-flare-session-lifecycle.json | 6 +-
.../diagnose-attention-bottlenecks.json | 6 +-
.../cuda/skills/host-overhead-profiling.json | 6 +-
.../cuda/skills/reason-hopper-kernels.json | 6 +-
81 files changed, 359 insertions(+), 273 deletions(-)
diff --git a/docs/skill-coverage.md b/docs/skill-coverage.md
index 853eb321c..3b0c56ccf 100644
--- a/docs/skill-coverage.md
+++ b/docs/skill-coverage.md
@@ -4,7 +4,7 @@ Generated from the registry, the skill sidecars and the two maintained
matrices. **Do not edit by hand**; regenerate after any registry, skill,
matrix or frozen-plan change.
-- Input digest: `sha256:946ec9585c7e530125131a12ef50666c490fb436c4a2c60c0a638ea7c88bcdf5`
+- Input digest: `sha256:a314b58c61ab9e0afcbfff44d1d16f1a775d392e3b7133006619994e3529814a`
- Regenerate: `python tools/build_skill_coverage.py`
- Verify without writing: `python tools/build_skill_coverage.py --check`
@@ -32,17 +32,17 @@ matrix: every field is copied from an owner that already maintains it.
| Measure | Value |
|---|---|
-| Registered skill ids | 77 |
-| By category | frameworks 3, hardware 22, methods 20, repositories 26, workflows 6 |
-| By backend | ascend 14, common 58, cuda 5 |
+| Registered skill ids | 79 |
+| By category | frameworks 5, hardware 22, methods 20, repositories 26, workflows 6 |
+| By backend | ascend 15, common 59, cuda 5 |
| Recorded on a numbered SearchSpace resource layer | 49 |
-| Recorded only in a supplementary guidance table (†) | 13 |
+| Recorded only in a supplementary guidance table (†) | 15 |
| Recorded on two or more resource layers | 24 |
-| Recorded in the Reference Matrix | 69 |
-| With a `metadata.json` sidecar | 77 |
+| Recorded in the Reference Matrix | 71 |
+| With a `metadata.json` sidecar | 79 |
| Selected by a frozen evaluation plan | 2 |
-| `SKILL.md` documents in the library | 124 |
-| Unregistered documents (excluded from every count above) | 47 |
+| `SKILL.md` documents in the library | 142 |
+| Unregistered documents (excluded from every count above) | 63 |
A complete inventory is not a coverage claim. Every registered row below
has an entry; that says the listing is exhaustive, not that inference
@@ -75,13 +75,15 @@ optimization is fully covered or that any procedure has measured benefit.
| [torchnpu-device-memory-lifecycle](../src/inferrsi/data/skill_library/hardware/ascend/torchnpu-device-memory-lifecycle/SKILL.md) | ascend | Diagnose TorchNPU device-HBM reuse, cross-stream lifetimes, cache retention and fragmentation, then propose ... | **unknown** | — | 1 Accelerator · Diagnose / Configure / Optimize / Verify | 1 · Systems sources (Ascend/pytorch (TorchNPU) @ `67511c53cd4c88014f6123de7097677e60e85faa` ...) | candidate · not_run · 1 readset(s) | no-applicability, no-eval |
| [torchnpu-pinned-host-memory-lifecycle](../src/inferrsi/data/skill_library/hardware/ascend/torchnpu-pinned-host-memory-lifecycle/SKILL.md) | ascend | Diagnose TorchNPU pinned-host RAM allocation, asynchronous-copy lifetime, reuse and cache controls, then propose ... | **unknown** | — | 2 RAM/Host · Diagnose / Configure / Optimize / Verify | 1 · Systems sources (Ascend/pytorch (TorchNPU) @ `67511c53cd4c88014f6123de7097677e60e85faa` ...) | candidate · not_run · 1 readset(s) | no-applicability, no-eval |
-## Registered: Frameworks (3)
+## Registered: Frameworks (5)
| Skill | Backend | Declared task | Applicability | Conditions | SearchSpace layer / stage | Reference Matrix | State | Gaps |
|---|---|---|---|---|---|---|---|---|
-| [dcp-async-save-performance](../src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/SKILL.md) | common | Diagnose training-visible checkpoint pauses using PyTorch Distributed Checkpoint asynchronous saving and staging ... | Source-supported diagnosis of training-visible PyTorch DCP save pauses ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator · Diagnose / Configure / Optimize
2 RAM/Host · Observe / Profile / Configure / Optimize / Verify
3 SSD · Observe / Profile / Configure / Optimize / Verify | 1 · Systems sources (C1 / DCP async-save-only approval — pytorch/pytorch @ ...) | candidate · not_run · 1 readset(s) | no-eval |
+| [dcp-async-save-performance](../src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/SKILL.md) | common | Diagnose training-visible checkpoint pauses using PyTorch Distributed Checkpoint asynchronous saving and staging ... | Source-supported diagnosis of training-visible PyTorch DCP save pauses ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator · Diagnose / Configure / Optimize
2 RAM/Host · Observe / Profile / Configure / Optimize / Verify †guidance: PyTorch framework source procedures
3 SSD · Observe / Profile / Configure / Optimize / Verify †guidance: PyTorch framework source procedures | 1 · Systems sources (C1 / DCP async-save-only approval — pytorch/pytorch @ ...) | candidate · not_run · 1 readset(s) | no-eval |
| [diffusers-invocation-state](../src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/SKILL.md) | common | Trace Diffusers pipeline-wrapper and scheduler invocation state when repeated calls, from_pipe reuse or interruption ... | FluxPipeline and FlowMatchEulerDiscreteScheduler invocation ownership at this source pin ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 2 RAM/Host †guidance: Diffusers invocation guidance (source-derived) | Diffusers invocation source extraction (huggingface/diffusers @ `aea7b70be0f594bcdc15a6f2677ee2ca85e49e8e` ...) | candidate · not_run · 1 readset(s) | no-eval |
| [lazy-evaluation-stream-lifetimes](../src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/SKILL.md) | common | Diagnose MLX lazy evaluation, cross-stream ordering and array-buffer lifetimes when timing, memory use or asynchronous ... | MLX evaluation/stream/buffer diagnosis at the pinned implementation; qualify the actual ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator †guidance: MLX framework guidance (source-derived) | MLX framework source extraction (ml-explore/mlx @ `bff10220c742907f0c68e55ff174aaea56f74e3d` (MIT)) | candidate · not_run · 1 readset(s) | no-eval |
+| [pytorch](../src/inferrsi/data/skill_library/frameworks/pytorch/pytorch/SKILL.md) | common | Operate and diagnose PyTorch installation, tensors and autograd, training and data, compilation and export, accelerator ... | Major user workflows and selected deep mechanisms of pinned pytorch/pytorch source; exact ... | failure_modes, input_contract, produced_artifact, refusal_point | 1 Accelerator †guidance: PyTorch framework source procedures
2 RAM/Host †guidance: PyTorch framework source procedures
3 SSD †guidance: PyTorch framework source procedures
4 Network †guidance: PyTorch framework source procedures | PyTorch framework operating graph (pytorch/pytorch @ `9a90a4fefcaafa63ac984621963a6fcc66aee68c`, source ...) | candidate · not_run · 1 readset(s) | no-eval |
+| [torch-npu](../src/inferrsi/data/skill_library/frameworks/torch-npu/torch-npu/SKILL.md) | ascend | Operate and diagnose TorchNPU on Ascend across installation, PrivateUse1 dispatch, operator development, memory and ... | Source-grounded TorchNPU operation and diagnosis; qualify the selected installed stack ... | failure_modes, input_contract, produced_artifact, refusal_point | 1 Accelerator †guidance: TorchNPU framework source procedures
2 RAM/Host †guidance: TorchNPU framework source procedures
4 Network †guidance: TorchNPU framework source procedures | TorchNPU framework operating graph (Ascend/pytorch @ `046934140b96836c3fcef14d0f39ae54dc0a06bf` ...) | candidate · not_run · 2 readset(s) | no-eval |
## Registered: Repositories (26)
@@ -126,7 +128,7 @@ optimization is fully covered or that any procedure has measured benefit.
| [choose-parallel-training-plan](../src/inferrsi/data/skill_library/methods/choose-parallel-training-plan/SKILL.md) | common | Compare stage, device-mesh and operator-sharding candidates for static distributed training while accounting for ... | Comparing stage, device-mesh and operator-sharding candidates for static distributed ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator · Diagnose / Configure / Optimize / Verify
4 Network · Diagnose / Configure / Optimize / Verify | 1 · Systems sources (Alpa — [2201.12023v3](https://arxiv.org/abs/2201.12023v3)) | candidate · not_run · 1 readset(s) | no-eval |
| [dense-performance-evidence](../src/inferrsi/data/skill_library/methods/dense-performance-evidence/SKILL.md) | common | Taxonomy of performance evidence an optimization agent needs, ordered by collection cost and diagnostic power. Maps ... | Deciding which performance evidence to collect next and which claim each evidence level ... | failure_modes, input_contract, produced_artifact, refusal_point
4 advice | 1 Accelerator · Observe / Profile / Verify
2 RAM/Host · Verify
4 Network · Verify | 1 · Systems sources (NVIDIA/TensorRT-LLM @ f33e0e5fd)
2 · Methodology and corpus references (JiusiServe/vllm-evolve @ 43cc82c)
2 · Methodology and corpus references (microsoft/SkillOpt @ 343db22) | candidate · not_run · 0 readset(s) | — |
| [evaluate-elastic-serving-placement](../src/inferrsi/data/skill_library/methods/evaluate-elastic-serving-placement/SKILL.md) | common | Compare migrating an LLM request between model replicas with changing its sequence-parallel execution group, including ... | Comparing migration of a running LLM request between compatible model replicas with ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator · Diagnose / Configure / Optimize / Verify
2 RAM/Host · Diagnose / Configure / Optimize / Verify
4 Network · Diagnose / Configure / Optimize / Verify | 1 · Systems sources (Llumnix — [2406.03243v1](https://arxiv.org/abs/2406.03243v1)) | candidate · not_run · 1 readset(s) | no-eval |
-| [inference-benchmark-design](../src/inferrsi/data/skill_library/methods/inference-benchmark-design/SKILL.md) | common | Select task quality gates, reference parity checks and performance workloads for text, agentic coding, work agents ... | Selecting task quality gates, reference parity checks and performance workloads for text ... | failure_modes, input_contract, produced_artifact, refusal_point
4 advice | — | 2 · Methodology and corpus references (JiusiServe/vllm-evolve @ 43cc82c)
Task-specific benchmark selection (issue #60) ([Issue #60 report](https://github.com/InferMatrixLab/InferRSI/issues/6 ...)
Task-specific benchmark selection (issue #60) ([NVIDIA/NeMo-Speech.cpp @ 20999cc9](https://github.com/NVIDIA/NeMo-Spe ...) | candidate · not_run · 5 readset(s) | no-SearchSpace-layer |
+| [inference-benchmark-design](../src/inferrsi/data/skill_library/methods/inference-benchmark-design/SKILL.md) | common | Select task quality gates, reference parity checks and performance workloads for text, agentic coding, work agents ... | Selecting task quality gates, reference parity checks and performance workloads for text ... | failure_modes, input_contract, produced_artifact, refusal_point
5 advice | — | 2 · Methodology and corpus references (JiusiServe/vllm-evolve @ 43cc82c)
Task-specific benchmark selection (issue #60) ([Issue #60 report](https://github.com/InferMatrixLab/InferRSI/issues/6 ...)
Task-specific benchmark selection (issue #60) ([NVIDIA/NeMo-Speech.cpp @ 20999cc9](https://github.com/NVIDIA/NeMo-Spe ...) | candidate · not_run · 7 readset(s) | no-SearchSpace-layer |
| [inference-optimization-casebook](../src/inferrsi/data/skill_library/methods/inference-optimization-casebook/SKILL.md) | common | How to structure and use a casebook of past optimization decisions as a decision-support tool. Extracts the casebook ... | Structuring and querying a casebook of past optimization decisions as decision support ... | failure_modes, input_contract, produced_artifact, refusal_point
4 advice | 1 Accelerator · Optimize | 1 · Systems sources (NVIDIA/TensorRT-LLM @ f33e0e5fd) | candidate · not_run · 0 readset(s) | no-eval |
| [kernel-writing-guide](../src/inferrsi/data/skill_library/methods/kernel-writing-guide/SKILL.md) | common | Decision guide for writing custom inference kernels: when to write one, which framework to choose (Triton, CuTe DSL ... | Deciding whether to write a custom inference kernel, routing a kernel pattern to Triton ... | failure_modes, input_contract, produced_artifact, refusal_point
6 advice | 1 Accelerator · Optimize / Diagnose / Configure / Verify | 1 · Systems sources (Dao-AILab/causal-conv1d @ `cd81f0413cad2fc1e6f17e785ac39f59aae690cd` ...)
1 · Systems sources (NVIDIA/TensorRT-LLM @ f33e0e5fd)
1 · Systems sources (huggingface/transformers @ `0dad7b822255a0ae261ec45ae937371e859ffd1a` ...)
1 · Systems sources (mindspore-ai/akg @ `5aa15f3fe4d856eb60cd3b27447e04f5f1fb4dff` ...)
1 · Systems sources (sgl-project/sglang @ `16176278d0180f1a1a578e53e6eef432577876c2` ...)
1 · Systems sources (vllm-project/vllm @ `537a79f0fc5ba564d2bef43d54915b2f80f9f77d` (root ...) | candidate · not_run · 7 readset(s) | no-eval |
| [optimization-search-policy](../src/inferrsi/data/skill_library/methods/optimization-search-policy/SKILL.md) | common | Policy for how an optimization agent should select, order, and stop optimization experiments. Covers candidate ... | Selecting, ordering and stopping optimization experiments: candidate generation ... | failure_modes, input_contract, produced_artifact, refusal_point
4 advice | 1 Accelerator · Optimize | 2 · Methodology and corpus references (JiusiServe/vllm-evolve @ 43cc82c)
2 · Methodology and corpus references (RRSI — arXiv 2609.24972v3)
2 · Methodology and corpus references (arXiv 2608.01867 (CRISP))
2 · Methodology and corpus references (microsoft/SkillOpt @ 343db22) | candidate · not_run · 9 readset(s) | no-eval |
@@ -152,13 +154,29 @@ optimization is fully covered or that any procedure has measured benefit.
## Documents outside the registry
-47 `SKILL.md` documents are not registered, so they have
+63 `SKILL.md` documents are not registered, so they have
no stable id and no row above. They are engine entrypoints, their
sub-skills, and the research router. Counting them would inflate coverage:
an entrypoint routes to procedures below it rather than adding one.
| Directory | Kind |
|---|---|
+| `frameworks/pytorch/pytorch/sub-skills/compiler-and-export` | sub-skill |
+| `frameworks/pytorch/pytorch/sub-skills/custom-operators-and-extension` | sub-skill |
+| `frameworks/pytorch/pytorch/sub-skills/distributed-and-checkpoint` | sub-skill |
+| `frameworks/pytorch/pytorch/sub-skills/installation-and-build` | sub-skill |
+| `frameworks/pytorch/pytorch/sub-skills/memory-streams-and-graphs` | sub-skill |
+| `frameworks/pytorch/pytorch/sub-skills/profiling-and-debugging` | sub-skill |
+| `frameworks/pytorch/pytorch/sub-skills/tensor-dispatch-autograd` | sub-skill |
+| `frameworks/pytorch/pytorch/sub-skills/training-and-data` | sub-skill |
+| `frameworks/torch-npu/torch-npu/sub-skills/distributed-and-hccl` | sub-skill |
+| `frameworks/torch-npu/torch-npu/sub-skills/graph-and-compile` | sub-skill |
+| `frameworks/torch-npu/torch-npu/sub-skills/installation-and-compatibility` | sub-skill |
+| `frameworks/torch-npu/torch-npu/sub-skills/memory-streams-and-events` | sub-skill |
+| `frameworks/torch-npu/torch-npu/sub-skills/operator-development` | sub-skill |
+| `frameworks/torch-npu/torch-npu/sub-skills/privateuse1-and-dispatch` | sub-skill |
+| `frameworks/torch-npu/torch-npu/sub-skills/profiling-and-debugging` | sub-skill |
+| `frameworks/torch-npu/torch-npu/sub-skills/training-and-precision` | sub-skill |
| `repositories/dwarfstar/sub-skills/qualification` | sub-skill |
| `repositories/dwarfstar/sub-skills/runtime-memory` | sub-skill |
| `repositories/dwarfstar/sub-skills/serving-state` | sub-skill |
@@ -209,7 +227,7 @@ an entrypoint routes to procedures below it rather than adding one.
## Integrity checks
-- Registered ids resolve to an existing `SKILL.md`: 77 of 77.
+- Registered ids resolve to an existing `SKILL.md`: 79 of 79.
- Duplicate ids or paths: none; `skill_catalog` rejects them before this view is built.
- Skill links in the two matrices that do not resolve: 0.
- Registered skills without a `metadata.json` sidecar: 0.
@@ -222,7 +240,7 @@ an entrypoint routes to procedures below it rather than adding one.
| `no-Reference-Matrix-row` | 8: ascend-video-diffusion-optimization, audit-fixed-workload-claim, cite-check-downstream-pr, claim-contract, competitor-scan, draft-acceptance-issue, rsi-evidence, video-diffusion-optimization |
| `no-sidecar` | 0: — |
| `no-applicability` | 20: ascend-aclgraph-capture-replay, ascend-device-health, ascend-distributed-parallel, ascend-graph-mode, ascend-model-adaptation, ascend-operator-profiling, ascend-operator-tuning, ascend-optimization-casebook, ascend-performance-analysis, ascendc-tiling-design, cann-attribution-overlap, cgroup-resource-controls, diagnose-attention-bottlenecks, host-memory-management, host-overhead-optimization, host-overhead-profiling, os-process-placement, storage-io-measurement, torchnpu-device-memory-lifecycle, torchnpu-pinned-host-memory-lifecycle |
-| `no-eval` | 75: as listed in the Gaps column above |
+| `no-eval` | 77: as listed in the Gaps column above |
These are inventory gaps, not defects. A workflow skill may legitimately
have no hardware resource layer; a candidate may legitimately be untested.
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/README.md b/src/inferrsi/data/repo_packs/inferrsi/knowledge/README.md
index daaa46804..400fbecbc 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/README.md
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/README.md
@@ -1,6 +1,6 @@
# InferRSI skill discovery index
-Generated abstracts for all 77 registered skills in `data/skill_library/`.
+Generated abstracts for all 79 registered skills in `data/skill_library/`.
Their complete procedures, references and cases ship in the same package.
`skill-index.json` maps stable skill IDs to categorized directories. The
reference and searchspace matrices
@@ -15,7 +15,7 @@ These are knowledge abstracts (`--kind knowledge`), with namespaced package
IDs distinct from explicitly selected `checkout_ref` skills. Ascend/CUDA
procedures use their corresponding backend; shared methods use `common`.
-Source snapshot: [78002cb43fdea1956876880ab251f9f1cd9a367f](https://github.com/InferMatrixLab/InferRSI/tree/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library).
+Source snapshot: [ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90](https://github.com/InferMatrixLab/InferRSI/tree/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library).
```bash
inferrsi knowledge search 'profiling' --repo-type inferrsi --limit 50
@@ -29,7 +29,7 @@ contain the current input paths and bytes (Git checkout filters are respected).
```bash
python tools/build_skill_index.py --source-revision
-python tools/build_skill_index.py --source-revision 78002cb43fdea1956876880ab251f9f1cd9a367f --check
+python tools/build_skill_index.py --source-revision ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90 --check
```
`--check` detects missing, stale and unexpected generated records without writing.
@@ -101,12 +101,14 @@ References and cases ship with their owning skill, not as duplicate knowledge ab
| [performance-ceiling-analysis](common/skills/performance-ceiling-analysis.json) | common | candidate |
| [plan-activation-residency](common/skills/plan-activation-residency.json) | common | candidate |
| [project-maya](common/skills/project-maya.json) | common | candidate |
+| [pytorch](common/skills/pytorch.json) | common | candidate |
| [reason-hopper-kernels](cuda/skills/reason-hopper-kernels.json) | cuda | candidate |
| [repository-evolution-analysis](common/skills/repository-evolution-analysis.json) | common | candidate |
| [rsi-evidence](common/skills/rsi-evidence.json) | common | candidate |
| [skill-evolution-playbook](common/skills/skill-evolution-playbook.json) | common | candidate |
| [splash](common/skills/splash.json) | common | candidate |
| [storage-io-measurement](common/skills/storage-io-measurement.json) | common | candidate |
+| [torch-npu](ascend/skills/torch-npu.json) | ascend | candidate |
| [torchnpu-device-memory-lifecycle](ascend/skills/torchnpu-device-memory-lifecycle.json) | ascend | candidate |
| [torchnpu-pinned-host-memory-lifecycle](ascend/skills/torchnpu-pinned-host-memory-lifecycle.json) | ascend | candidate |
| [training-input-pipeline](common/skills/training-input-pipeline.json) | common | candidate |
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-aclgraph-capture-replay.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-aclgraph-capture-replay.json
index 391ded6cd..f8103539f 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-aclgraph-capture-replay.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-aclgraph-capture-replay.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/hardware/ascend/ascend-aclgraph-capture-replay/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/hardware/ascend/ascend-aclgraph-capture-replay/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-aclgraph-capture-replay/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-aclgraph-capture-replay/metadata.json",
@@ -41,7 +41,7 @@
"src/inferrsi/data/skill_library/hardware/ascend/ascend-aclgraph-capture-replay/SKILL.md": "d494fc7e539cbb7388189bd7b3f04bb5c4983fdd7a82484348aad1cee7f741f2",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-aclgraph-capture-replay/metadata.json": "71c68431663ed2336cccc985a354adab6b13b4cfd6123a18a5d8d53c28268509",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-aclgraph-capture-replay/references/source-notes.md": "6ffbfb68cc7e04923b4fbca2a73a7555b7c3b45a6f40819bb5f3213ef05cf88d",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "ascend-aclgraph-capture-replay",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-device-health.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-device-health.json
index eef063cf4..3f68d516f 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-device-health.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-device-health.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/hardware/ascend/ascend-device-health/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/hardware/ascend/ascend-device-health/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-device-health/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-device-health/metadata.json",
@@ -41,7 +41,7 @@
"source_files": {
"src/inferrsi/data/skill_library/hardware/ascend/ascend-device-health/SKILL.md": "03829039860ece35519f2e4c339a48b5014ec25e07a2c150a3efb6a3a4e35b4a",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-device-health/metadata.json": "80d80c15514919826fa72d6f6cf26b76db8accd85d21187fc239232ca454e7bf",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "ascend-device-health",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-distributed-parallel.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-distributed-parallel.json
index 527153054..cc7383994 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-distributed-parallel.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-distributed-parallel.json
@@ -27,8 +27,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/hardware/ascend/ascend-distributed-parallel/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/hardware/ascend/ascend-distributed-parallel/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-distributed-parallel/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-distributed-parallel/metadata.json",
@@ -53,7 +53,7 @@
"src/inferrsi/data/skill_library/hardware/ascend/ascend-distributed-parallel/references/hccl-LICENSE.md": "05205a02acebd671c78d3b876c1f6097b84cce89780f5d9bdc204bfbea92d0c1",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-distributed-parallel/references/hccl-allreduce-readset.json": "4a3d403986dfceaa28605f2f30b79545e1efdc27f35c88755402b75b4d13d7e2",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-distributed-parallel/references/hccl-allreduce.md": "4770bba0090ed5ead3d2fd3b2cce2d21ba708acae3e0b4d859cc503499da5b93",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "ascend-distributed-parallel",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-graph-mode.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-graph-mode.json
index 41d95e6f2..41b4aa8e4 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-graph-mode.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-graph-mode.json
@@ -26,8 +26,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/hardware/ascend/ascend-graph-mode/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/hardware/ascend/ascend-graph-mode/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-graph-mode/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-graph-mode/metadata.json",
@@ -52,7 +52,7 @@
"src/inferrsi/data/skill_library/hardware/ascend/ascend-graph-mode/references/torchair-LICENSE.md": "bc07d81fa332b6543463588a120518c57b5ce8582d3747de75d2f6b09b068c01",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-graph-mode/references/torchair-graph-path.md": "272424571ff439278c8f2675406eae7da9a6482c581973e24319739fa0e58348",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-graph-mode/references/torchair-graph-readset.json": "eb84cbd5c355a88ee013dc7e428b274162e0e13e85845b58526113b0d6a2b46b",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "ascend-graph-mode",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-model-adaptation.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-model-adaptation.json
index 8414cfc82..b872e64ab 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-model-adaptation.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-model-adaptation.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/hardware/ascend/ascend-model-adaptation/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/hardware/ascend/ascend-model-adaptation/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-model-adaptation/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-model-adaptation/metadata.json",
@@ -40,7 +40,7 @@
"source_files": {
"src/inferrsi/data/skill_library/hardware/ascend/ascend-model-adaptation/SKILL.md": "8fdf46fb2743b67a66bc069d22a45305a222debfb606777bc451b4b8f31d9cc6",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-model-adaptation/metadata.json": "be681a3206e5e183b4c9670c9d7784bcffce18c891449e5b15f94ca5f1324b9a",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "ascend-model-adaptation",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-profiling.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-profiling.json
index 603c07ca8..761cc93a9 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-profiling.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-profiling.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-profiling/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-profiling/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-profiling/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-profiling/metadata.json",
@@ -41,7 +41,7 @@
"source_files": {
"src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-profiling/SKILL.md": "401981cf7828edcab0c9ea882f7ecd83badfa3c2a95006e912cf99d67b3d6a12",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-profiling/metadata.json": "efd37bb23acb500dad4ec8ebe7ddfe5c3f196374052825f923ac85da8445f06b",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "ascend-operator-profiling",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-tuning.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-tuning.json
index 35921208e..5d00e5016 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-tuning.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-tuning.json
@@ -24,8 +24,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-tuning/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-tuning/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-tuning/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-tuning/metadata.json",
@@ -42,7 +42,7 @@
"source_files": {
"src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-tuning/SKILL.md": "38b151bc6b4e9734c6fa41efb7a30343559400c0c75bbdd26bd8a968b8ea4592",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-tuning/metadata.json": "04d2002fc8a0a7becf86b0b73daaf2da8716ab466d7f988da43d41a10f26680e",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "ascend-operator-tuning",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-optimization-casebook.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-optimization-casebook.json
index 654424c77..639dfd7d5 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-optimization-casebook.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-optimization-casebook.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/hardware/ascend/ascend-optimization-casebook/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/hardware/ascend/ascend-optimization-casebook/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-optimization-casebook/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-optimization-casebook/metadata.json",
@@ -40,7 +40,7 @@
"source_files": {
"src/inferrsi/data/skill_library/hardware/ascend/ascend-optimization-casebook/SKILL.md": "647f5e9a288f2c4e1617ba53df59d9d827df749eb4fe86182d21a9c41175caf8",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-optimization-casebook/metadata.json": "f425f7c1abbd616387273e8932c6bd02f4b1da012f099286ad4171e0f6037b22",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "ascend-optimization-casebook",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-performance-analysis.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-performance-analysis.json
index fc8237e50..9ddb517d3 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-performance-analysis.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-performance-analysis.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/hardware/ascend/ascend-performance-analysis/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/hardware/ascend/ascend-performance-analysis/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-performance-analysis/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-performance-analysis/metadata.json",
@@ -44,7 +44,7 @@
"src/inferrsi/data/skill_library/hardware/ascend/ascend-performance-analysis/metadata.json": "9e705625c8308013c287576b0247da6f6116d817a4a535cbecff80abd6ea35ee",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-performance-analysis/references/cannbot-CANN-2.0-LICENSE.md": "89b1c114df62cfd183004f38d18d07f186436c3b37203dd69c3e84becf9d2282",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-performance-analysis/references/cannbot-stage-evidence.md": "f56e4125325c9c1b9235a1d64fd732861c7686444492422daff44ba8dd648be4",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "ascend-performance-analysis",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-video-diffusion-optimization.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-video-diffusion-optimization.json
index 572a93c5a..5863ada55 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-video-diffusion-optimization.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-video-diffusion-optimization.json
@@ -26,8 +26,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/hardware/ascend/ascend-video-diffusion-optimization/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/hardware/ascend/ascend-video-diffusion-optimization/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-video-diffusion-optimization/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-video-diffusion-optimization/metadata.json",
@@ -46,7 +46,7 @@
"src/inferrsi/data/skill_library/hardware/ascend/ascend-video-diffusion-optimization/SKILL.md": "53c21e64f0dd47f9510faef1f2c01e1283899787ba7ebfe70cdabd8e57a76f00",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-video-diffusion-optimization/metadata.json": "8c2b364298330751fd1252516159b16c27e90354e2a724782d2851e68e43cdb3",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-video-diffusion-optimization/references/field-notes-video-diffusion.md": "dadac7b23462fdc325218c72ddf8a2ac553151761a5d3e45b854b53d2c8c6908",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "ascend-video-diffusion-optimization",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascendc-tiling-design.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascendc-tiling-design.json
index bbc49aec6..53eca492c 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascendc-tiling-design.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascendc-tiling-design.json
@@ -24,8 +24,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/hardware/ascend/ascendc-tiling-design/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/hardware/ascend/ascendc-tiling-design/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascendc-tiling-design/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascendc-tiling-design/metadata.json",
@@ -42,7 +42,7 @@
"source_files": {
"src/inferrsi/data/skill_library/hardware/ascend/ascendc-tiling-design/SKILL.md": "93aa1b775ced304cebfe6c4dec0503ddb2260ecb3cb01edc17ce329446c5ce2b",
"src/inferrsi/data/skill_library/hardware/ascend/ascendc-tiling-design/metadata.json": "f6d94f104cb70969024b2f4542078a240e200712c66279eca8b94f60bee1dd71",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "ascendc-tiling-design",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/cann-attribution-overlap.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/cann-attribution-overlap.json
index 939f791d3..92e1ceb1d 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/cann-attribution-overlap.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/cann-attribution-overlap.json
@@ -24,8 +24,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/hardware/ascend/cann-attribution-overlap/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/hardware/ascend/cann-attribution-overlap/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/cann-attribution-overlap/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/cann-attribution-overlap/metadata.json",
@@ -42,7 +42,7 @@
"source_files": {
"src/inferrsi/data/skill_library/hardware/ascend/cann-attribution-overlap/SKILL.md": "0e6f094dc554abea12e07def50f20b233b47127779a8aefad595b4ab1efd24c2",
"src/inferrsi/data/skill_library/hardware/ascend/cann-attribution-overlap/metadata.json": "acccc8eb86de0df6caf78a9b3c6db76a2a686adc4a5ba0029a4d62d50ff9a8f8",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "cann-attribution-overlap",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torch-npu.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torch-npu.json
index deb5a8d02..b7b8dc41e 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torch-npu.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torch-npu.json
@@ -20,8 +20,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/8115da15b9ef34978de500b08aa9f14ea790c35b/src/inferrsi/data/skill_library/frameworks/torch-npu/torch-npu/SKILL.md",
- "commit": "8115da15b9ef34978de500b08aa9f14ea790c35b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/frameworks/torch-npu/torch-npu/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/frameworks/torch-npu/torch-npu/SKILL.md",
"src/inferrsi/data/skill_library/frameworks/torch-npu/torch-npu/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-device-memory-lifecycle.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-device-memory-lifecycle.json
index 2c44b362d..9ac5c3784 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-device-memory-lifecycle.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-device-memory-lifecycle.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/hardware/ascend/torchnpu-device-memory-lifecycle/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/hardware/ascend/torchnpu-device-memory-lifecycle/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/torchnpu-device-memory-lifecycle/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/torchnpu-device-memory-lifecycle/metadata.json",
@@ -42,7 +42,7 @@
"src/inferrsi/data/skill_library/hardware/ascend/torchnpu-device-memory-lifecycle/SKILL.md": "8428df7157af024fa17cfe0b3818ed54bd9f468306dda46eb8d4f0edf4f3f230",
"src/inferrsi/data/skill_library/hardware/ascend/torchnpu-device-memory-lifecycle/metadata.json": "c650dce88c07b636d68e063dbda4d319f6f7f3ecbcdd5e47f393cbfc3af7c71d",
"src/inferrsi/data/skill_library/hardware/ascend/torchnpu-device-memory-lifecycle/references/source-notes.md": "80cba4d3a884e7ee2618326cf74b9f9c3a34e4d9cf443707f6bbe673f2d5ba65",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "torchnpu-device-memory-lifecycle",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-pinned-host-memory-lifecycle.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-pinned-host-memory-lifecycle.json
index f574ae8ab..5f4f82ec1 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-pinned-host-memory-lifecycle.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-pinned-host-memory-lifecycle.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/hardware/ascend/torchnpu-pinned-host-memory-lifecycle/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/hardware/ascend/torchnpu-pinned-host-memory-lifecycle/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/torchnpu-pinned-host-memory-lifecycle/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/torchnpu-pinned-host-memory-lifecycle/metadata.json",
@@ -42,7 +42,7 @@
"src/inferrsi/data/skill_library/hardware/ascend/torchnpu-pinned-host-memory-lifecycle/SKILL.md": "23e93ea7c98f99743509fdc04139d78ed2e7123c63e80b366c6beaba7eb70d0d",
"src/inferrsi/data/skill_library/hardware/ascend/torchnpu-pinned-host-memory-lifecycle/metadata.json": "97836a38465106d816a9ace894a7b277d6d330f9accade8f8493ce6f0fb57ef9",
"src/inferrsi/data/skill_library/hardware/ascend/torchnpu-pinned-host-memory-lifecycle/references/source-notes.md": "435193590b4a2ba32e2e4c6fa9d7e8f889528d412cbbb48e35ec5433f7dab384",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "torchnpu-pinned-host-memory-lifecycle",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-serving-preemption.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-serving-preemption.json
index 91d99b710..6469ebff5 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-serving-preemption.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-serving-preemption.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/methods/assess-serving-preemption/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/methods/assess-serving-preemption/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/methods/assess-serving-preemption/SKILL.md",
"src/inferrsi/data/skill_library/methods/assess-serving-preemption/metadata.json",
@@ -36,7 +36,7 @@
"src/inferrsi/data/skill_library/methods/assess-serving-preemption/SKILL.md": "0a5f48d9352c5bf460775ef47f9136829978811f4a16e84983e299782cae305d",
"src/inferrsi/data/skill_library/methods/assess-serving-preemption/metadata.json": "3e2cea37f5669e7c60f7c01c390a48ea960b15cbd6de8f5340cf8fe2afb9e1a1",
"src/inferrsi/data/skill_library/methods/assess-serving-preemption/references/source-notes.md": "f29fcecfba0490008a354c11b42403983fe165e56ff907ad61cc6040d2b6407f",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "assess-serving-preemption",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-split-backward-schedule.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-split-backward-schedule.json
index 3f809f979..816d251e3 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-split-backward-schedule.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-split-backward-schedule.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/methods/assess-split-backward-schedule/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/methods/assess-split-backward-schedule/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/methods/assess-split-backward-schedule/SKILL.md",
"src/inferrsi/data/skill_library/methods/assess-split-backward-schedule/metadata.json",
@@ -36,7 +36,7 @@
"src/inferrsi/data/skill_library/methods/assess-split-backward-schedule/SKILL.md": "1e64f28ea71672fb53c56f6108b87864738c6d03d6dcd32eedaef9e0625e96e3",
"src/inferrsi/data/skill_library/methods/assess-split-backward-schedule/metadata.json": "a089dd2c2e5a4b41e970e426c63b5b8f65acc77d2e5a40a44b38543b5d730a00",
"src/inferrsi/data/skill_library/methods/assess-split-backward-schedule/references/source-notes.md": "61b1609f71c37b61d136b8c4d39f793de87cffb1cf31c8ec737a1c3fa6fb1b4e",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "assess-split-backward-schedule",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-fixed-workload-claim.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-fixed-workload-claim.json
index 46bd61d02..d719fe7fa 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-fixed-workload-claim.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-fixed-workload-claim.json
@@ -20,8 +20,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/methods/audit-fixed-workload-claim/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/methods/audit-fixed-workload-claim/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/methods/audit-fixed-workload-claim/SKILL.md",
"src/inferrsi/data/skill_library/methods/audit-fixed-workload-claim/metadata.json",
@@ -38,7 +38,7 @@
"source_files": {
"src/inferrsi/data/skill_library/methods/audit-fixed-workload-claim/SKILL.md": "2be89b1340a3deb978c8e1792d18aeecbe8f5de81d4daf542389f53826b8ebee",
"src/inferrsi/data/skill_library/methods/audit-fixed-workload-claim/metadata.json": "d36e8a6148b7ab830f1ddf6e573a865965502709a835ce460c37e3d4c386de95",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "audit-fixed-workload-claim",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-metal-fasth3-semantics.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-metal-fasth3-semantics.json
index 1941bc103..04c533d97 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-metal-fasth3-semantics.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-metal-fasth3-semantics.json
@@ -29,8 +29,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/LICENSE",
"src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/SKILL.md",
@@ -65,7 +65,7 @@
"src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/references/source-notes.md": "f9430103644eacd9c55ace6f84d8587fc0bc063bfcd9cb60e6045b139a9fee05",
"src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/references/troubleshooting.md": "ed552d87f9fb9fb2c56de64df61f3fc6e2333cb049116cd0736f784c8a327612",
"src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/scripts/audit_fasth3.py": "05bfb9348cf88a76f4fba8bc8a600fe54246d83a98c1685bc4847ace28e2fbac",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "audit-h3-metal-fasth3-semantics",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-runtime-residency.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-runtime-residency.json
index 3cab95bbb..b9973b506 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-runtime-residency.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-runtime-residency.json
@@ -30,8 +30,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/LICENSE",
"src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/SKILL.md",
@@ -66,7 +66,7 @@
"src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/references/source-identity.json": "730f8dd23648a3f8f47b9726cc613d893a79d1817d515eec609a9b06d8a9ff90",
"src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/references/source-notes.md": "7304bf9369efc157dc7aa40686dd828cc14db45c49442e908b3d3cc4b329eb86",
"src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/references/validation.md": "0754873a758e048480938314894e25b504ae0c269549312e7f8e6020d83fe294",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "audit-h3-runtime-residency",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-llama-runtime-state.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-llama-runtime-state.json
index 3f3701e61..047037bcb 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-llama-runtime-state.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-llama-runtime-state.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/SKILL.md",
"src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/metadata.json",
@@ -43,7 +43,7 @@
"src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/SKILL.md": "95e81b6bc096a954fa73139ea5bc001d1efa6c99747161da62f012b494e5c013",
"src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/metadata.json": "22f1d01b48ea6e2b2ef2deeb20c1404ce8b586eb18442b5adb1390310b4b7503",
"src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/references/source-notes.md": "3d12a401e9decb55a2f4f3b2bfa0d360ef5110f63a0469dfe285654e78054190",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "audit-llama-runtime-state",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-recurrent-kv-finalization.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-recurrent-kv-finalization.json
index 1c2fa168b..1ec0620cf 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-recurrent-kv-finalization.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-recurrent-kv-finalization.json
@@ -26,8 +26,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/LICENSE",
"src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/SKILL.md",
@@ -52,7 +52,7 @@
"src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/references/NOTICE.md": "cca794656897ecdc37f983f7f23c4cd12306dbe3c9cd2f54fece6340753ae4cf",
"src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/references/source-identity.json": "5e52db0a0bda4673b9aac46a3fc035b824780def72921fee6796edb2ed9fca65",
"src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/references/source-notes.md": "0956f59cd9a1fac27da53228dd60c51a07b87639a43c09bfab296c4aeeceaab9",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "audit-recurrent-kv-finalization",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sd-cpp-residency.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sd-cpp-residency.json
index 4e73951f0..5d2b6cef5 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sd-cpp-residency.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sd-cpp-residency.json
@@ -25,8 +25,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/LICENSE",
"src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/SKILL.md",
@@ -55,7 +55,7 @@
"src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/references/source-identity.json": "af4830370b6eddc2d176a9ae14da4ab56a90b0fe048a01a89ee051df4365d8e9",
"src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/references/source-notes.md": "c493af8f6fd11ab7c9a37c9eec78cac31d57c04403a790035efbb4fff99f4669",
"src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/references/validation-boundaries.md": "3aaf79b36192bf8ff3e8f95ac90f2e769aa065dfac1fe70368824f3591849b8d",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "audit-sd-cpp-residency",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sglang-diffusion-lifetimes.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sglang-diffusion-lifetimes.json
index ba11f6422..984809c22 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sglang-diffusion-lifetimes.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sglang-diffusion-lifetimes.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/LICENSE",
"src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/SKILL.md",
@@ -45,7 +45,7 @@
"src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/SKILL.md": "b0a4ba4751087126143ef86983798deb0ed00e24b6ff06bf6ef76a9df6e274f2",
"src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/metadata.json": "e2cb137978960d29b3644967a29dab930c768d3453bce8aff2aad96fd90a56d9",
"src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/references/source-notes.md": "d0fcfd1177ae2f351041c5b2abba7f5121cec133006d9e4dca8f74ddb567332d",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "audit-sglang-diffusion-lifetimes",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-vllm-kv-lifetimes.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-vllm-kv-lifetimes.json
index af929ae7e..3199dbc0f 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-vllm-kv-lifetimes.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-vllm-kv-lifetimes.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/SKILL.md",
"src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/metadata.json",
@@ -41,7 +41,7 @@
"src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/SKILL.md": "4c3f27459df239f9f4fc971c559d4c9ddfbfebd70507a8a93be8eae59a30b69d",
"src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/metadata.json": "8ded6e60c562131616dc8cdcf0d7efc04aef1884885da0b30fb16f42d7433b7c",
"src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/references/source-notes.md": "3665b034594fc8cc767c408002f4e16f8803369242405027c7f37ce6c9ecebe9",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "audit-vllm-kv-lifetimes",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/bound-windowed-attention-replay.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/bound-windowed-attention-replay.json
index 439cf4f58..7d41bea5d 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/bound-windowed-attention-replay.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/bound-windowed-attention-replay.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/SKILL.md",
"src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/metadata.json",
@@ -45,7 +45,7 @@
"src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/metadata.json": "c68254b1a96a8834ae46c7a1d9035856fa4783315b219aa95e31461bd8ac58c9",
"src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/references/source-notes.md": "fd917bd5ac80bee9c9f4525f8e0300f6d67dc144e2a16632af42a67f3d53d098",
"src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/references/validation.md": "60e71e70c5519bcdf6c171eec2be2f342e8801bedd746d6005f11adc646dfaff",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "bound-windowed-attention-replay",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/budget-speculative-verification.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/budget-speculative-verification.json
index 06a71583a..85695fd40 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/budget-speculative-verification.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/budget-speculative-verification.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/SKILL.md",
"src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/metadata.json",
@@ -45,7 +45,7 @@
"src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/metadata.json": "1aa80a3109072c1ec211f38dbeb9e9cb9edb56aeac187946b43e65dd7502e46f",
"src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/references/source-notes.md": "6fa955022f16cd598f44fd6521901d1cb868b06cec545cabe2b82769a641b5d6",
"src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/references/validation.md": "673d93f6700052974b340fab25b3438e0e6344960569ac9abec7d7720f662288",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "budget-speculative-verification",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/calibrate-serving-simulation.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/calibrate-serving-simulation.json
index cb09587b8..07b837586 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/calibrate-serving-simulation.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/calibrate-serving-simulation.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/methods/calibrate-serving-simulation/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/methods/calibrate-serving-simulation/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/methods/calibrate-serving-simulation/SKILL.md",
"src/inferrsi/data/skill_library/methods/calibrate-serving-simulation/metadata.json",
@@ -36,7 +36,7 @@
"src/inferrsi/data/skill_library/methods/calibrate-serving-simulation/SKILL.md": "ac773c6f1c110a664dfca28a3ae9129b10c07614e60ca3640dea776220e7b352",
"src/inferrsi/data/skill_library/methods/calibrate-serving-simulation/metadata.json": "96771fb6fecf5fc7773bb7cf84038d8eaf436add1a0362dc26110c82037f8320",
"src/inferrsi/data/skill_library/methods/calibrate-serving-simulation/references/source-notes.md": "9696d3c0ef96b6611787446bd6f96d4d6255ca8d107105de8c1d9af83edb9fa5",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "calibrate-serving-simulation",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cgroup-resource-controls.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cgroup-resource-controls.json
index b26d1e4b5..8e05d49ee 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cgroup-resource-controls.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cgroup-resource-controls.json
@@ -25,8 +25,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/hardware/host/cgroup-resource-controls/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/hardware/host/cgroup-resource-controls/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/hardware/host/cgroup-resource-controls/SKILL.md",
"src/inferrsi/data/skill_library/hardware/host/cgroup-resource-controls/metadata.json",
@@ -43,7 +43,7 @@
"source_files": {
"src/inferrsi/data/skill_library/hardware/host/cgroup-resource-controls/SKILL.md": "fd1f5a7e6a55714730044155f5b38c1d700383d4e16c3e91963bbfbe1c773af5",
"src/inferrsi/data/skill_library/hardware/host/cgroup-resource-controls/metadata.json": "19e40763fedc20990651bb989c32a7693306c5c54af46c3d4be77fd4f87b6c03",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "cgroup-resource-controls",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/checkpoint-recovery-consistency.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/checkpoint-recovery-consistency.json
index dd85bc623..36d983000 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/checkpoint-recovery-consistency.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/checkpoint-recovery-consistency.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/methods/checkpoint-recovery-consistency/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/methods/checkpoint-recovery-consistency/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/methods/checkpoint-recovery-consistency/SKILL.md",
"src/inferrsi/data/skill_library/methods/checkpoint-recovery-consistency/metadata.json",
@@ -44,7 +44,7 @@
"src/inferrsi/data/skill_library/methods/checkpoint-recovery-consistency/metadata.json": "4185ff9f8b4c06d2a984753127780f29ceebf50d87b225a7697bfecccc2a4a60",
"src/inferrsi/data/skill_library/methods/checkpoint-recovery-consistency/references/dlrover-persistence-replicas.md": "055f4357f79026fcbcc766bbe77db610376ec353623f8898f5d14a30ee69e7bc",
"src/inferrsi/data/skill_library/methods/checkpoint-recovery-consistency/references/sources.md": "9425b1aab963a2c8d35f3db897bd15ee69699bf5b0d0c8425c3c8bba3e9a0233",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "checkpoint-recovery-consistency",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/choose-parallel-training-plan.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/choose-parallel-training-plan.json
index 7c2f13e5d..b1445e424 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/choose-parallel-training-plan.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/choose-parallel-training-plan.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/methods/choose-parallel-training-plan/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/methods/choose-parallel-training-plan/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/methods/choose-parallel-training-plan/SKILL.md",
"src/inferrsi/data/skill_library/methods/choose-parallel-training-plan/metadata.json",
@@ -36,7 +36,7 @@
"src/inferrsi/data/skill_library/methods/choose-parallel-training-plan/SKILL.md": "25d04aed9782b2a8ff1186828becdc7fce93c4418039ceaef9b8dcc5ec9bcd24",
"src/inferrsi/data/skill_library/methods/choose-parallel-training-plan/metadata.json": "4990d0a2f1a269111a680f45a5d91221af37e3f8b3ce514fea84e97878e767cd",
"src/inferrsi/data/skill_library/methods/choose-parallel-training-plan/references/source-notes.md": "9f9184d46b19312138a6e805e439e43dc0594a0cc6b46675255a582dc946ca5b",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "choose-parallel-training-plan",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cite-check-downstream-pr.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cite-check-downstream-pr.json
index 98628f65a..4a0e3343e 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cite-check-downstream-pr.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cite-check-downstream-pr.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/workflows/cite-check-downstream-pr/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/workflows/cite-check-downstream-pr/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/skill-index.json",
"src/inferrsi/data/skill_library/workflows/cite-check-downstream-pr/SKILL.md",
@@ -38,7 +38,7 @@
},
"checkout_ref": "checkout:cite-check-downstream-pr",
"source_files": {
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317",
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db",
"src/inferrsi/data/skill_library/workflows/cite-check-downstream-pr/SKILL.md": "485932cdf3b1fba5dc69be73e75d38f86a7b79442d26e82a296c963a7598d226",
"src/inferrsi/data/skill_library/workflows/cite-check-downstream-pr/metadata.json": "7a18f9dbb4894216802a44929b913990fc4bd8521cb60976d9435ee8f37e4265"
},
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/claim-contract.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/claim-contract.json
index 96e9330f7..b2d17c5e1 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/claim-contract.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/claim-contract.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/workflows/claim-contract/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/workflows/claim-contract/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/skill-index.json",
"src/inferrsi/data/skill_library/workflows/claim-contract/SKILL.md",
@@ -38,7 +38,7 @@
},
"checkout_ref": "checkout:claim-contract",
"source_files": {
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317",
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db",
"src/inferrsi/data/skill_library/workflows/claim-contract/SKILL.md": "3af1dbbe81d91cb05fc2ef15a2ae25d7592a08558e1a5535f49c52f33a488c55",
"src/inferrsi/data/skill_library/workflows/claim-contract/metadata.json": "25c0137cf347f75d7ff8f8305d9ff5f055c6881ddf4b708f923c3176de16c153"
},
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/comfy-ui.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/comfy-ui.json
index d40ae76ab..fb4006d68 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/comfy-ui.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/comfy-ui.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/repositories/comfy-ui/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/repositories/comfy-ui/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/repositories/comfy-ui/LICENSE",
"src/inferrsi/data/skill_library/repositories/comfy-ui/SKILL.md",
@@ -50,7 +50,7 @@
"src/inferrsi/data/skill_library/repositories/comfy-ui/references/model-residency.md": "014c68386e7cdfdb7b4f566836e8b2ff61bd5a5a506e7b4723f83fe80c5d4de7",
"src/inferrsi/data/skill_library/repositories/comfy-ui/references/repo-provenance.md": "ca996ef515af42a3ce304747fa3f12a31c00533deaace5ba2ce21aefa0a62e2e",
"src/inferrsi/data/skill_library/repositories/comfy-ui/references/validation-boundaries.md": "1913dedbeb558d872d6a0e015cc11d822910c3cb259a236db29fb60ad8a64263",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "comfy-ui",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/competitor-scan.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/competitor-scan.json
index 42280bcf2..ed41c713f 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/competitor-scan.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/competitor-scan.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/workflows/competitor-scan/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/workflows/competitor-scan/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/skill-index.json",
"src/inferrsi/data/skill_library/workflows/competitor-scan/SKILL.md",
@@ -37,7 +37,7 @@
},
"checkout_ref": "checkout:competitor-scan",
"source_files": {
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317",
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db",
"src/inferrsi/data/skill_library/workflows/competitor-scan/SKILL.md": "0ede58ace65a9324a538b4816add1684f230ba536d4364fe5a5dd2c0bbed3776",
"src/inferrsi/data/skill_library/workflows/competitor-scan/metadata.json": "74b387e5df7b40cce9fd1ceee7ea1a26c258a4ed091144f2b366ef971895a2e8"
},
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dcp-async-save-performance.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dcp-async-save-performance.json
index f7cc2da85..b0a6f8e10 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dcp-async-save-performance.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dcp-async-save-performance.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/SKILL.md",
"src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/metadata.json",
@@ -43,7 +43,7 @@
"src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/SKILL.md": "645139ee0c88ec865cad952e24cda35f4648d4b777b732c9b7d274b6c7830845",
"src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/metadata.json": "49764fa454403fb38891d82ee63de1e07a90d3ff6a3cd28550aee7ea34cbdcb0",
"src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/references/sources.md": "22afe4f4d77f441096cc657d55d763473bfcbea8fb531c8e493fe49ab84955f7",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "dcp-async-save-performance",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dense-performance-evidence.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dense-performance-evidence.json
index e6edc9c66..72bc12779 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dense-performance-evidence.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dense-performance-evidence.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/methods/dense-performance-evidence/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/methods/dense-performance-evidence/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/methods/dense-performance-evidence/SKILL.md",
"src/inferrsi/data/skill_library/methods/dense-performance-evidence/metadata.json",
@@ -39,7 +39,7 @@
"source_files": {
"src/inferrsi/data/skill_library/methods/dense-performance-evidence/SKILL.md": "53cc1867a13f54d1006d50517dea8848869ca50df3f0cd610b2424cc8d18c933",
"src/inferrsi/data/skill_library/methods/dense-performance-evidence/metadata.json": "041780985dcaf76809d29dc0d4e242359b17b6bcfaadab2d05674fa4699a5427",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "dense-performance-evidence",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diagnose-vllm-admission-pressure.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diagnose-vllm-admission-pressure.json
index b7bac3770..1a4341c68 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diagnose-vllm-admission-pressure.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diagnose-vllm-admission-pressure.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/SKILL.md",
"src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/metadata.json",
@@ -67,7 +67,7 @@
"src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/references/papers-20261015/references/loongserve-readset.md": "f1a383e7bdbbd5a5a472d7f6e1292c8a17d4910ff736ca5a6aa422e1bee74f8c",
"src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/references/papers-20261015/references/orca-readset.md": "a497ef23b7e60c0fbc15d95b7e07cd20547cfe56f37c24dfe6ed70d79ef7a3cf",
"src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/references/source-notes.md": "84033fd514773f10c0bddd9914fabb980497ba2b308f226a922c3e1888413827",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "diagnose-vllm-admission-pressure",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diffusers-invocation-state.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diffusers-invocation-state.json
index b88df15c5..f30022b8e 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diffusers-invocation-state.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diffusers-invocation-state.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/SKILL.md",
"src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/metadata.json",
@@ -46,7 +46,7 @@
"src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/references/ownership.md": "40688cc44183027efbf2fca8d3548c299cdf2e33fef54f76ab3549353eb9bd1b",
"src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/references/source-map.md": "10562a75683bd9bd07ec2e37a1cbd0f93f2534d6f85b55a566adb2d97485215f",
"src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/references/validation.md": "5a0b854b87d091d7d505b0bdd5c03330052507a851ab6edf35b93a6675f2d4f7",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "diffusers-invocation-state",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/draft-acceptance-issue.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/draft-acceptance-issue.json
index eca015bea..a5ace976e 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/draft-acceptance-issue.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/draft-acceptance-issue.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/workflows/draft-acceptance-issue/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/workflows/draft-acceptance-issue/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/skill-index.json",
"src/inferrsi/data/skill_library/workflows/draft-acceptance-issue/SKILL.md",
@@ -38,7 +38,7 @@
},
"checkout_ref": "checkout:draft-acceptance-issue",
"source_files": {
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317",
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db",
"src/inferrsi/data/skill_library/workflows/draft-acceptance-issue/SKILL.md": "ce983acfb7d7d3272775e7cf00c92e4ddadd97511b602575a76e5ab7ade01b56",
"src/inferrsi/data/skill_library/workflows/draft-acceptance-issue/metadata.json": "e97a665349f234d74baeaa20c6996bff990d0f45c40c0b7ac7283aa599cb951e"
},
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dwarfstar.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dwarfstar.json
index 7495c14b1..5e0da828f 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dwarfstar.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dwarfstar.json
@@ -25,8 +25,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/repositories/dwarfstar/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/repositories/dwarfstar/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/repositories/dwarfstar/LICENSE",
"src/inferrsi/data/skill_library/repositories/dwarfstar/SKILL.md",
@@ -71,7 +71,7 @@
"src/inferrsi/data/skill_library/repositories/dwarfstar/sub-skills/serving-state/SKILL.md": "71c27212847146376d72e6cc6d22abe2f7eed56402c5917cd7dfcda0b2aa5982",
"src/inferrsi/data/skill_library/repositories/dwarfstar/sub-skills/serving-state/references/state-contracts.md": "8f74a07c947021e3069468268f6b5aa2548a17db518c5c0505b84dde73a1f31c",
"src/inferrsi/data/skill_library/repositories/dwarfstar/sub-skills/serving-state/references/troubleshooting.md": "c9eebe4109d418c4dfb9bb2425c145ebc1a52da9b665e0556149bc31a442e2fc",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "dwarfstar",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/evaluate-elastic-serving-placement.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/evaluate-elastic-serving-placement.json
index 2b69cb795..0b6140435 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/evaluate-elastic-serving-placement.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/evaluate-elastic-serving-placement.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/methods/evaluate-elastic-serving-placement/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/methods/evaluate-elastic-serving-placement/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/methods/evaluate-elastic-serving-placement/SKILL.md",
"src/inferrsi/data/skill_library/methods/evaluate-elastic-serving-placement/metadata.json",
@@ -36,7 +36,7 @@
"src/inferrsi/data/skill_library/methods/evaluate-elastic-serving-placement/SKILL.md": "82d0099d3c33706f46c657d4fa7f76d3d486e019b36760d1de001921328a60cc",
"src/inferrsi/data/skill_library/methods/evaluate-elastic-serving-placement/metadata.json": "091a3b95d1a8996b5ad917e7948efb4ed598e31c7163709377f8e9199cebfe26",
"src/inferrsi/data/skill_library/methods/evaluate-elastic-serving-placement/references/source-notes.md": "ea416831fd7d0874e06d0a76fb0c7196df53072b717673a0cdf635db6bfd917b",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "evaluate-elastic-serving-placement",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/generation-cache-lifecycle.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/generation-cache-lifecycle.json
index 3f8d133e7..99de5ec47 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/generation-cache-lifecycle.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/generation-cache-lifecycle.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/SKILL.md",
"src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/metadata.json",
@@ -48,7 +48,7 @@
"src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/references/generation-batch.md": "f16d3c57d45e63bcaffd3871aa41c76972ee11f7dbf1e12b45d9b170303cad9b",
"src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/references/source-map.md": "91d26181f246307f2113f94b12ac9587d689b654dcd216e9f14b11cf5ebb650c",
"src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/references/validation.md": "73b5dd347017baba519ecc85415aead9993c1751d116b9ba4f07689f58c5f783",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "generation-cache-lifecycle",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-memory-management.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-memory-management.json
index 48329956a..a097c6a6a 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-memory-management.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-memory-management.json
@@ -30,8 +30,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/hardware/host/host-memory-management/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/hardware/host/host-memory-management/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/hardware/host/host-memory-management/SKILL.md",
"src/inferrsi/data/skill_library/hardware/host/host-memory-management/metadata.json",
@@ -66,7 +66,7 @@
"src/inferrsi/data/skill_library/hardware/host/host-memory-management/references/vllm-ascend-LICENSE.md": "c71d239df91726fc519c6eb72d318ec65820627232b2f796219e87dcf35d0ab4",
"src/inferrsi/data/skill_library/hardware/host/host-memory-management/references/vllm-ascend-native-offload-readset.json": "cc6c2011be746002ba3ff96ccefab1809863748ddd17eb102b75e7b0e134fe55",
"src/inferrsi/data/skill_library/hardware/host/host-memory-management/references/vllm-ascend-native-offload.md": "6c2a72b5ba3708a01d229e0576570837a617e6a42dbb0b5312a26a65a6c31542",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "host-memory-management",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-overhead-optimization.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-overhead-optimization.json
index 137cf6178..329ff605d 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-overhead-optimization.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-overhead-optimization.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/hardware/host/host-overhead-optimization/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/hardware/host/host-overhead-optimization/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/hardware/host/host-overhead-optimization/SKILL.md",
"src/inferrsi/data/skill_library/hardware/host/host-overhead-optimization/metadata.json",
@@ -42,7 +42,7 @@
"src/inferrsi/data/skill_library/hardware/host/host-overhead-optimization/SKILL.md": "36ef88826a32ac3147b869001004ded17a8d5a491e25bfe105005ba39ea9d1eb",
"src/inferrsi/data/skill_library/hardware/host/host-overhead-optimization/metadata.json": "ae71c389dfdb1a19d267feecb4c6f1fbb4c9afad501ff6d8fab7e38b29bcc37a",
"src/inferrsi/data/skill_library/hardware/host/host-overhead-optimization/references/deepinfer-zero-overhead-scheduling.md": "f452bf40132c9195d739f1c86b3aa18857138c7ed0f1861af5853d93b9955924",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "host-overhead-optimization",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/husky.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/husky.json
index fb9e22206..37e502f7f 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/husky.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/husky.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/repositories/husky/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/repositories/husky/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/repositories/husky/SKILL.md",
"src/inferrsi/data/skill_library/repositories/husky/metadata.json",
@@ -51,7 +51,7 @@
"src/inferrsi/data/skill_library/repositories/husky/references/troubleshooting.md": "3948711437d25822d389dd857c72dc7fcc15727c9048d2e3eae3317d97aca6dd",
"src/inferrsi/data/skill_library/repositories/husky/references/verification-protocol.md": "83c5a966ed3f721b412b8d8417af8336c5a8c1e46044a888720936bf2f1deeab",
"src/inferrsi/data/skill_library/repositories/husky/scripts/audit_evidence_card.py": "291e62701ef7a37a5bfab963ea4e20abb94245bc015584223a101f18b1490895",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "husky",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-benchmark-design.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-benchmark-design.json
index 00314c24f..e8b219314 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-benchmark-design.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-benchmark-design.json
@@ -24,7 +24,10 @@
"image-generation",
"video-generation",
"asr",
- "stream-boundaries"
+ "stream-boundaries",
+ "treatment-delivery",
+ "trial-validity",
+ "protocol-audit"
],
"origin": [
"external"
@@ -32,11 +35,13 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/methods/inference-benchmark-design/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/methods/inference-benchmark-design/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/methods/inference-benchmark-design/SKILL.md",
"src/inferrsi/data/skill_library/methods/inference-benchmark-design/metadata.json",
+ "src/inferrsi/data/skill_library/methods/inference-benchmark-design/references/autotunebench-source.json",
+ "src/inferrsi/data/skill_library/methods/inference-benchmark-design/references/autotunebench-trial-validity.md",
"src/inferrsi/data/skill_library/methods/inference-benchmark-design/references/benchmark-commands.md",
"src/inferrsi/data/skill_library/methods/inference-benchmark-design/references/cache-aware-asr-stream-boundaries.md",
"src/inferrsi/data/skill_library/methods/inference-benchmark-design/references/justgoal-evaluation-contracts.md",
@@ -44,7 +49,7 @@
"src/inferrsi/data/skill_library/methods/inference-benchmark-design/references/task-benchmark-selection.md",
"src/inferrsi/data/skill_library/skill-index.json"
],
- "content_sha256": "a977e81fb83e3e476f66dc97133f1a60a35983bdfaec538ad5063defda41d4b7"
+ "content_sha256": "7fd7f557f5c99ca4ec4c46f2d97e67ecf280f5f0fdf071871a14cc08a7cb1c84"
}
],
"index_type": "packaged-skill",
@@ -53,14 +58,16 @@
},
"checkout_ref": "checkout:inference-benchmark-design",
"source_files": {
- "src/inferrsi/data/skill_library/methods/inference-benchmark-design/SKILL.md": "a977e81fb83e3e476f66dc97133f1a60a35983bdfaec538ad5063defda41d4b7",
- "src/inferrsi/data/skill_library/methods/inference-benchmark-design/metadata.json": "3570ee36df8037c97c4692afe39316b62c0c91c28b2ea29ab9b72c5ae6740bb6",
+ "src/inferrsi/data/skill_library/methods/inference-benchmark-design/SKILL.md": "7fd7f557f5c99ca4ec4c46f2d97e67ecf280f5f0fdf071871a14cc08a7cb1c84",
+ "src/inferrsi/data/skill_library/methods/inference-benchmark-design/metadata.json": "0880a9915d931a1b232d2250ff32790407803adf65f08020573d6d2cc6a18ff8",
+ "src/inferrsi/data/skill_library/methods/inference-benchmark-design/references/autotunebench-source.json": "bc343ece2bf40c4e8aa4bd8d84f32465d00633255557feda5fe7799ce24e8688",
+ "src/inferrsi/data/skill_library/methods/inference-benchmark-design/references/autotunebench-trial-validity.md": "cb3a7a5d94d056f868af07f92a16a6fef31eafd3199ed07ae79e93798e320b1d",
"src/inferrsi/data/skill_library/methods/inference-benchmark-design/references/benchmark-commands.md": "0898424e0bff3a6246e1eeeda6abd6c5bd974c047fe32f3bbd5ab9d31ce17cd3",
"src/inferrsi/data/skill_library/methods/inference-benchmark-design/references/cache-aware-asr-stream-boundaries.md": "4e0ea9c1bf6b096a49a91ad5266e90b8a456f4f5aca15eba8ac270a53c2309b5",
"src/inferrsi/data/skill_library/methods/inference-benchmark-design/references/justgoal-evaluation-contracts.md": "d25df795fd301e8a1bb436e5f8122d0bac2441c43104e59a3b310fd5a64f2f3e",
"src/inferrsi/data/skill_library/methods/inference-benchmark-design/references/justgoal-source.json": "60f2784384ea1a3b08a2ecafb60207d8e3707247dc7bc177b95b463d81f1cc16",
"src/inferrsi/data/skill_library/methods/inference-benchmark-design/references/task-benchmark-selection.md": "0d4bb080a754e26846e81680023db0dff5e6a4c96be4327c411c3b72ac4ef8e2",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "inference-benchmark-design",
@@ -80,7 +87,10 @@
"image-generation",
"video-generation",
"asr",
- "stream-boundaries"
+ "stream-boundaries",
+ "treatment-delivery",
+ "trial-validity",
+ "protocol-audit"
],
"status": "candidate",
"profile": "local-skill-v1",
@@ -256,14 +266,58 @@
"readset": "references/justgoal-source.json",
"license": "UNKNOWN",
"scope": "Complete 21-page paper and appendices: controlled comparison, independent correctness, repair/coverage and timing evidence; author-reported, no PDF/figure redistribution"
+ },
+ {
+ "type": "git",
+ "url": "https://github.com/li-ch/autotunebench",
+ "commit": "64cc8d7ca5d0ee40c4e2189b2c2c8dfb1a2a7c2f",
+ "paths": [
+ "ARTIFACT.md",
+ "README.md",
+ "CITATION.cff",
+ "LICENSE",
+ "config/protocol.yaml",
+ "pdb/ingest_validate.py",
+ "pdb/store.py",
+ "agents/loop.py",
+ "agents/adapter.py",
+ "orchestrator/cli.py",
+ "gates/g2_runtime.py",
+ "gates/g3_correctness.py",
+ "gates/g5_graduation.py",
+ "gates/schema.py",
+ "harness/kernel_bench.py",
+ "figs/plot_cand_ab_saturation.py",
+ "scripts/kernelbench_tabulate.py",
+ "experiments/kernelbench_l1/results/README.md",
+ "experiments/kernelbench_l1/results/TABLE.md",
+ "experiments/astra_parity/README.md",
+ "observability/iterations.jsonl",
+ "experiments/kernelbench_l1/kernelbench_tasks/UPSTREAM_LICENSE",
+ "experiments/kernelbench_l1/kernelbench_tasks/README.md"
+ ],
+ "readset": "references/autotunebench-trial-validity.md"
+ },
+ {
+ "type": "paper",
+ "id": "arxiv:2609.18123",
+ "version": "v1",
+ "url": "https://arxiv.org/abs/2609.18123v1",
+ "sections": [
+ "1-8",
+ "References",
+ "Figures 1-6",
+ "Tables 1-7"
+ ],
+ "readset": "references/autotunebench-trial-validity.md"
}
],
"empirical_validation": "not_run",
"applicability": {
- "scope": "Selecting task quality gates, reference parity checks and performance workloads for text, agentic coding, work agents, cache-aware ASR, TTS, full-duplex audio/video, image and video inference, and designing goodput-based multi-regime comparisons that fit InferRSI's evidence trail",
+ "scope": "Selecting task quality gates, reference parity checks and performance workloads for text, agentic coding, work agents, cache-aware ASR, TTS, full-duplex audio/video, image and video inference, and designing goodput-based multi-regime comparisons that fit InferRSI's evidence trail; interpreting treatment delivery, arm validity and stated-versus-observed protocol enforcement in supplied tuning reports",
"prerequisites": "The primary objective, timing boundary, quality gates, sample/failure accounting, required regimes, repetitions, seeds and decision rule are frozen before optimization; baseline and candidate share hardware, software stack, input distribution, benchmark configuration, warmup and measurement protocol, and on a shared machine also one time window and dedicated devices; cache-aware ASR also requires the selected model/config and first/steady geometry, decoded clock and reset/EOU/EOF boundaries to be established",
- "input_contract": "The task family and deployment use case; candidate quality evaluators and reference parity; the declared SLO and the caller's policy on which regimes must improve or must not regress; for an audit, a supplied fixed-workload report with request-to-trial grouping, exclusion and failure counts and the declared aggregation rule",
- "produced_artifact": "A frozen benchmark design with evaluators, reference parity, primary objective, timing boundary, quality gates, required regimes, repetition and seed rules and an explicit decision config, plus the evidence-tree layout a PASS claim must produce; for a supplied report, the computed threshold result and missing evidence. Candidate knowledge, not qualification of any model or backend",
+ "input_contract": "The task family and deployment use case; candidate quality evaluators and reference parity; the declared SLO and the caller's policy on which regimes must improve or must not regress; for an audit, a supplied fixed-workload report with request-to-trial grouping, exclusion and failure counts and the declared aggregation rule; for a treatment audit, the frozen delivery/arm-validity rules, run/candidate identities, feedback and consumer-input evidence, and per-arm raw records",
+ "produced_artifact": "A frozen benchmark design with evaluators, reference parity, primary objective, timing boundary, quality gates, required regimes, repetition and seed rules and an explicit decision config, plus the evidence-tree layout a PASS claim must produce; for a supplied report, the computed threshold result and missing evidence. Candidate knowledge, not qualification of any model or backend; for a treatment audit, a contract-bound eligibility and interpretation report separating treatment exposure, observed checks, comparator identity, anomaly disposition and source conflicts",
"failure_modes": [
"Raw throughput is optimized in place of goodput, so requests served outside the SLO count toward a win",
"One concurrency level or one undisclosed aggregate score stands in for all required regimes",
@@ -272,15 +326,19 @@
"Runs on the same hardware class, in different time windows or on shared devices are compared as if pinned",
"A quality pass is taken to certify that the timed work actually occurred",
"Best-seen results are reported as delivered without restarting and validating the final code/config artifact",
- "A cache-aware ASR EOU is treated as EOF or full reset, nominal chunk duration as the decoded mel frontier, or shape-only geometry PASS as a native model-state test"
+ "A cache-aware ASR EOU is treated as EOF or full reset, nominal chunk duration as the decoded mel frontier, or shape-only geometry PASS as a native model-state test",
+ "An enabled treatment or stored feedback summary is taken as proof of delivery to the next eligible consumer",
+ "A failed arm-validity condition is overridden by a favorable best timing or a saturation/null interpretation",
+ "Complete gate keys or zero recorded failures are read as proof that every declared protocol check ran"
],
- "refusal_point": "Records missing coverage and high-variance results as INCONCLUSIVE rather than a win; a run without a frozen RunSpec, or whose validator changed after the plan was frozen, is INCONCLUSIVE regardless of metric values; sets no universal numeric gate or noise threshold"
+ "refusal_point": "Records missing coverage and high-variance results as INCONCLUSIVE rather than a win; a run without a frozen RunSpec, or whose validator changed after the plan was frozen, is INCONCLUSIVE regardless of metric values; sets no universal numeric gate or noise threshold; withholds causal treatment benefit/null conclusions when required delivery or arm eligibility is unverified, and preserves conflicts between stated protocol, exported code and source reports"
},
"recommendations": [
"Freeze one primary objective, timing boundary, quality gates and sample/failure accounting before optimization.",
"Choose the required regimes from the deployment use case, report each separately, and record which must improve or must not regress in a versioned decision config.",
"Use the replay trace that matches the actual deployment pattern.",
- "State the concurrency, trace, model, SKU and microbenchmark scope boundaries in any report that cites benchmark results."
+ "State the concurrency, trace, model, SKU and microbenchmark scope boundaries in any report that cites benchmark results.",
+ "Trace treatment to its eligible consumer and apply the frozen arm-validity precedence before attributing benefit, saturation or a null."
],
"limitations": [
"Source-derived candidate; no local efficacy validation.",
@@ -288,14 +346,17 @@
"Task benchmark knowledge is distilled from issue #60 with selected pinned-source rechecks; no all-task runtime qualification or optimization benefit is established.",
"Benchmark feasibility and package-consumption checks do not promote this candidate or replace task quality, repeated A/B and service qualification.",
"The pinned NeMo-Speech.cpp ASR reference covers selected source/test definitions and author-reported benchmark boundaries; native model/device, decoder internals, quality, long-audio/load and service behavior remain not_run.",
- "JustGoal contributes source-only controlled-evaluation and kernel-acceptance guidance; its repaired/retested timings, released-pass counts and protocol differences are author-reported, not locally reproduced or a complete campaign runner."
+ "JustGoal contributes source-only controlled-evaluation and kernel-acceptance guidance; its repaired/retested timings, released-pass counts and protocol differences are author-reported, not locally reproduced or a complete campaign runner.",
+ "The AutoTuneBench reference is source-only: original operational checkpoint/prompt receipts remain unavailable, source contradictions remain explicit, and no tuning, device/performance or agent-efficacy validation was run."
],
"provenance_readsets": [
"references/task-benchmark-selection.md",
"references/benchmark-commands.md",
"references/cache-aware-asr-stream-boundaries.md",
"references/justgoal-evaluation-contracts.md",
- "references/justgoal-source.json"
+ "references/justgoal-source.json",
+ "references/autotunebench-trial-validity.md",
+ "references/autotunebench-source.json"
],
"references": [
"../inference-optimization-casebook/cases/admission-drain-occupancy.md",
@@ -304,10 +365,10 @@
"consumption_mode": "packaged"
},
"applicability": {
- "scope": "Selecting task quality gates, reference parity checks and performance workloads for text, agentic coding, work agents, cache-aware ASR, TTS, full-duplex audio/video, image and video inference, and designing goodput-based multi-regime comparisons that fit InferRSI's evidence trail",
+ "scope": "Selecting task quality gates, reference parity checks and performance workloads for text, agentic coding, work agents, cache-aware ASR, TTS, full-duplex audio/video, image and video inference, and designing goodput-based multi-regime comparisons that fit InferRSI's evidence trail; interpreting treatment delivery, arm validity and stated-versus-observed protocol enforcement in supplied tuning reports",
"prerequisites": "The primary objective, timing boundary, quality gates, sample/failure accounting, required regimes, repetitions, seeds and decision rule are frozen before optimization; baseline and candidate share hardware, software stack, input distribution, benchmark configuration, warmup and measurement protocol, and on a shared machine also one time window and dedicated devices; cache-aware ASR also requires the selected model/config and first/steady geometry, decoded clock and reset/EOU/EOF boundaries to be established",
- "input_contract": "The task family and deployment use case; candidate quality evaluators and reference parity; the declared SLO and the caller's policy on which regimes must improve or must not regress; for an audit, a supplied fixed-workload report with request-to-trial grouping, exclusion and failure counts and the declared aggregation rule",
- "produced_artifact": "A frozen benchmark design with evaluators, reference parity, primary objective, timing boundary, quality gates, required regimes, repetition and seed rules and an explicit decision config, plus the evidence-tree layout a PASS claim must produce; for a supplied report, the computed threshold result and missing evidence. Candidate knowledge, not qualification of any model or backend",
+ "input_contract": "The task family and deployment use case; candidate quality evaluators and reference parity; the declared SLO and the caller's policy on which regimes must improve or must not regress; for an audit, a supplied fixed-workload report with request-to-trial grouping, exclusion and failure counts and the declared aggregation rule; for a treatment audit, the frozen delivery/arm-validity rules, run/candidate identities, feedback and consumer-input evidence, and per-arm raw records",
+ "produced_artifact": "A frozen benchmark design with evaluators, reference parity, primary objective, timing boundary, quality gates, required regimes, repetition and seed rules and an explicit decision config, plus the evidence-tree layout a PASS claim must produce; for a supplied report, the computed threshold result and missing evidence. Candidate knowledge, not qualification of any model or backend; for a treatment audit, a contract-bound eligibility and interpretation report separating treatment exposure, observed checks, comparator identity, anomaly disposition and source conflicts",
"failure_modes": [
"Raw throughput is optimized in place of goodput, so requests served outside the SLO count toward a win",
"One concurrency level or one undisclosed aggregate score stands in for all required regimes",
@@ -316,15 +377,19 @@
"Runs on the same hardware class, in different time windows or on shared devices are compared as if pinned",
"A quality pass is taken to certify that the timed work actually occurred",
"Best-seen results are reported as delivered without restarting and validating the final code/config artifact",
- "A cache-aware ASR EOU is treated as EOF or full reset, nominal chunk duration as the decoded mel frontier, or shape-only geometry PASS as a native model-state test"
+ "A cache-aware ASR EOU is treated as EOF or full reset, nominal chunk duration as the decoded mel frontier, or shape-only geometry PASS as a native model-state test",
+ "An enabled treatment or stored feedback summary is taken as proof of delivery to the next eligible consumer",
+ "A failed arm-validity condition is overridden by a favorable best timing or a saturation/null interpretation",
+ "Complete gate keys or zero recorded failures are read as proof that every declared protocol check ran"
],
- "refusal_point": "Records missing coverage and high-variance results as INCONCLUSIVE rather than a win; a run without a frozen RunSpec, or whose validator changed after the plan was frozen, is INCONCLUSIVE regardless of metric values; sets no universal numeric gate or noise threshold"
+ "refusal_point": "Records missing coverage and high-variance results as INCONCLUSIVE rather than a win; a run without a frozen RunSpec, or whose validator changed after the plan was frozen, is INCONCLUSIVE regardless of metric values; sets no universal numeric gate or noise threshold; withholds causal treatment benefit/null conclusions when required delivery or arm eligibility is unverified, and preserves conflicts between stated protocol, exported code and source reports"
},
"recommendations": [
"Freeze one primary objective, timing boundary, quality gates and sample/failure accounting before optimization.",
"Choose the required regimes from the deployment use case, report each separately, and record which must improve or must not regress in a versioned decision config.",
"Use the replay trace that matches the actual deployment pattern.",
- "State the concurrency, trace, model, SKU and microbenchmark scope boundaries in any report that cites benchmark results."
+ "State the concurrency, trace, model, SKU and microbenchmark scope boundaries in any report that cites benchmark results.",
+ "Trace treatment to its eligible consumer and apply the frozen arm-validity precedence before attributing benefit, saturation or a null."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
@@ -336,7 +401,8 @@
"Task benchmark knowledge is distilled from issue #60 with selected pinned-source rechecks; no all-task runtime qualification or optimization benefit is established.",
"Benchmark feasibility and package-consumption checks do not promote this candidate or replace task quality, repeated A/B and service qualification.",
"The pinned NeMo-Speech.cpp ASR reference covers selected source/test definitions and author-reported benchmark boundaries; native model/device, decoder internals, quality, long-audio/load and service behavior remain not_run.",
- "JustGoal contributes source-only controlled-evaluation and kernel-acceptance guidance; its repaired/retested timings, released-pass counts and protocol differences are author-reported, not locally reproduced or a complete campaign runner."
+ "JustGoal contributes source-only controlled-evaluation and kernel-acceptance guidance; its repaired/retested timings, released-pass counts and protocol differences are author-reported, not locally reproduced or a complete campaign runner.",
+ "The AutoTuneBench reference is source-only: original operational checkpoint/prompt receipts remain unavailable, source contradictions remain explicit, and no tuning, device/performance or agent-efficacy validation was run."
],
"verification": {
"locally_verified": false,
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-optimization-casebook.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-optimization-casebook.json
index 4cf356416..08b281116 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-optimization-casebook.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-optimization-casebook.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/methods/inference-optimization-casebook/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/methods/inference-optimization-casebook/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/methods/inference-optimization-casebook/SKILL.md",
"src/inferrsi/data/skill_library/methods/inference-optimization-casebook/cases/admission-drain-occupancy.json",
@@ -55,7 +55,7 @@
"src/inferrsi/data/skill_library/methods/inference-optimization-casebook/cases/qwen-b300-rejected-directions.json": "146ea1564a46ed03092c7dc5796b318fc05b04b1821105ee00eb7fa7fa1333e2",
"src/inferrsi/data/skill_library/methods/inference-optimization-casebook/cases/qwen-b300-rejected-directions.md": "cce2930f544a0f78f21797fba4d8ded298e4a81f76c34d7977b0e31d361bc738",
"src/inferrsi/data/skill_library/methods/inference-optimization-casebook/metadata.json": "963488658f327d2379183d70357eabf832d70a6a70db2aa67f548493f4088d97",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "inference-optimization-casebook",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kernel-writing-guide.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kernel-writing-guide.json
index ad463bd9b..301c5980e 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kernel-writing-guide.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kernel-writing-guide.json
@@ -39,8 +39,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/methods/kernel-writing-guide/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/methods/kernel-writing-guide/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/methods/kernel-writing-guide/SKILL.md",
"src/inferrsi/data/skill_library/methods/kernel-writing-guide/cases/causal-conv1d-dispatch-state.md",
@@ -89,7 +89,7 @@
"src/inferrsi/data/skill_library/methods/kernel-writing-guide/references/sglang-gdn-causal-conv-contract.md": "c9f27b0231ddec3d14d6fa0831d1494ffd8e11bd8ad704e04be3bc4a4ffc0e58",
"src/inferrsi/data/skill_library/methods/kernel-writing-guide/references/sglang-gdn-source-files.json": "4c7ec2cfd2c29297e185ce09b4d9b2040fbe3635943c112f9ebb71dca8ab381b",
"src/inferrsi/data/skill_library/methods/kernel-writing-guide/references/vllm-gdn-causal-conv-contract.md": "eec09a6e95983c46b67d5f066d34419c17819899f5d8e3c65f2f163f2ada0fde",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "kernel-writing-guide",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kv-tier-management.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kv-tier-management.json
index 3e8c398cf..3abfdb434 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kv-tier-management.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kv-tier-management.json
@@ -24,8 +24,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/repositories/lmcache/kv-tier-management/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/repositories/lmcache/kv-tier-management/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/repositories/lmcache/kv-tier-management/SKILL.md",
"src/inferrsi/data/skill_library/repositories/lmcache/kv-tier-management/metadata.json",
@@ -42,7 +42,7 @@
"source_files": {
"src/inferrsi/data/skill_library/repositories/lmcache/kv-tier-management/SKILL.md": "e4b78f368d70c8abb74a935acb910414015f1aeec1b50f7a295e836822ec4b27",
"src/inferrsi/data/skill_library/repositories/lmcache/kv-tier-management/metadata.json": "85f7242b83334af11f57799875738b498bfe4680a51f36727792dfbe91806704",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "kv-tier-management",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lazy-evaluation-stream-lifetimes.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lazy-evaluation-stream-lifetimes.json
index 4c97a5bb2..24f59676b 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lazy-evaluation-stream-lifetimes.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lazy-evaluation-stream-lifetimes.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/SKILL.md",
"src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/metadata.json",
@@ -46,7 +46,7 @@
"src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/references/execution-lifetime.md": "c4bc41e08aca68f661325f61e6f0db1760a079dc362f295c847bdc374fee3bf6",
"src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/references/source-map.md": "798b9cb89ed6b885eb47464ca9f137d2802a54571b9327e631d7742d3e5abd5d",
"src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/references/validation.md": "ebfc00bf7d6ab0a1dc415953ae84d56ccdc713d5614aff85e433c4a843793bff",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "lazy-evaluation-stream-lifetimes",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lily.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lily.json
index e7875efa9..3f384c085 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lily.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lily.json
@@ -25,8 +25,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/repositories/lily/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/repositories/lily/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/repositories/lily/LICENSE",
"src/inferrsi/data/skill_library/repositories/lily/NOTICE.md",
@@ -77,7 +77,7 @@
"src/inferrsi/data/skill_library/repositories/lily/sub-skills/serving-cache/references/api-and-session.md": "7262bb077c24f2f7549d450473ec7944829940d09cca89db9c12f37ac90d5db0",
"src/inferrsi/data/skill_library/repositories/lily/sub-skills/serving-cache/references/troubleshooting.md": "6e2f36d2f7ab9eae665d87a676e34c0ebd2f46f3df6fa5f2e589a306749a87b5",
"src/inferrsi/data/skill_library/repositories/lily/sub-skills/serving-cache/scripts/request_template.py": "d6b00c9f0b38c7fb0a0e52e396993a832542ea0aa6e2456026fabc0c39a4d33a",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "lily",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lithos-metal.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lithos-metal.json
index d3aa5fa09..bdf86884d 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lithos-metal.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lithos-metal.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/repositories/lithos-metal/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/repositories/lithos-metal/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/repositories/lithos-metal/LICENSE",
"src/inferrsi/data/skill_library/repositories/lithos-metal/SKILL.md",
@@ -81,7 +81,7 @@
"src/inferrsi/data/skill_library/repositories/lithos-metal/sub-skills/serving-cache/metadata.json": "2fca222efac7752e29f236f40a0bc38275045826b55955687e39c64ffb69f58f",
"src/inferrsi/data/skill_library/repositories/lithos-metal/sub-skills/serving-cache/references/troubleshooting.md": "52c6e27a76547475b5c66121f873f4d2c1662c811a69da64ce85614f1bd44749",
"src/inferrsi/data/skill_library/repositories/lithos-metal/sub-skills/serving-cache/references/workflows.md": "298d27d3c748fe22774b872ed0465b707e53812379a7a538c95f25d1727fc796",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "lithos-metal",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/local-dream.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/local-dream.json
index 5d641cc45..22a714f55 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/local-dream.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/local-dream.json
@@ -29,8 +29,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/repositories/local-dream/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/repositories/local-dream/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/repositories/local-dream/SKILL.md",
"src/inferrsi/data/skill_library/repositories/local-dream/metadata.json",
@@ -61,7 +61,7 @@
"src/inferrsi/data/skill_library/repositories/local-dream/references/source-identity.json": "b85f028bb96a37852e9d9780b5d5052fdd9c8f3a704ad391b6381adb58291347",
"src/inferrsi/data/skill_library/repositories/local-dream/references/upstream-license.md": "41003d4a74749c0220e33dd415042164b5a1093ed401f36277234f772d22d3d0",
"src/inferrsi/data/skill_library/repositories/local-dream/scripts/screen_declarations.py": "ccd7d0d55be9a47484ec3a74498d8edef8aba55287f92287775c8d82050c8932",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "local-dream",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mistral-rs.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mistral-rs.json
index 55fb63635..3e21ec778 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mistral-rs.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mistral-rs.json
@@ -26,8 +26,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/repositories/mistral-rs/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/repositories/mistral-rs/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/repositories/mistral-rs/LICENSE",
"src/inferrsi/data/skill_library/repositories/mistral-rs/SKILL.md",
@@ -90,7 +90,7 @@
"src/inferrsi/data/skill_library/repositories/mistral-rs/sub-skills/qualification/references/troubleshooting.md": "d933bea6e9e493faa8cd0f4103ea33197ace7f7f901359f8cf0d6bab2edcdb4d",
"src/inferrsi/data/skill_library/repositories/mistral-rs/sub-skills/qualification/references/warmup-inventory.md": "1e7af7d63c6853ede4232b5a12cf68aba6597c4a305015366f1c2c917e3caf01",
"src/inferrsi/data/skill_library/repositories/mistral-rs/sub-skills/qualification/scripts/warmup_shape_inventory.py": "6aa920ca1cb534dec41f5dd2d833c3e760b234ab2c6439b6c4250b8a30b74055",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "mistral-rs",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlc-llm.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlc-llm.json
index 00a38685c..7032a96d5 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlc-llm.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlc-llm.json
@@ -27,8 +27,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/repositories/mlc-llm/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/repositories/mlc-llm/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/repositories/mlc-llm/LICENSE",
"src/inferrsi/data/skill_library/repositories/mlc-llm/SKILL.md",
@@ -97,7 +97,7 @@
"src/inferrsi/data/skill_library/repositories/mlc-llm/sub-skills/runtime-kv-limits/references/limits-and-workspace.md": "77105db95de2b8ea27ed6953b5441c8af061fb745ddb13d3904f8773fc21e879",
"src/inferrsi/data/skill_library/repositories/mlc-llm/sub-skills/runtime-kv-limits/references/troubleshooting.md": "b89255fc04507185bc8bcf74de779730b009b12d13bb38ad0193205dfed8e921",
"src/inferrsi/data/skill_library/repositories/mlc-llm/sub-skills/runtime-kv-limits/scripts/audit_kv_limit_declarations.py": "d82a6dda5f5931eb369003a5e1f419bfbdd82304766a18be8ea1e1849e26b77d",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "mlc-llm",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlx-swift-lm.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlx-swift-lm.json
index 1dced88de..cc8e664c1 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlx-swift-lm.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlx-swift-lm.json
@@ -27,8 +27,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/repositories/mlx-swift-lm/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/repositories/mlx-swift-lm/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/repositories/mlx-swift-lm/LICENSE",
"src/inferrsi/data/skill_library/repositories/mlx-swift-lm/SKILL.md",
@@ -79,7 +79,7 @@
"src/inferrsi/data/skill_library/repositories/mlx-swift-lm/sub-skills/generation-lifecycle/references/ownership-and-termination.md": "56fdec7ebc89b5c217c7e5c203886bef1d3c2eb5794852897bdb5941ab1db2a0",
"src/inferrsi/data/skill_library/repositories/mlx-swift-lm/sub-skills/generation-lifecycle/references/troubleshooting.md": "de28b8ca9cf95be2a4e5762f938acfadb6d4b53459dbf86df4b2b2ec64ec347f",
"src/inferrsi/data/skill_library/repositories/mlx-swift-lm/sub-skills/generation-lifecycle/scripts/audit_generation_evidence.py": "2aba6caf42ca1b3fc55299fd96419af8681c37750c21a7ba75205a43baa9c7b1",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "mlx-swift-lm",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/npu-skill-factory.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/npu-skill-factory.json
index 4d6ec218d..9cb886cf4 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/npu-skill-factory.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/npu-skill-factory.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/workflows/npu-skill-factory/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/workflows/npu-skill-factory/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/skill-index.json",
"src/inferrsi/data/skill_library/workflows/npu-skill-factory/SKILL.md",
@@ -54,7 +54,7 @@
},
"checkout_ref": "checkout:npu-skill-factory",
"source_files": {
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317",
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db",
"src/inferrsi/data/skill_library/workflows/npu-skill-factory/SKILL.md": "74cd863df2e80dc2bda1713d24207809173a3116f30325e3487e1e3517cfec1e",
"src/inferrsi/data/skill_library/workflows/npu-skill-factory/metadata.json": "073174963ca5c19c63d68843ccabdbf758f4c8b7ca61e3d7283affd6f88a753c",
"src/inferrsi/data/skill_library/workflows/npu-skill-factory/references/papers-20261015/COMPARISON.md": "38587678a96581f5c2dc697d982127343565504bffa191352811dffcc165aa23",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/optimization-search-policy.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/optimization-search-policy.json
index 13a8349da..bc67998a1 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/optimization-search-policy.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/optimization-search-policy.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/methods/optimization-search-policy/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/methods/optimization-search-policy/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/methods/optimization-search-policy/SKILL.md",
"src/inferrsi/data/skill_library/methods/optimization-search-policy/cases/sarathi-chunk-budget.json",
@@ -59,7 +59,7 @@
"src/inferrsi/data/skill_library/methods/optimization-search-policy/references/papers-20261015/knowledge/adaptation-interfaces.md": "5e761957cb5ecd93926e6ffc13deee7c830c6e8f0ee3c3846e9dd3b50dc74446",
"src/inferrsi/data/skill_library/methods/optimization-search-policy/references/papers-20261015/references/source-notes.md": "54d83c3713045d59a48821bcf93598374ba4a630e3891ad39498f919317c4fb8",
"src/inferrsi/data/skill_library/methods/optimization-search-policy/references/papers-20261015/updates/optimization-search-policy.md": "d3db44fa9a6831b86a02c638b323ed943e69fe256866213f10d68c72d762879e",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "optimization-search-policy",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/os-process-placement.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/os-process-placement.json
index d91899f66..9c2378939 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/os-process-placement.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/os-process-placement.json
@@ -24,8 +24,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/hardware/host/os-process-placement/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/hardware/host/os-process-placement/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/hardware/host/os-process-placement/SKILL.md",
"src/inferrsi/data/skill_library/hardware/host/os-process-placement/metadata.json",
@@ -44,7 +44,7 @@
"src/inferrsi/data/skill_library/hardware/host/os-process-placement/SKILL.md": "2003492e4d89fde0443b180d74933e764bd1979435e109c8d26e20eb8f6e5053",
"src/inferrsi/data/skill_library/hardware/host/os-process-placement/metadata.json": "ec02995c075f0a540921f57b4846f2cafe334eb90f44389f17c3c50457c3e9c5",
"src/inferrsi/data/skill_library/hardware/host/os-process-placement/references/llama-numa-thread-readset.md": "748b254b3b043a5ddb0f480eb4a566854d27eedd7b111499863e9bf0a5a6951e",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "os-process-placement",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/performance-ceiling-analysis.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/performance-ceiling-analysis.json
index 251b7694e..ea5027a26 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/performance-ceiling-analysis.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/performance-ceiling-analysis.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/SKILL.md",
"src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/metadata.json",
@@ -82,7 +82,7 @@
"src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/references/roofline-contract.md": "419ec4bcaec45ac579a133bce083d4f48779f26e664d7a86f720fa65f343b1d1",
"src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/references/roofline-example.json": "597cd3205df043cff3ae9d15924c77934e8d8eb300c00c4840f6bb2921f90faf",
"src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/scripts/roofline.py": "524de48a6764401695fe1444f1fa3d38f363069b0cd2fba4715a86ca4b28a5d4",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "performance-ceiling-analysis",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/plan-activation-residency.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/plan-activation-residency.json
index a99ef44cb..04f7cd997 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/plan-activation-residency.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/plan-activation-residency.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/methods/plan-activation-residency/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/methods/plan-activation-residency/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/methods/plan-activation-residency/SKILL.md",
"src/inferrsi/data/skill_library/methods/plan-activation-residency/metadata.json",
@@ -36,7 +36,7 @@
"src/inferrsi/data/skill_library/methods/plan-activation-residency/SKILL.md": "542d8a1e6963f5019986e1fbd79d3a247a045a51487c09313026087ac457c15e",
"src/inferrsi/data/skill_library/methods/plan-activation-residency/metadata.json": "0289c62f93d087cd6a4ba4140487925c4b3258a9db1f7004ff1dfb22fd6d4582",
"src/inferrsi/data/skill_library/methods/plan-activation-residency/references/source-notes.md": "5ff52655cfd06fcf970ed2a7d4d4fe1c31115b102af2d41497de0c74848ed2b3",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "plan-activation-residency",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/project-maya.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/project-maya.json
index 85f6a9fe5..20cc1edd9 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/project-maya.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/project-maya.json
@@ -25,8 +25,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/repositories/project-maya/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/repositories/project-maya/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/repositories/project-maya/LICENSE",
"src/inferrsi/data/skill_library/repositories/project-maya/SKILL.md",
@@ -83,7 +83,7 @@
"src/inferrsi/data/skill_library/repositories/project-maya/sub-skills/serving-state/references/native-state.md": "e5a7ee5e777998db0fb5422cf5b1ab18dbd67b4cbbfcfd6628ed6e40d8709c61",
"src/inferrsi/data/skill_library/repositories/project-maya/sub-skills/serving-state/references/troubleshooting.md": "97dea7b2d26fe165d2f64a1a63c91d099dddb2d3b95c16b667f5173b3ffa9d2d",
"src/inferrsi/data/skill_library/repositories/project-maya/sub-skills/serving-state/scripts/request_template.py": "d91b03fbbd7003012a92bd7259d39604b9a91ee5707da2d3366e7c0fb47edd6a",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "project-maya",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/pytorch.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/pytorch.json
index 09306d843..e5bf6c045 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/pytorch.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/pytorch.json
@@ -24,8 +24,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/8115da15b9ef34978de500b08aa9f14ea790c35b/src/inferrsi/data/skill_library/frameworks/pytorch/pytorch/SKILL.md",
- "commit": "8115da15b9ef34978de500b08aa9f14ea790c35b",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/frameworks/pytorch/pytorch/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/frameworks/pytorch/pytorch/LICENSE",
"src/inferrsi/data/skill_library/frameworks/pytorch/pytorch/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/repository-evolution-analysis.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/repository-evolution-analysis.json
index f1811d204..d7a9ffd12 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/repository-evolution-analysis.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/repository-evolution-analysis.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/methods/repository-evolution-analysis/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/methods/repository-evolution-analysis/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/methods/repository-evolution-analysis/SKILL.md",
"src/inferrsi/data/skill_library/methods/repository-evolution-analysis/metadata.json",
@@ -43,7 +43,7 @@
"src/inferrsi/data/skill_library/methods/repository-evolution-analysis/metadata.json": "98ef753a0e74c46f59a69ae267bdc0a5d293761e6f27edc1822fa8bd4548c4d6",
"src/inferrsi/data/skill_library/methods/repository-evolution-analysis/references/capability-presence-audit.md": "c21eb4a6aabe811a49a6540a34c54dbcfdcf7efae3052a79b54034c7838026a5",
"src/inferrsi/data/skill_library/methods/repository-evolution-analysis/references/rtk-consumer-path-transition.md": "a70b1d128fb5e0ac12292a290d624f218a0831d7581fe8980b62661473e0f2b7",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "repository-evolution-analysis",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/rsi-evidence.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/rsi-evidence.json
index bcdcc3afd..55b99bea3 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/rsi-evidence.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/rsi-evidence.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/workflows/rsi-evidence/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/workflows/rsi-evidence/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/skill-index.json",
"src/inferrsi/data/skill_library/workflows/rsi-evidence/SKILL.md",
@@ -45,7 +45,7 @@
},
"checkout_ref": "checkout:rsi-evidence",
"source_files": {
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317",
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db",
"src/inferrsi/data/skill_library/workflows/rsi-evidence/SKILL.md": "08f76e12f6f8e8893f03bcdc8152a96c0c99442e7faecbc2a67a18c632fcb243",
"src/inferrsi/data/skill_library/workflows/rsi-evidence/metadata.json": "caf3ef265df71b3cbec31b6c4affb6ebe9278fba0626b23bc124825d2761e49a",
"src/inferrsi/data/skill_library/workflows/rsi-evidence/references/loopx-goal-governance/LICENSE": "cfc7749b96f63bd31c3c42b5c471bf756814053e847c10f3eb003417bc523d30",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/skill-evolution-playbook.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/skill-evolution-playbook.json
index 2289e3110..a34a155e9 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/skill-evolution-playbook.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/skill-evolution-playbook.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/methods/skill-evolution-playbook/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/methods/skill-evolution-playbook/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/methods/skill-evolution-playbook/SKILL.md",
"src/inferrsi/data/skill_library/methods/skill-evolution-playbook/metadata.json",
@@ -39,7 +39,7 @@
"source_files": {
"src/inferrsi/data/skill_library/methods/skill-evolution-playbook/SKILL.md": "9cd37f21fe0d862069e4798edca8a04d59f8b840e55d521901594a63dddd9ba3",
"src/inferrsi/data/skill_library/methods/skill-evolution-playbook/metadata.json": "d3e790614e09d2d57acb24bfe8d2c2a524cc36bd667bd3867c557554bcbb605e",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "skill-evolution-playbook",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/splash.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/splash.json
index 8e8de52f3..f738604c4 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/splash.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/splash.json
@@ -25,8 +25,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/repositories/splash/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/repositories/splash/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/repositories/splash/LICENSE",
"src/inferrsi/data/skill_library/repositories/splash/SKILL.md",
@@ -77,7 +77,7 @@
"src/inferrsi/data/skill_library/repositories/splash/sub-skills/serving-cache/SKILL.md": "c16f911498e116ff9b31b5a6f0ddd3ae6517b51c0e9d079c337f36a5bcac1df8",
"src/inferrsi/data/skill_library/repositories/splash/sub-skills/serving-cache/references/state-and-api.md": "592603177473ac8e024b7dc19a3d3faedd7c1e0312137eab8bf9a62d2f567be6",
"src/inferrsi/data/skill_library/repositories/splash/sub-skills/serving-cache/references/troubleshooting.md": "4fda461042a3de43c109d7a3cec67542c7d61fe68412bf91606220f2e13e576e",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "splash",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/storage-io-measurement.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/storage-io-measurement.json
index 8bc9231e4..8463b604c 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/storage-io-measurement.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/storage-io-measurement.json
@@ -24,8 +24,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/hardware/host/storage-io-measurement/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/hardware/host/storage-io-measurement/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/hardware/host/storage-io-measurement/SKILL.md",
"src/inferrsi/data/skill_library/hardware/host/storage-io-measurement/metadata.json",
@@ -42,7 +42,7 @@
"source_files": {
"src/inferrsi/data/skill_library/hardware/host/storage-io-measurement/SKILL.md": "63b883c10a8d3e7d918abfc7175b72e2b401ff1e02697f94b8382ec1c8f369b7",
"src/inferrsi/data/skill_library/hardware/host/storage-io-measurement/metadata.json": "0a41c7b00978332b54ee1e4a7c1ade88ecde3e961128f24d1b4f7708caff5ff5",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "storage-io-measurement",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-input-pipeline.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-input-pipeline.json
index ce75694ad..732366ace 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-input-pipeline.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-input-pipeline.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/methods/training-input-pipeline/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/methods/training-input-pipeline/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/methods/training-input-pipeline/SKILL.md",
"src/inferrsi/data/skill_library/methods/training-input-pipeline/metadata.json",
@@ -43,7 +43,7 @@
"src/inferrsi/data/skill_library/methods/training-input-pipeline/SKILL.md": "64bd2c6b795d966045d1d3cca55976d3d49b08b06df7c7e06a70c2824cf55b7e",
"src/inferrsi/data/skill_library/methods/training-input-pipeline/metadata.json": "debad47896f600c7a1e19e477e34026c975dfde320a786dca1ddd02030a5feb1",
"src/inferrsi/data/skill_library/methods/training-input-pipeline/references/sources.md": "60e0f3e88655d8e1d5eb4797f39ca4370deda01614f265b8d2ea4559d8ac020f",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "training-input-pipeline",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-resource-budget.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-resource-budget.json
index 4dd3878e3..bf058278c 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-resource-budget.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-resource-budget.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/methods/training-resource-budget/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/methods/training-resource-budget/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/methods/training-resource-budget/SKILL.md",
"src/inferrsi/data/skill_library/methods/training-resource-budget/metadata.json",
@@ -59,7 +59,7 @@
"src/inferrsi/data/skill_library/methods/training-resource-budget/references/papers-20261015/references/source-readset.md": "2a5c166767f56fa47e7eba7386f1f36da8cb73702e5b5bbe14d6a456ca9ebfae",
"src/inferrsi/data/skill_library/methods/training-resource-budget/references/sources.md": "53d70671a183de4aaa3369f9caf2383a23b6f8a0a7b04464a8d8a639150afc6a",
"src/inferrsi/data/skill_library/methods/training-resource-budget/references/stas-decision-notes.md": "f1a993e6c4ca9a035a2effcaf5500081f227b0cc56cc885a6b22fda0041fc744",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "training-resource-budget",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/video-diffusion-optimization.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/video-diffusion-optimization.json
index bb3e3aa4d..05cc3a848 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/video-diffusion-optimization.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/video-diffusion-optimization.json
@@ -26,8 +26,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/methods/video-diffusion-optimization/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/methods/video-diffusion-optimization/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/methods/video-diffusion-optimization/SKILL.md",
"src/inferrsi/data/skill_library/methods/video-diffusion-optimization/assets/campaign.template.json",
@@ -48,7 +48,7 @@
"src/inferrsi/data/skill_library/methods/video-diffusion-optimization/assets/campaign.template.json": "1a37f1d136612143ddab8f0cbbf424c80433b4da5312edb18d07c0b9d87ef657",
"src/inferrsi/data/skill_library/methods/video-diffusion-optimization/metadata.json": "b27d69142dfea5dd2820170475c2bf6cced2d5eefd1c43b3413c41849af0fd02",
"src/inferrsi/data/skill_library/methods/video-diffusion-optimization/scripts/evaluate_target.py": "9e84df051e03d567746fd73ae7243efb2f3c5ec806fef824d3c4d42a8eb9704c",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "video-diffusion-optimization",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/vllm-metal.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/vllm-metal.json
index 3776d52d5..4ed28de43 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/vllm-metal.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/vllm-metal.json
@@ -26,8 +26,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/repositories/vllm-metal/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/repositories/vllm-metal/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/repositories/vllm-metal/LICENSE",
"src/inferrsi/data/skill_library/repositories/vllm-metal/SKILL.md",
@@ -86,7 +86,7 @@
"src/inferrsi/data/skill_library/repositories/vllm-metal/sub-skills/qualification/references/speculation-and-pipeline.md": "64faced7f31d489e9b0c4b9523a803ff1a76dca0f9315bdc6d0b83935bd338a4",
"src/inferrsi/data/skill_library/repositories/vllm-metal/sub-skills/qualification/references/troubleshooting.md": "a2e85696120ffc2af4249ac9c1e1b96320e02287e5099577c0474c025ab4f5f6",
"src/inferrsi/data/skill_library/repositories/vllm-metal/sub-skills/qualification/scripts/summarize_serving_records.py": "78e660a13a6a6941d006cb11fc2f401a34f489e54273860ae6a208dca6a6a3c2",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "vllm-metal",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/weight-loading-optimization.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/weight-loading-optimization.json
index 9e0124b10..b61f018de 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/weight-loading-optimization.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/weight-loading-optimization.json
@@ -29,8 +29,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/methods/weight-loading-optimization/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/methods/weight-loading-optimization/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/methods/weight-loading-optimization/SKILL.md",
"src/inferrsi/data/skill_library/methods/weight-loading-optimization/metadata.json",
@@ -53,7 +53,7 @@
"src/inferrsi/data/skill_library/methods/weight-loading-optimization/references/fastsafetensors-transfer-readset.md": "5a39f14b102b0cee304541e011aa31e44e6b13527d1bee9987c3f3b66c064d3d",
"src/inferrsi/data/skill_library/methods/weight-loading-optimization/references/flare-preparation-boundary.md": "1d7922785a196bdbafdc9791fe79428465a5acc0292ca8f99446f8c764b35454",
"src/inferrsi/data/skill_library/methods/weight-loading-optimization/references/llama-loading-readset.md": "6c640f28ee3f0253a17b81461c865cb9f947367002cb75bf72550c4572d02e97",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "weight-loading-optimization",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-afd-p2p-exchanges.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-afd-p2p-exchanges.json
index 1f7e75553..ae9d02b96 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-afd-p2p-exchanges.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-afd-p2p-exchanges.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/SKILL.md",
"src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/metadata.json",
@@ -45,7 +45,7 @@
"src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/metadata.json": "ae8c7582265dd6902cc4172293fc709244cc915df9390672fde25687b9e79b4b",
"src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/references/afd-plugin-LICENSE.md": "0e8566ea362276878e1f38f97da49964dbea301c15d4aa9d727f97a4c39fd943",
"src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/references/source-notes.md": "5073d87450cb60d1b245308dff8807e73749c07de976d50f2fa26694fad81a72",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "audit-afd-p2p-exchanges",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-flare-session-lifecycle.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-flare-session-lifecycle.json
index f0443368f..e0e77dc79 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-flare-session-lifecycle.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-flare-session-lifecycle.json
@@ -29,8 +29,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/LICENSE",
"src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/SKILL.md",
@@ -57,7 +57,7 @@
"src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/references/lifecycle.md": "107ebe3c22e5dba8007c3b8ea61149f10bc55c51b557a4230f3e5de859bb0f7c",
"src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/references/source-ledger.json": "61e51a97ffd7c1fa7866bdd440b024d0c5e0065635ac8d0ba53d2c69c7f9e470",
"src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/references/source-notes.md": "fc7ceb819b91ef1731d8b7783f543c92ef6bb5c97ef5f897c055824f8a560168",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "audit-flare-session-lifecycle",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/diagnose-attention-bottlenecks.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/diagnose-attention-bottlenecks.json
index 280f92791..7d4d08677 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/diagnose-attention-bottlenecks.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/diagnose-attention-bottlenecks.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/hardware/cuda/diagnose-attention-bottlenecks/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/hardware/cuda/diagnose-attention-bottlenecks/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/hardware/cuda/diagnose-attention-bottlenecks/SKILL.md",
"src/inferrsi/data/skill_library/hardware/cuda/diagnose-attention-bottlenecks/metadata.json",
@@ -52,7 +52,7 @@
"src/inferrsi/data/skill_library/hardware/cuda/diagnose-attention-bottlenecks/references/papers-20261015/cases/partitioning-after-fusion.md": "4d4cabe0991eab554dcfbd4fa9c8199e61fa4767709b0c4cd646e2588bb398a2",
"src/inferrsi/data/skill_library/hardware/cuda/diagnose-attention-bottlenecks/references/papers-20261015/knowledge/attention-cost-model-boundaries.md": "5ed8f888443e38ccb72545b6025447aab7b92936f78c151fed680e582126239b",
"src/inferrsi/data/skill_library/hardware/cuda/diagnose-attention-bottlenecks/references/sources.md": "53436e0de4e6dda2d29426591bf61f8526378cdc6f9ec7967b88b1bcd4c61e6f",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "diagnose-attention-bottlenecks",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/host-overhead-profiling.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/host-overhead-profiling.json
index 48930d1e2..92ef2edb8 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/host-overhead-profiling.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/host-overhead-profiling.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/hardware/cuda/host-overhead-profiling/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/hardware/cuda/host-overhead-profiling/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/hardware/cuda/host-overhead-profiling/SKILL.md",
"src/inferrsi/data/skill_library/hardware/cuda/host-overhead-profiling/metadata.json",
@@ -40,7 +40,7 @@
"source_files": {
"src/inferrsi/data/skill_library/hardware/cuda/host-overhead-profiling/SKILL.md": "326e566fbe30d3ba04e4b50c3e87aad9df037f0ac4ad6fe129165ae97dc5f91d",
"src/inferrsi/data/skill_library/hardware/cuda/host-overhead-profiling/metadata.json": "76e34c1c64aba473a9e126b032a776417caa0f913f6ac3cf4dba3a5d123b1ee0",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "host-overhead-profiling",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/reason-hopper-kernels.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/reason-hopper-kernels.json
index 10dba6944..6bfdb950e 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/reason-hopper-kernels.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/reason-hopper-kernels.json
@@ -36,8 +36,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/78002cb43fdea1956876880ab251f9f1cd9a367f/src/inferrsi/data/skill_library/hardware/cuda/reason-hopper-kernels/SKILL.md",
- "commit": "78002cb43fdea1956876880ab251f9f1cd9a367f",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/hardware/cuda/reason-hopper-kernels/SKILL.md",
+ "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
"paths": [
"src/inferrsi/data/skill_library/hardware/cuda/reason-hopper-kernels/SKILL.md",
"src/inferrsi/data/skill_library/hardware/cuda/reason-hopper-kernels/metadata.json",
@@ -72,7 +72,7 @@
"src/inferrsi/data/skill_library/hardware/cuda/reason-hopper-kernels/references/source-inventory.json": "1dc09c895c970a0c9708158bcc6dd65cdb736274fd2c35d44542c70407d3849d",
"src/inferrsi/data/skill_library/hardware/cuda/reason-hopper-kernels/references/sources.md": "6b74b9e774f2a80db0860510d9e6940d36284430f29c1e124f115a9eaf25833d",
"src/inferrsi/data/skill_library/hardware/cuda/reason-hopper-kernels/references/synchronization.md": "4322f6b3c0e4368bf8c192b4ec752529767f50d19c0c38dad8fdf9cd781e41d9",
- "src/inferrsi/data/skill_library/skill-index.json": "b5a9f05bd8179644f3fea79d7eb04fe4a8baad9e8d68aa8d0f1db213cf633317"
+ "src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
"name": "reason-hopper-kernels",
From 9c45004af9045af31c7af406df9a920a4754c10d Mon Sep 17 00:00:00 2001
From: Dong1017 <55723162+Dong1017@users.noreply.github.com>
Date: Tue, 3 Nov 2026 13:31:47 +0800
Subject: [PATCH 11/11] docs: regenerate discovery for combined hardware and
repository metadata
Bind projections to combined source snapshot 1bada78556d695a01cf2bc3b0c117f65855cc177.
Agent-assisted-by: Codex / gpt-6-astra [effort=xhigh] (author)
Signed-off-by: Dong1017 <55723162+Dong1017@users.noreply.github.com>
---
docs/skill-coverage.md | 48 +++++++++----------
.../repo_packs/inferrsi/knowledge/README.md | 4 +-
.../ascend-aclgraph-capture-replay.json | 33 ++++++++++---
.../ascend/skills/ascend-device-health.json | 33 ++++++++++---
.../skills/ascend-distributed-parallel.json | 31 +++++++++---
.../ascend/skills/ascend-graph-mode.json | 31 +++++++++---
.../skills/ascend-model-adaptation.json | 33 ++++++++++---
.../skills/ascend-operator-profiling.json | 33 ++++++++++---
.../ascend/skills/ascend-operator-tuning.json | 35 +++++++++++---
.../skills/ascend-optimization-casebook.json | 31 +++++++++---
.../skills/ascend-performance-analysis.json | 33 ++++++++++---
.../ascend-video-diffusion-optimization.json | 28 +++++++----
.../ascend/skills/ascendc-tiling-design.json | 31 +++++++++---
.../skills/cann-attribution-overlap.json | 35 +++++++++++---
.../knowledge/ascend/skills/torch-npu.json | 4 +-
.../torchnpu-device-memory-lifecycle.json | 31 +++++++++---
...torchnpu-pinned-host-memory-lifecycle.json | 31 +++++++++---
.../skills/assess-serving-preemption.json | 4 +-
.../assess-split-backward-schedule.json | 4 +-
.../skills/audit-fixed-workload-claim.json | 4 +-
.../audit-h3-metal-fasth3-semantics.json | 4 +-
.../skills/audit-h3-runtime-residency.json | 4 +-
.../skills/audit-llama-runtime-state.json | 4 +-
.../audit-recurrent-kv-finalization.json | 4 +-
.../common/skills/audit-sd-cpp-residency.json | 4 +-
.../audit-sglang-diffusion-lifetimes.json | 4 +-
.../skills/audit-vllm-kv-lifetimes.json | 4 +-
.../bound-windowed-attention-replay.json | 4 +-
.../budget-speculative-verification.json | 4 +-
.../skills/calibrate-serving-simulation.json | 4 +-
.../skills/cgroup-resource-controls.json | 35 +++++++++++---
.../checkpoint-recovery-consistency.json | 4 +-
.../skills/choose-parallel-training-plan.json | 4 +-
.../skills/cite-check-downstream-pr.json | 4 +-
.../common/skills/claim-contract.json | 4 +-
.../knowledge/common/skills/comfy-ui.json | 4 +-
.../common/skills/competitor-scan.json | 4 +-
.../skills/dcp-async-save-performance.json | 4 +-
.../skills/dense-performance-evidence.json | 4 +-
.../diagnose-vllm-admission-pressure.json | 4 +-
.../skills/diffusers-invocation-state.json | 4 +-
.../common/skills/draft-acceptance-issue.json | 4 +-
.../knowledge/common/skills/dwarfstar.json | 4 +-
.../evaluate-elastic-serving-placement.json | 4 +-
.../skills/generation-cache-lifecycle.json | 4 +-
.../common/skills/host-memory-management.json | 31 +++++++++---
.../skills/host-overhead-optimization.json | 33 ++++++++++---
.../knowledge/common/skills/husky.json | 4 +-
.../skills/inference-benchmark-design.json | 4 +-
.../inference-optimization-casebook.json | 4 +-
.../common/skills/kernel-writing-guide.json | 4 +-
.../common/skills/kv-tier-management.json | 4 +-
.../lazy-evaluation-stream-lifetimes.json | 4 +-
.../knowledge/common/skills/lily.json | 4 +-
.../knowledge/common/skills/lithos-metal.json | 4 +-
.../knowledge/common/skills/local-dream.json | 4 +-
.../knowledge/common/skills/mistral-rs.json | 4 +-
.../knowledge/common/skills/mlc-llm.json | 4 +-
.../knowledge/common/skills/mlx-swift-lm.json | 4 +-
.../common/skills/npu-skill-factory.json | 4 +-
.../skills/optimization-search-policy.json | 4 +-
.../common/skills/os-process-placement.json | 33 ++++++++++---
.../skills/performance-ceiling-analysis.json | 4 +-
.../skills/plan-activation-residency.json | 4 +-
.../knowledge/common/skills/project-maya.json | 4 +-
.../knowledge/common/skills/pytorch.json | 4 +-
.../skills/repository-evolution-analysis.json | 4 +-
.../knowledge/common/skills/rsi-evidence.json | 4 +-
.../skills/skill-evolution-playbook.json | 4 +-
.../knowledge/common/skills/splash.json | 4 +-
.../common/skills/storage-io-measurement.json | 35 +++++++++++---
.../skills/training-input-pipeline.json | 4 +-
.../skills/training-resource-budget.json | 4 +-
.../skills/video-diffusion-optimization.json | 4 +-
.../knowledge/common/skills/vllm-metal.json | 4 +-
.../skills/weight-loading-optimization.json | 4 +-
.../cuda/skills/audit-afd-p2p-exchanges.json | 4 +-
.../skills/audit-flare-session-lifecycle.json | 4 +-
.../diagnose-attention-bottlenecks.json | 31 +++++++++---
.../cuda/skills/host-overhead-profiling.json | 33 ++++++++++---
.../cuda/skills/reason-hopper-kernels.json | 28 +++++++----
81 files changed, 694 insertions(+), 294 deletions(-)
diff --git a/docs/skill-coverage.md b/docs/skill-coverage.md
index 3b0c56ccf..c2bd0ca49 100644
--- a/docs/skill-coverage.md
+++ b/docs/skill-coverage.md
@@ -4,7 +4,7 @@ Generated from the registry, the skill sidecars and the two maintained
matrices. **Do not edit by hand**; regenerate after any registry, skill,
matrix or frozen-plan change.
-- Input digest: `sha256:a314b58c61ab9e0afcbfff44d1d16f1a775d392e3b7133006619994e3529814a`
+- Input digest: `sha256:b60b6ee6224999ed51c45d93648bb7c22108e0522ad44b1471bed5ce5fb26481`
- Regenerate: `python tools/build_skill_coverage.py`
- Verify without writing: `python tools/build_skill_coverage.py --check`
@@ -52,28 +52,28 @@ optimization is fully covered or that any procedure has measured benefit.
| Skill | Backend | Declared task | Applicability | Conditions | SearchSpace layer / stage | Reference Matrix | State | Gaps |
|---|---|---|---|---|---|---|---|---|
-| [ascend-aclgraph-capture-replay](../src/inferrsi/data/skill_library/hardware/ascend/ascend-aclgraph-capture-replay/SKILL.md) | ascend | Diagnose vLLM-Ascend ACLGraph capture/replay eligibility, stable-input and ordering contracts, and capture-resource ... | **unknown** | — | 1 Accelerator · Diagnose / Configure / Optimize / Verify | 1 · Systems sources (vllm-project/vllm-ascend @ `8a8e1e05ecb64a30804c99c8f362909f9a0cb5bf` ...) | candidate · not_run · 1 readset(s) | no-applicability, no-eval |
-| [ascend-device-health](../src/inferrsi/data/skill_library/hardware/ascend/ascend-device-health/SKILL.md) | ascend | Pre-run health checks and recovery procedures for Ascend NPU devices on shared machines: card-group topology ... | **unknown** | — | 4 Network · Diagnose
5 OS · Observe / Profile / Diagnose / Configure / Verify | 1 · Systems sources (Ascend/MindIE-SD @ 8481d39) | candidate · not_run · 0 readset(s) | no-applicability, no-eval |
-| [ascend-distributed-parallel](../src/inferrsi/data/skill_library/hardware/ascend/ascend-distributed-parallel/SKILL.md) | ascend | Select and validate distributed parallelism strategy on Ascend NPU: topology-aware card-group selection, USP/CP/TP/PP ... | **unknown** | — | 1 Accelerator · Configure
4 Network · Diagnose / Configure / Optimize / Verify / Observe | 1 · Systems sources (Ascend/MindIE-SD @ 8481d39)
1 · Systems sources (ascend-ai-coding/awesome-ascend-skills @ 75ec89c)
1 · Systems sources (cann/hccl @ `96fab0a2acb761e7f8ab0f94aeb5138f200a345b` (CANN Open ...) | candidate · not_run · 3 readset(s) | no-applicability, no-eval |
-| [ascend-graph-mode](../src/inferrsi/data/skill_library/hardware/ascend/ascend-graph-mode/SKILL.md) | ascend | Diagnose an Ascend TorchAir ACL graph path from versioned configuration through FX compilation, capture-key selection ... | **unknown** | — | 1 Accelerator · Diagnose / Configure / Verify
3 SSD · Diagnose | 1 · Systems sources (Ascend/torchair @ `e1118e8b0fc1b95190ab53b732c0872989550e20` ...)
1 · Systems sources (ascend-ai-coding/awesome-ascend-skills @ 75ec89c) | candidate · not_run · 3 readset(s) | no-applicability, no-eval |
-| [ascend-model-adaptation](../src/inferrsi/data/skill_library/hardware/ascend/ascend-model-adaptation/SKILL.md) | ascend | Patterns for porting LLM/VLM model adapters to Ascend NPU: FP8 checkpoint handling, KV replication with TP, multimodal ... | **unknown** | — | 1 Accelerator · Configure
3 SSD · Diagnose / Configure | 1 · Systems sources (vllm-project/vllm-ascend @ bc8115ed) | candidate · not_run · 0 readset(s) | no-applicability, no-eval |
-| [ascend-operator-profiling](../src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-profiling/SKILL.md) | ascend | Benchmark a single Ascend NPU operator using msprof op: build, collect with warmup, read the 8 CSV metrics files ... | **unknown** | — | 1 Accelerator · Observe / Profile
5 OS · Observe / Profile | 1 · Systems sources (ascend-ai-coding/awesome-ascend-skills @ 75ec89c) | candidate · not_run · 0 readset(s) | no-applicability, no-eval |
-| [ascend-operator-tuning](../src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-tuning/SKILL.md) | ascend | Run a staged, accuracy-gated tuning campaign on a single Ascend NPU operator: set a numeric target up front, block ... | **unknown** | — | 1 Accelerator · Optimize | 1 · Systems sources (ascend-ai-coding/awesome-ascend-skills @ 75ec89c) | candidate · not_run · 0 readset(s) | no-applicability, no-eval |
-| [ascend-optimization-casebook](../src/inferrsi/data/skill_library/hardware/ascend/ascend-optimization-casebook/SKILL.md) | ascend | Casebook of NPU optimization cases for Ascend inference systems: symptom → diagnosis → action → evidence records for ... | **unknown** | — | 1 Accelerator · Optimize / Verify | 1 · Systems sources (Ascend/MindIE-SD @ 8481d39) | candidate · not_run · 0 readset(s) | no-applicability, no-eval |
-| [ascend-performance-analysis](../src/inferrsi/data/skill_library/hardware/ascend/ascend-performance-analysis/SKILL.md) | ascend | Diagnose performance bottlenecks on Ascend NPU systems using CANN profiling data. Maps GPU profiling concepts (nsys ... | **unknown** | — | 1 Accelerator · Observe / Profile / Diagnose / Verify
4 Network · Observe / Profile / Diagnose / Verify
5 OS · Diagnose | 1 · Systems sources (Ascend/MindIE-SD @ 8481d39)
1 · Systems sources (cann/cannbot @ `c0e2ad23ee1a73e3c0f84c8df1cc459e2aae3110` (selected ...) | candidate · not_run · 1 readset(s) | no-applicability, no-eval |
-| [ascend-video-diffusion-optimization](../src/inferrsi/data/skill_library/hardware/ascend/ascend-video-diffusion-optimization/SKILL.md) | ascend | Run and verify goal-driven Ascend video diffusion optimization. Specializes the backend-agnostic ... | Ascend NPU video diffusion inference and serving optimization | — | — | — | candidate · not_run · 1 readset(s) | no-SearchSpace-layer, no-Reference-Matrix-row, no-eval |
-| [ascendc-tiling-design](../src/inferrsi/data/skill_library/hardware/ascend/ascendc-tiling-design/SKILL.md) | ascend | Design tiling for a custom AscendC operator: normalize the shape first (axis-collapse or flatten), answer the ... | **unknown** | — | 1 Accelerator · Optimize | 1 · Systems sources (ascend-ai-coding/awesome-ascend-skills @ 75ec89c) | candidate · not_run · 0 readset(s) | no-applicability, no-eval |
-| [cann-attribution-overlap](../src/inferrsi/data/skill_library/hardware/ascend/cann-attribution-overlap/SKILL.md) | ascend | Attribute NPU trace kernels to model source and plan bounded multistream overlap: select a representative step, map ... | **unknown** | — | 1 Accelerator · Observe / Profile / Diagnose / Optimize / Verify
4 Network · Diagnose / Optimize / Verify | 1 · Systems sources (cann/oam-tools @ `8440935f` (!302 import; later master `f3cf4115` ...) | candidate · not_run · 0 readset(s) | no-applicability, no-eval |
-| [cgroup-resource-controls](../src/inferrsi/data/skill_library/hardware/host/cgroup-resource-controls/SKILL.md) | common | Diagnose and configure cgroup v2 resource controls for serving/training workloads: distinguish quota stalls from ... | **unknown** | — | 2 RAM/Host · Observe / Profile / Configure / Verify
5 OS · Observe / Profile / Diagnose / Configure / Optimize / Verify | 1 · Systems sources (Linux cgroup v2 docs — kernel.org live 2026-10-15 (revision ...) | candidate · not_run · 0 readset(s) | no-applicability, no-eval |
-| [diagnose-attention-bottlenecks](../src/inferrsi/data/skill_library/hardware/cuda/diagnose-attention-bottlenecks/SKILL.md) | cuda | Diagnose the next attention-kernel bottleneck after IO reduction using evidence about HBM traffic, work ownership ... | **unknown** | — | 1 Accelerator · Diagnose / Optimize / Verify | 1 · Systems sources (FlashAttention — [2205.14135v2](https://arxiv.org/abs/2205.14135v2)) | candidate · not_run · 9 readset(s) | no-applicability, no-eval |
-| [host-memory-management](../src/inferrsi/data/skill_library/hardware/host/host-memory-management/SKILL.md) | common | Audit pinned host buffers and asynchronous CUDA transfers: distinguish copy ordering, host mutation/reuse and device ... | **unknown** | — | 2 RAM/Host · Configure / Verify | 1 · Systems sources (python/cpython @ `60403a5409ff2c3f3b07dd2ca91a7a3e096839c7` (v3.13.0 ...)
1 · Systems sources (pytorch/pytorch @ `ba56102387ef21a3b04b357e5b183d48f0afefc7` (v2.8.0 ...)
1 · Systems sources (vllm-project/vllm-ascend @ `8a8e1e05ecb64a30804c99c8f362909f9a0cb5bf` ...) | candidate · not_run · 4 readset(s) | no-applicability, no-eval |
-| [host-overhead-optimization](../src/inferrsi/data/skill_library/hardware/host/host-overhead-optimization/SKILL.md) | common | Profile and reduce CPU/host overhead in inference serving using line_profiler. Iterative ... | **unknown** | — | 2 RAM/Host · Optimize / Diagnose / Verify | 1 · Systems sources (NVIDIA/TensorRT-LLM @ f33e0e5fd) | candidate · not_run · 0 readset(s) | no-applicability, no-eval |
-| [host-overhead-profiling](../src/inferrsi/data/skill_library/hardware/cuda/host-overhead-profiling/SKILL.md) | cuda | Detect and diagnose CPU/host overhead in inference serving from nsys traces. Answers: is host overhead the bottleneck? ... | **unknown** | — | 2 RAM/Host · Observe / Profile / Diagnose | 1 · Systems sources (NVIDIA/TensorRT-LLM @ f33e0e5fd) | candidate · not_run · 0 readset(s) | no-applicability, no-eval |
-| [os-process-placement](../src/inferrsi/data/skill_library/hardware/host/os-process-placement/SKILL.md) | common | Place and configure inference/training processes at the OS level before measuring or optimizing: check then pin CPU ... | **unknown** | — | 5 OS · Optimize | 1 · Systems sources (NVIDIA/TensorRT-LLM @ f33e0e5fd)
1 · Systems sources (ascend-ai-coding/awesome-ascend-skills @ 75ec89c)
1 · Systems sources (ggml-org/llama.cpp @ `3e390af18de1131dda27913c00511725b4274c8a` (MIT ...)
1 · Systems sources (vllm-project/vllm-ascend @ bc8115ed) | candidate · not_run · 1 readset(s) | no-applicability, no-eval |
-| [reason-hopper-kernels](../src/inferrsi/data/skill_library/hardware/cuda/reason-hopper-kernels/SKILL.md) | cuda | Reason about Hopper SM90 CUDA GEMM kernels using TMA and mbarrier lifetimes, WGMMA warpgroup participation, register ... | Source-derived Hopper SM90/SM90a kernel reasoning for TMA/mbarrier, WGMMA ... | architectures, excluded_mechanisms | 1 Accelerator †guidance: Hopper kernel reasoning (source-derived) | 1 · Systems sources (NVlabs/kda @ `acdd79343c5e9d0c7cb96289f26fd3f1051ce785` → ...) | candidate · not_run · 3 readset(s) | no-eval |
-| [storage-io-measurement](../src/inferrsi/data/skill_library/hardware/host/storage-io-measurement/SKILL.md) | common | Measure storage I/O for LLM/diffusion serving without conflating device metrics with application outcomes: match the ... | **unknown** | — | 3 SSD · Observe / Profile / Diagnose / Verify | 1 · Systems sources (axboe/fio @ acb7a302)
1 · Systems sources (sysstat/sysstat @ efd0c445) | candidate · not_run · 0 readset(s) | no-applicability, no-eval |
-| [torchnpu-device-memory-lifecycle](../src/inferrsi/data/skill_library/hardware/ascend/torchnpu-device-memory-lifecycle/SKILL.md) | ascend | Diagnose TorchNPU device-HBM reuse, cross-stream lifetimes, cache retention and fragmentation, then propose ... | **unknown** | — | 1 Accelerator · Diagnose / Configure / Optimize / Verify | 1 · Systems sources (Ascend/pytorch (TorchNPU) @ `67511c53cd4c88014f6123de7097677e60e85faa` ...) | candidate · not_run · 1 readset(s) | no-applicability, no-eval |
-| [torchnpu-pinned-host-memory-lifecycle](../src/inferrsi/data/skill_library/hardware/ascend/torchnpu-pinned-host-memory-lifecycle/SKILL.md) | ascend | Diagnose TorchNPU pinned-host RAM allocation, asynchronous-copy lifetime, reuse and cache controls, then propose ... | **unknown** | — | 2 RAM/Host · Diagnose / Configure / Optimize / Verify | 1 · Systems sources (Ascend/pytorch (TorchNPU) @ `67511c53cd4c88014f6123de7097677e60e85faa` ...) | candidate · not_run · 1 readset(s) | no-applicability, no-eval |
+| [ascend-aclgraph-capture-replay](../src/inferrsi/data/skill_library/hardware/ascend/ascend-aclgraph-capture-replay/SKILL.md) | ascend | Diagnose vLLM-Ascend ACLGraph capture/replay eligibility, stable-input and ordering contracts, and capture-resource ... | Diagnose vLLM-Ascend ACLGraph eligibility, capture/replay dispatch, caller-buffer ... | input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator · Diagnose / Configure / Optimize / Verify | 1 · Systems sources (vllm-project/vllm-ascend @ `8a8e1e05ecb64a30804c99c8f362909f9a0cb5bf` ...) | candidate · not_run · 1 readset(s) | no-eval |
+| [ascend-device-health](../src/inferrsi/data/skill_library/hardware/ascend/ascend-device-health/SKILL.md) | ascend | Pre-run health checks and recovery procedures for Ascend NPU devices on shared machines: card-group topology ... | Pre-run Ascend device-health, topology, occupancy and timing checks before a trusted ... | failure_modes, input_contract, produced_artifact
3 advice | 4 Network · Diagnose
5 OS · Observe / Profile / Diagnose / Configure / Verify | 1 · Systems sources (Ascend/MindIE-SD @ 8481d39) | candidate · not_run · 0 readset(s) | no-eval |
+| [ascend-distributed-parallel](../src/inferrsi/data/skill_library/hardware/ascend/ascend-distributed-parallel/SKILL.md) | ascend | Select and validate distributed parallelism strategy on Ascend NPU: topology-aware card-group selection, USP/CP/TP/PP ... | Select and diagnose Ascend parallel strategies and their activation/overlap, with a ... | failure_modes, input_contract, produced_artifact
4 advice | 1 Accelerator · Configure
4 Network · Diagnose / Configure / Optimize / Verify / Observe | 1 · Systems sources (Ascend/MindIE-SD @ 8481d39)
1 · Systems sources (ascend-ai-coding/awesome-ascend-skills @ 75ec89c)
1 · Systems sources (cann/hccl @ `96fab0a2acb761e7f8ab0f94aeb5138f200a345b` (CANN Open ...) | candidate · not_run · 3 readset(s) | no-eval |
+| [ascend-graph-mode](../src/inferrsi/data/skill_library/hardware/ascend/ascend-graph-mode/SKILL.md) | ascend | Diagnose an Ascend TorchAir ACL graph path from versioned configuration through FX compilation, capture-key selection ... | Diagnose the TorchAir Python ACL graph path from configuration and FX compilation through ... | failure_modes, input_contract, produced_artifact
3 advice | 1 Accelerator · Diagnose / Configure / Verify
3 SSD · Diagnose | 1 · Systems sources (Ascend/torchair @ `e1118e8b0fc1b95190ab53b732c0872989550e20` ...)
1 · Systems sources (ascend-ai-coding/awesome-ascend-skills @ 75ec89c) | candidate · not_run · 3 readset(s) | no-eval |
+| [ascend-model-adaptation](../src/inferrsi/data/skill_library/hardware/ascend/ascend-model-adaptation/SKILL.md) | ascend | Patterns for porting LLM/VLM model adapters to Ascend NPU: FP8 checkpoint handling, KV replication with TP, multimodal ... | Checkpoint onboarding for LLM/VLM adapters on Ascend: weight/scale loading, KV-head ... | failure_modes, input_contract, produced_artifact
3 advice | 1 Accelerator · Configure
3 SSD · Diagnose / Configure | 1 · Systems sources (vllm-project/vllm-ascend @ bc8115ed) | candidate · not_run · 0 readset(s) | no-eval |
+| [ascend-operator-profiling](../src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-profiling/SKILL.md) | ascend | Benchmark a single Ascend NPU operator using msprof op: build, collect with warmup, read the 8 CSV metrics files ... | Build, measure and diagnose one Ascend operator in isolation using msprof op, before ... | failure_modes, input_contract, produced_artifact
3 advice | 1 Accelerator · Observe / Profile
5 OS · Observe / Profile | 1 · Systems sources (ascend-ai-coding/awesome-ascend-skills @ 75ec89c) | candidate · not_run · 0 readset(s) | no-eval |
+| [ascend-operator-tuning](../src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-tuning/SKILL.md) | ascend | Run a staged, accuracy-gated tuning campaign on a single Ascend NPU operator: set a numeric target up front, block ... | Run a bounded, accuracy-gated tuning campaign for a single Ascend operator already ... | failure_modes, input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator · Optimize | 1 · Systems sources (ascend-ai-coding/awesome-ascend-skills @ 75ec89c) | candidate · not_run · 0 readset(s) | no-eval |
+| [ascend-optimization-casebook](../src/inferrsi/data/skill_library/hardware/ascend/ascend-optimization-casebook/SKILL.md) | ascend | Casebook of NPU optimization cases for Ascend inference systems: symptom → diagnosis → action → evidence records for ... | Find source-derived prior cases for an already classified Ascend inference bottleneck ... | input_contract, produced_artifact, refusal_point
2 advice | 1 Accelerator · Optimize / Verify | 1 · Systems sources (Ascend/MindIE-SD @ 8481d39) | candidate · not_run · 0 readset(s) | no-eval |
+| [ascend-performance-analysis](../src/inferrsi/data/skill_library/hardware/ascend/ascend-performance-analysis/SKILL.md) | ascend | Diagnose performance bottlenecks on Ascend NPU systems using CANN profiling data. Maps GPU profiling concepts (nsys ... | Interpret Ascend CANN profiling evidence from step breakdown through ... | failure_modes, input_contract, produced_artifact
3 advice | 1 Accelerator · Observe / Profile / Diagnose / Verify
4 Network · Observe / Profile / Diagnose / Verify
5 OS · Diagnose | 1 · Systems sources (Ascend/MindIE-SD @ 8481d39)
1 · Systems sources (cann/cannbot @ `c0e2ad23ee1a73e3c0f84c8df1cc459e2aae3110` (selected ...) | candidate · not_run · 1 readset(s) | no-eval |
+| [ascend-video-diffusion-optimization](../src/inferrsi/data/skill_library/hardware/ascend/ascend-video-diffusion-optimization/SKILL.md) | ascend | Run and verify goal-driven Ascend video diffusion optimization. Specializes the backend-agnostic ... | Ascend NPU video diffusion inference and serving optimization | failure_modes, input_contract, produced_artifact
3 advice | — | — | candidate · not_run · 1 readset(s) | no-SearchSpace-layer, no-Reference-Matrix-row, no-eval |
+| [ascendc-tiling-design](../src/inferrsi/data/skill_library/hardware/ascend/ascendc-tiling-design/SKILL.md) | ascend | Design tiling for a custom AscendC operator: normalize the shape first (axis-collapse or flatten), answer the ... | Design or restructure AscendC tiling for the covered reduction, broadcast and elementwise ... | failure_modes, input_contract, produced_artifact
3 advice | 1 Accelerator · Optimize | 1 · Systems sources (ascend-ai-coding/awesome-ascend-skills @ 75ec89c) | candidate · not_run · 0 readset(s) | no-eval |
+| [cann-attribution-overlap](../src/inferrsi/data/skill_library/hardware/ascend/cann-attribution-overlap/SKILL.md) | ascend | Attribute NPU trace kernels to model source and plan bounded multistream overlap: select a representative step, map ... | Attribute representative CANN trace kernels to source owners and plan bounded multistream ... | failure_modes, input_contract, produced_artifact
4 advice | 1 Accelerator · Observe / Profile / Diagnose / Optimize / Verify
4 Network · Diagnose / Optimize / Verify | 1 · Systems sources (cann/oam-tools @ `8440935f` (!302 import; later master `f3cf4115` ...) | candidate · not_run · 0 readset(s) | no-eval |
+| [cgroup-resource-controls](../src/inferrsi/data/skill_library/hardware/host/cgroup-resource-controls/SKILL.md) | common | Diagnose and configure cgroup v2 resource controls for serving/training workloads: distinguish quota stalls from ... | Audit Linux cgroup v2 CPU quota/shares, effective placement, memory pressure/protection ... | failure_modes, input_contract, produced_artifact
4 advice | 2 RAM/Host · Observe / Profile / Configure / Verify
5 OS · Observe / Profile / Diagnose / Configure / Optimize / Verify | 1 · Systems sources (Linux cgroup v2 docs — kernel.org live 2026-10-15 (revision ...) | candidate · not_run · 0 readset(s) | no-eval |
+| [diagnose-attention-bottlenecks](../src/inferrsi/data/skill_library/hardware/cuda/diagnose-attention-bottlenecks/SKILL.md) | cuda | Diagnose the next attention-kernel bottleneck after IO reduction using evidence about HBM traffic, work ownership ... | Diagnose attention-kernel bottlenecks after IO reduction, tile/pipeline regressions and ... | input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator · Diagnose / Optimize / Verify | 1 · Systems sources (FlashAttention — [2205.14135v2](https://arxiv.org/abs/2205.14135v2)) | candidate · not_run · 9 readset(s) | no-eval |
+| [host-memory-management](../src/inferrsi/data/skill_library/hardware/host/host-memory-management/SKILL.md) | common | Audit pinned host buffers and asynchronous CUDA transfers: distinguish copy ordering, host mutation/reuse and device ... | Audit host preparation, pinned-host pressure and asynchronous CUDA-copy ... | failure_modes, input_contract, produced_artifact
3 advice | 2 RAM/Host · Configure / Verify | 1 · Systems sources (python/cpython @ `60403a5409ff2c3f3b07dd2ca91a7a3e096839c7` (v3.13.0 ...)
1 · Systems sources (pytorch/pytorch @ `ba56102387ef21a3b04b357e5b183d48f0afefc7` (v2.8.0 ...)
1 · Systems sources (vllm-project/vllm-ascend @ `8a8e1e05ecb64a30804c99c8f362909f9a0cb5bf` ...) | candidate · not_run · 4 readset(s) | no-eval |
+| [host-overhead-optimization](../src/inferrsi/data/skill_library/hardware/host/host-overhead-optimization/SKILL.md) | common | Profile and reduce CPU/host overhead in inference serving using line_profiler. Iterative ... | Use line-level profiling to reduce exposed Python request-preparation, scheduling and ... | failure_modes, input_contract, produced_artifact
3 advice | 2 RAM/Host · Optimize / Diagnose / Verify | 1 · Systems sources (NVIDIA/TensorRT-LLM @ f33e0e5fd) | candidate · not_run · 0 readset(s) | no-eval |
+| [host-overhead-profiling](../src/inferrsi/data/skill_library/hardware/cuda/host-overhead-profiling/SKILL.md) | cuda | Detect and diagnose CPU/host overhead in inference serving from nsys traces. Answers: is host overhead the bottleneck? ... | Detect exposed host overhead and diagnose per-step host regressions in CUDA inference ... | failure_modes, input_contract, produced_artifact
3 advice | 2 RAM/Host · Observe / Profile / Diagnose | 1 · Systems sources (NVIDIA/TensorRT-LLM @ f33e0e5fd) | candidate · not_run · 0 readset(s) | no-eval |
+| [os-process-placement](../src/inferrsi/data/skill_library/hardware/host/os-process-placement/SKILL.md) | common | Place and configure inference/training processes at the OS level before measuring or optimizing: check then pin CPU ... | Prepare and record CPU/NUMA process placement, phase-specific thread settings, device ... | failure_modes, input_contract, produced_artifact
3 advice | 5 OS · Optimize | 1 · Systems sources (NVIDIA/TensorRT-LLM @ f33e0e5fd)
1 · Systems sources (ascend-ai-coding/awesome-ascend-skills @ 75ec89c)
1 · Systems sources (ggml-org/llama.cpp @ `3e390af18de1131dda27913c00511725b4274c8a` (MIT ...)
1 · Systems sources (vllm-project/vllm-ascend @ bc8115ed) | candidate · not_run · 1 readset(s) | no-eval |
+| [reason-hopper-kernels](../src/inferrsi/data/skill_library/hardware/cuda/reason-hopper-kernels/SKILL.md) | cuda | Reason about Hopper SM90 CUDA GEMM kernels using TMA and mbarrier lifetimes, WGMMA warpgroup participation, register ... | Source-derived Hopper SM90/SM90a kernel reasoning for TMA/mbarrier, WGMMA ... | architectures, excluded_mechanisms, failure_modes, input_contract, produced_artifact
3 advice | 1 Accelerator †guidance: Hopper kernel reasoning (source-derived) | 1 · Systems sources (NVlabs/kda @ `acdd79343c5e9d0c7cb96289f26fd3f1051ce785` → ...) | candidate · not_run · 3 readset(s) | no-eval |
+| [storage-io-measurement](../src/inferrsi/data/skill_library/hardware/host/storage-io-measurement/SKILL.md) | common | Measure storage I/O for LLM/diffusion serving without conflating device metrics with application outcomes: match the ... | Measure and attribute storage traffic for model loading or serving while separating ... | input_contract, produced_artifact, refusal_point
4 advice | 3 SSD · Observe / Profile / Diagnose / Verify | 1 · Systems sources (axboe/fio @ acb7a302)
1 · Systems sources (sysstat/sysstat @ efd0c445) | candidate · not_run · 0 readset(s) | no-eval |
+| [torchnpu-device-memory-lifecycle](../src/inferrsi/data/skill_library/hardware/ascend/torchnpu-device-memory-lifecycle/SKILL.md) | ascend | Diagnose TorchNPU device-HBM reuse, cross-stream lifetimes, cache retention and fragmentation, then propose ... | Diagnose TorchNPU device-HBM reuse, stream lifetimes, fragmentation and ... | input_contract, produced_artifact, refusal_point
3 advice | 1 Accelerator · Diagnose / Configure / Optimize / Verify | 1 · Systems sources (Ascend/pytorch (TorchNPU) @ `67511c53cd4c88014f6123de7097677e60e85faa` ...) | candidate · not_run · 1 readset(s) | no-eval |
+| [torchnpu-pinned-host-memory-lifecycle](../src/inferrsi/data/skill_library/hardware/ascend/torchnpu-pinned-host-memory-lifecycle/SKILL.md) | ascend | Diagnose TorchNPU pinned-host RAM allocation, asynchronous-copy lifetime, reuse and cache controls, then propose ... | Diagnose TorchNPU pinned-host RAM allocation, asynchronous-copy ownership/readiness ... | input_contract, produced_artifact, refusal_point
3 advice | 2 RAM/Host · Diagnose / Configure / Optimize / Verify | 1 · Systems sources (Ascend/pytorch (TorchNPU) @ `67511c53cd4c88014f6123de7097677e60e85faa` ...) | candidate · not_run · 1 readset(s) | no-eval |
## Registered: Frameworks (5)
@@ -239,7 +239,7 @@ an entrypoint routes to procedures below it rather than adding one.
| `no-SearchSpace-layer` | 15: ascend-video-diffusion-optimization, audit-fixed-workload-claim, audit-recurrent-kv-finalization, cite-check-downstream-pr, competitor-scan, draft-acceptance-issue, dwarfstar, husky, inference-benchmark-design, lily, npu-skill-factory, repository-evolution-analysis, skill-evolution-playbook, splash, video-diffusion-optimization |
| `no-Reference-Matrix-row` | 8: ascend-video-diffusion-optimization, audit-fixed-workload-claim, cite-check-downstream-pr, claim-contract, competitor-scan, draft-acceptance-issue, rsi-evidence, video-diffusion-optimization |
| `no-sidecar` | 0: — |
-| `no-applicability` | 20: ascend-aclgraph-capture-replay, ascend-device-health, ascend-distributed-parallel, ascend-graph-mode, ascend-model-adaptation, ascend-operator-profiling, ascend-operator-tuning, ascend-optimization-casebook, ascend-performance-analysis, ascendc-tiling-design, cann-attribution-overlap, cgroup-resource-controls, diagnose-attention-bottlenecks, host-memory-management, host-overhead-optimization, host-overhead-profiling, os-process-placement, storage-io-measurement, torchnpu-device-memory-lifecycle, torchnpu-pinned-host-memory-lifecycle |
+| `no-applicability` | 0: — |
| `no-eval` | 77: as listed in the Gaps column above |
These are inventory gaps, not defects. A workflow skill may legitimately
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/README.md b/src/inferrsi/data/repo_packs/inferrsi/knowledge/README.md
index 400fbecbc..bbd74d7a5 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/README.md
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/README.md
@@ -15,7 +15,7 @@ These are knowledge abstracts (`--kind knowledge`), with namespaced package
IDs distinct from explicitly selected `checkout_ref` skills. Ascend/CUDA
procedures use their corresponding backend; shared methods use `common`.
-Source snapshot: [ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90](https://github.com/InferMatrixLab/InferRSI/tree/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library).
+Source snapshot: [1bada78556d695a01cf2bc3b0c117f65855cc177](https://github.com/InferMatrixLab/InferRSI/tree/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library).
```bash
inferrsi knowledge search 'profiling' --repo-type inferrsi --limit 50
@@ -29,7 +29,7 @@ contain the current input paths and bytes (Git checkout filters are respected).
```bash
python tools/build_skill_index.py --source-revision
-python tools/build_skill_index.py --source-revision ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90 --check
+python tools/build_skill_index.py --source-revision 1bada78556d695a01cf2bc3b0c117f65855cc177 --check
```
`--check` detects missing, stale and unexpected generated records without writing.
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-aclgraph-capture-replay.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-aclgraph-capture-replay.json
index f8103539f..38ede9b19 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-aclgraph-capture-replay.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-aclgraph-capture-replay.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/hardware/ascend/ascend-aclgraph-capture-replay/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/hardware/ascend/ascend-aclgraph-capture-replay/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-aclgraph-capture-replay/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-aclgraph-capture-replay/metadata.json",
@@ -39,7 +39,7 @@
"checkout_ref": "checkout:ascend-aclgraph-capture-replay",
"source_files": {
"src/inferrsi/data/skill_library/hardware/ascend/ascend-aclgraph-capture-replay/SKILL.md": "d494fc7e539cbb7388189bd7b3f04bb5c4983fdd7a82484348aad1cee7f741f2",
- "src/inferrsi/data/skill_library/hardware/ascend/ascend-aclgraph-capture-replay/metadata.json": "71c68431663ed2336cccc985a354adab6b13b4cfd6123a18a5d8d53c28268509",
+ "src/inferrsi/data/skill_library/hardware/ascend/ascend-aclgraph-capture-replay/metadata.json": "3ac69dd82ae798450092c830b62f4958dc2f5b1c7000f9c3f822f2b0693f8af4",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-aclgraph-capture-replay/references/source-notes.md": "6ffbfb68cc7e04923b4fbca2a73a7555b7c3b45a6f40819bb5f3213ef05cf88d",
"src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
@@ -67,12 +67,31 @@
"provenance_readsets": [
"references/source-notes.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "applicability": {
+ "scope": "Diagnose vLLM-Ascend ACLGraph eligibility, capture/replay dispatch, caller-buffer ownership and capture-resource failures.",
+ "prerequisites": "An existing eager correctness baseline; authorization before any new run or live configuration change.",
+ "input_contract": "Installed stack and device, requested/effective graph mode, runner/backend settings, capture sizes and batch descriptors, error phase, replay observations and HBM/stream evidence.",
+ "produced_artifact": "A bounded diagnosis, smallest reversible proposal, expected tradeoff and correctness/resource verification plan with unresolved compatibility or authorization questions.",
+ "refusal_point": "Stop on output mismatch, unstable replay addresses, capture exhaustion, unknown ordering or version drift; missing evidence permits only a conditional proposal."
+ },
+ "recommendations": [
+ "Diagnose the effective runtime mode and batch descriptor before changing capture sizes; intended warmup/profile bypass is not a cache defect.",
+ "Trace stable replay buffers to the owning caller and preserve synchronization/update ordering; the wrapper does not stage new inputs into persistent buffers.",
+ "Compare eager and graph execution on the same workload, separating initial capture, repeated replay, compiler changes and memory/stream costs; a replay log alone proves neither coverage nor speedup."
+ ]
+ },
+ "applicability": {
+ "scope": "Diagnose vLLM-Ascend ACLGraph eligibility, capture/replay dispatch, caller-buffer ownership and capture-resource failures.",
+ "prerequisites": "An existing eager correctness baseline; authorization before any new run or live configuration change.",
+ "input_contract": "Installed stack and device, requested/effective graph mode, runner/backend settings, capture sizes and batch descriptors, error phase, replay observations and HBM/stream evidence.",
+ "produced_artifact": "A bounded diagnosis, smallest reversible proposal, expected tradeoff and correctness/resource verification plan with unresolved compatibility or authorization questions.",
+ "refusal_point": "Stop on output mismatch, unstable replay addresses, capture exhaustion, unknown ordering or version drift; missing evidence permits only a conditional proposal."
},
- "applicability": {},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Diagnose the effective runtime mode and batch descriptor before changing capture sizes; intended warmup/profile bypass is not a cache defect.",
+ "Trace stable replay buffers to the owning caller and preserve synchronization/update ordering; the wrapper does not stage new inputs into persistent buffers.",
+ "Compare eager and graph execution on the same workload, separating initial capture, repeated replay, compiler changes and memory/stream costs; a replay log alone proves neither coverage nor speedup."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-device-health.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-device-health.json
index 3f68d516f..b7da6596e 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-device-health.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-device-health.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/hardware/ascend/ascend-device-health/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/hardware/ascend/ascend-device-health/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-device-health/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-device-health/metadata.json",
@@ -40,7 +40,7 @@
"checkout_ref": "checkout:ascend-device-health",
"source_files": {
"src/inferrsi/data/skill_library/hardware/ascend/ascend-device-health/SKILL.md": "03829039860ece35519f2e4c339a48b5014ec25e07a2c150a3efb6a3a4e35b4a",
- "src/inferrsi/data/skill_library/hardware/ascend/ascend-device-health/metadata.json": "80d80c15514919826fa72d6f6cf26b76db8accd85d21187fc239232ca454e7bf",
+ "src/inferrsi/data/skill_library/hardware/ascend/ascend-device-health/metadata.json": "f643185563dbf528a5a1f2b0561685ae34302a9efc9900c6753a7168b4e8b329",
"src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
@@ -67,12 +67,31 @@
"Original typed source inventory is not normalized; retain source identities and limits in the procedure/readsets."
],
"provenance_readsets": [],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "applicability": {
+ "scope": "Pre-run Ascend device-health, topology, occupancy and timing checks before a trusted baseline, multi-card run or graph capture, including after a killed job.",
+ "hardware": "Ascend NPU card groups with topology and health visible through npu-smi.",
+ "input_contract": "Intended card group, current topology/health and process occupancy, prior termination history and expected workload step timing.",
+ "produced_artifact": "A pre-run health/occupancy and step-timing record, with recovery observations and a baseline-remeasurement requirement after a card-group change.",
+ "failure_modes": "Health=OK can miss degraded post-kill timing; foreign processes, unwarmed clocks or comparing different card groups can contaminate the baseline."
+ },
+ "recommendations": [
+ "Read topology each session rather than hardcoding card numbers, then check health and foreign-process occupancy before the workload smoke check.",
+ "Use the short step-timing check to investigate post-kill degradation; record affected cards and timings instead of treating Health=OK as sufficient evidence.",
+ "Warm up operator measurements and remeasure the baseline after changing card groups; set only environment variables needed by the run."
+ ]
+ },
+ "applicability": {
+ "scope": "Pre-run Ascend device-health, topology, occupancy and timing checks before a trusted baseline, multi-card run or graph capture, including after a killed job.",
+ "hardware": "Ascend NPU card groups with topology and health visible through npu-smi.",
+ "input_contract": "Intended card group, current topology/health and process occupancy, prior termination history and expected workload step timing.",
+ "produced_artifact": "A pre-run health/occupancy and step-timing record, with recovery observations and a baseline-remeasurement requirement after a card-group change.",
+ "failure_modes": "Health=OK can miss degraded post-kill timing; foreign processes, unwarmed clocks or comparing different card groups can contaminate the baseline."
},
- "applicability": {},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Read topology each session rather than hardcoding card numbers, then check health and foreign-process occupancy before the workload smoke check.",
+ "Use the short step-timing check to investigate post-kill degradation; record affected cards and timings instead of treating Health=OK as sufficient evidence.",
+ "Warm up operator measurements and remeasure the baseline after changing card groups; set only environment variables needed by the run."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-distributed-parallel.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-distributed-parallel.json
index cc7383994..54cf6e9bf 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-distributed-parallel.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-distributed-parallel.json
@@ -27,8 +27,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/hardware/ascend/ascend-distributed-parallel/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/hardware/ascend/ascend-distributed-parallel/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-distributed-parallel/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-distributed-parallel/metadata.json",
@@ -48,7 +48,7 @@
"checkout_ref": "checkout:ascend-distributed-parallel",
"source_files": {
"src/inferrsi/data/skill_library/hardware/ascend/ascend-distributed-parallel/SKILL.md": "0b79d95e8d5abf0a822f8ae6bfa3779cbb40595c5b6f0fb934233275f5899ef8",
- "src/inferrsi/data/skill_library/hardware/ascend/ascend-distributed-parallel/metadata.json": "56ae2092fb4d6538603c1d3d3916f0a3f0eb443d23dad3217ebd74d48d208d94",
+ "src/inferrsi/data/skill_library/hardware/ascend/ascend-distributed-parallel/metadata.json": "096a12d4a5b98bd057e1a87f972038fd43556e82e484b5134943b6e45e880178",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-distributed-parallel/references/awesome-ascend-communication-tests.md": "0cb27a3d24409171bb414d7ebf0c49ad62be5f46482f29624695496ba9eb3e63",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-distributed-parallel/references/hccl-LICENSE.md": "05205a02acebd671c78d3b876c1f6097b84cce89780f5d9bdc204bfbea92d0c1",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-distributed-parallel/references/hccl-allreduce-readset.json": "4a3d403986dfceaa28605f2f30b79545e1efdc27f35c88755402b75b4d13d7e2",
@@ -139,12 +139,31 @@
"references/hccl-allreduce.md",
"references/hccl-allreduce-readset.json",
"references/hccl-LICENSE.md"
+ ],
+ "applicability": {
+ "scope": "Select and diagnose Ascend parallel strategies and their activation/overlap, with a separately bounded native multi-rank HCCL AllReduce dispatch and correctness procedure.",
+ "input_contract": "Installed collective/runtime builds, rank/device map and topology, workload/strategy settings, count/dtype/reduction, effective dispatch, per-rank outputs and completed-collective/application timing evidence.",
+ "produced_artifact": "A topology- and workload-qualified strategy/dispatch diagnosis, activation and correctness checks, stage-cost attribution and a target-validation proposal; isolated test completion is not a service result.",
+ "failure_modes": "Requested parallel settings can silently fall back; stale local-head/chunk/shard parameters can disable a path, and changed topology/card count can confound strategy comparisons."
+ },
+ "recommendations": [
+ "Confirm the final collective algorithm/engine after fallback and observe sharding/collective activation before interpreting correct outputs or tuning performance.",
+ "Preserve peer/thread waits and test distinct rank/position data with explicit completion on every rank; keep enqueue time, collective completion and application-stage timing separate.",
+ "Recompute topology-derived parameters and remeasure the anchor after changing card groups or parallel degree; inherited diffusion topology restrictions are not universal HCCL gates.",
+ "Require producer/consumer timeline and exposed-communication evidence for overlap; compare the real pipeline wait cost rather than extrapolating small collectives from bytes."
]
},
- "applicability": {},
+ "applicability": {
+ "scope": "Select and diagnose Ascend parallel strategies and their activation/overlap, with a separately bounded native multi-rank HCCL AllReduce dispatch and correctness procedure.",
+ "input_contract": "Installed collective/runtime builds, rank/device map and topology, workload/strategy settings, count/dtype/reduction, effective dispatch, per-rank outputs and completed-collective/application timing evidence.",
+ "produced_artifact": "A topology- and workload-qualified strategy/dispatch diagnosis, activation and correctness checks, stage-cost attribution and a target-validation proposal; isolated test completion is not a service result.",
+ "failure_modes": "Requested parallel settings can silently fall back; stale local-head/chunk/shard parameters can disable a path, and changed topology/card count can confound strategy comparisons."
+ },
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Confirm the final collective algorithm/engine after fallback and observe sharding/collective activation before interpreting correct outputs or tuning performance.",
+ "Preserve peer/thread waits and test distinct rank/position data with explicit completion on every rank; keep enqueue time, collective completion and application-stage timing separate.",
+ "Recompute topology-derived parameters and remeasure the anchor after changing card groups or parallel degree; inherited diffusion topology restrictions are not universal HCCL gates.",
+ "Require producer/consumer timeline and exposed-communication evidence for overlap; compare the real pipeline wait cost rather than extrapolating small collectives from bytes."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-graph-mode.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-graph-mode.json
index 41b4aa8e4..640f058fb 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-graph-mode.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-graph-mode.json
@@ -26,8 +26,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/hardware/ascend/ascend-graph-mode/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/hardware/ascend/ascend-graph-mode/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-graph-mode/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-graph-mode/metadata.json",
@@ -47,7 +47,7 @@
"checkout_ref": "checkout:ascend-graph-mode",
"source_files": {
"src/inferrsi/data/skill_library/hardware/ascend/ascend-graph-mode/SKILL.md": "a7341b5bf3b43f32d6114e2109e5043e46f3ddeef9b7bb3b20c1c50236bab638",
- "src/inferrsi/data/skill_library/hardware/ascend/ascend-graph-mode/metadata.json": "3912bdfc721a703580a97a42ca13ac708396b17c4b600bf9e76375dd061fd29c",
+ "src/inferrsi/data/skill_library/hardware/ascend/ascend-graph-mode/metadata.json": "017a288c514f6f0aab9d00247af7923258816f0987db15e09a5025a1f3937aa2",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-graph-mode/references/legacy-source-notes.md": "a51c655aff59741cd7421277bd71dffeb9907560950459ae0a776207e496de4a",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-graph-mode/references/torchair-LICENSE.md": "bc07d81fa332b6543463588a120518c57b5ce8582d3747de75d2f6b09b068c01",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-graph-mode/references/torchair-graph-path.md": "272424571ff439278c8f2675406eae7da9a6482c581973e24319739fa0e58348",
@@ -119,12 +119,31 @@
"references/torchair-LICENSE.md",
"references/legacy-source-notes.md",
"../ascend-aclgraph-capture-replay/SKILL.md"
+ ],
+ "applicability": {
+ "scope": "Diagnose the TorchAir Python ACL graph path from configuration and FX compilation through capture-key selection, fallback, input staging, replay and output lifetime.",
+ "prerequisites": "Eager correctness established first; configuration must match the installed backend, with authorization and workload-specific tolerances for any NPU comparison.",
+ "input_contract": "Installed hardware/runtime versions, actual compile backend/options, shape/dtype/stride and alias/mutation contract, retained outputs, capture/replay/fallback observations and workload mix.",
+ "produced_artifact": "A branch-specific diagnosis and bounded configuration or lifetime proposal with correctness, cold/warm cost, resource and concurrency checks still required on the actual stack.",
+ "failure_modes": "Shape equality or clone_input=False does not prove capture reuse or zero-copy; capture limits can trigger persistent fallback and later replay can overwrite retained outputs."
+ },
+ "recommendations": [
+ "Use the installed CompilerConfig/backend surface and a fresh compiled instance for option changes; do not mix registered-backend snippets with callable-backend configuration.",
+ "Observe capture, recapture, replay and persistent fallback separately, including changed addresses and revisiting old keys after a capture-limit event.",
+ "Preserve earlier results before another replay and check views/input-output aliases; keep cold capture, input/output copying, optional compiler passes and service timing separate."
]
},
- "applicability": {},
+ "applicability": {
+ "scope": "Diagnose the TorchAir Python ACL graph path from configuration and FX compilation through capture-key selection, fallback, input staging, replay and output lifetime.",
+ "prerequisites": "Eager correctness established first; configuration must match the installed backend, with authorization and workload-specific tolerances for any NPU comparison.",
+ "input_contract": "Installed hardware/runtime versions, actual compile backend/options, shape/dtype/stride and alias/mutation contract, retained outputs, capture/replay/fallback observations and workload mix.",
+ "produced_artifact": "A branch-specific diagnosis and bounded configuration or lifetime proposal with correctness, cold/warm cost, resource and concurrency checks still required on the actual stack.",
+ "failure_modes": "Shape equality or clone_input=False does not prove capture reuse or zero-copy; capture limits can trigger persistent fallback and later replay can overwrite retained outputs."
+ },
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Use the installed CompilerConfig/backend surface and a fresh compiled instance for option changes; do not mix registered-backend snippets with callable-backend configuration.",
+ "Observe capture, recapture, replay and persistent fallback separately, including changed addresses and revisiting old keys after a capture-limit event.",
+ "Preserve earlier results before another replay and check views/input-output aliases; keep cold capture, input/output copying, optional compiler passes and service timing separate."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-model-adaptation.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-model-adaptation.json
index b872e64ab..096ab6dcb 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-model-adaptation.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-model-adaptation.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/hardware/ascend/ascend-model-adaptation/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/hardware/ascend/ascend-model-adaptation/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-model-adaptation/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-model-adaptation/metadata.json",
@@ -39,7 +39,7 @@
"checkout_ref": "checkout:ascend-model-adaptation",
"source_files": {
"src/inferrsi/data/skill_library/hardware/ascend/ascend-model-adaptation/SKILL.md": "8fdf46fb2743b67a66bc069d22a45305a222debfb606777bc451b4b8f31d9cc6",
- "src/inferrsi/data/skill_library/hardware/ascend/ascend-model-adaptation/metadata.json": "be681a3206e5e183b4c9670c9d7784bcffce18c891449e5b15f94ca5f1324b9a",
+ "src/inferrsi/data/skill_library/hardware/ascend/ascend-model-adaptation/metadata.json": "989b91821b36c0b80caf1323ec527e7ac847219b8f4fe8ca4ea22d292365cc3c",
"src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
@@ -65,12 +65,31 @@
"Original typed source inventory is not normalized; retain source identities and limits in the procedure/readsets."
],
"provenance_readsets": [],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "applicability": {
+ "scope": "Checkpoint onboarding for LLM/VLM adapters on Ascend: weight/scale loading, KV-head replication under tensor parallelism and staged multimodal feature checks.",
+ "prerequisites": "Check the installed CANN and checkpoint format against the versioned guidance; final sign-off requires real weights and at least one full-layer validation.",
+ "input_contract": "Checkpoint weights/configuration and scale keys, attention-head/TP configuration, model modalities and expected graph/capacity features.",
+ "produced_artifact": "A checkpoint/feature report separating dummy and real-weight evidence, applicable feature outcomes, and theoretical versus demonstrated serving capacity.",
+ "failure_modes": "Dummy-only success misses scale/remapping/sharding defects; blindly sharding replicated KV heads or accepting reduced-layer results can conceal loading or full-model failures."
+ },
+ "recommendations": [
+ "Gate onboarding in order: dummy architecture check, real-weight loading, model endpoint, text request and then multimodal request when the model supports VL; retain the evidence each stage actually supplies.",
+ "Check weight/scale pairing and each attention tensor loading path, especially when TP exceeds the KV-head count; verify historical FP8 guidance against the actual stack.",
+ "Expand capacity/concurrency only after basic modality and expected replay checks; mark MoE-only features inapplicable for non-MoE models and require full-layer validation before sign-off."
+ ]
+ },
+ "applicability": {
+ "scope": "Checkpoint onboarding for LLM/VLM adapters on Ascend: weight/scale loading, KV-head replication under tensor parallelism and staged multimodal feature checks.",
+ "prerequisites": "Check the installed CANN and checkpoint format against the versioned guidance; final sign-off requires real weights and at least one full-layer validation.",
+ "input_contract": "Checkpoint weights/configuration and scale keys, attention-head/TP configuration, model modalities and expected graph/capacity features.",
+ "produced_artifact": "A checkpoint/feature report separating dummy and real-weight evidence, applicable feature outcomes, and theoretical versus demonstrated serving capacity.",
+ "failure_modes": "Dummy-only success misses scale/remapping/sharding defects; blindly sharding replicated KV heads or accepting reduced-layer results can conceal loading or full-model failures."
},
- "applicability": {},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Gate onboarding in order: dummy architecture check, real-weight loading, model endpoint, text request and then multimodal request when the model supports VL; retain the evidence each stage actually supplies.",
+ "Check weight/scale pairing and each attention tensor loading path, especially when TP exceeds the KV-head count; verify historical FP8 guidance against the actual stack.",
+ "Expand capacity/concurrency only after basic modality and expected replay checks; mark MoE-only features inapplicable for non-MoE models and require full-layer validation before sign-off."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-profiling.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-profiling.json
index 761cc93a9..f3b116c32 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-profiling.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-profiling.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-profiling/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-profiling/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-profiling/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-profiling/metadata.json",
@@ -40,7 +40,7 @@
"checkout_ref": "checkout:ascend-operator-profiling",
"source_files": {
"src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-profiling/SKILL.md": "401981cf7828edcab0c9ea882f7ecd83badfa3c2a95006e912cf99d67b3d6a12",
- "src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-profiling/metadata.json": "efd37bb23acb500dad4ec8ebe7ddfe5c3f196374052825f923ac85da8445f06b",
+ "src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-profiling/metadata.json": "b2b93637f01546ecb1b7076be6798d234988c594cccdcb764ad222af9ad21af9",
"src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
@@ -71,12 +71,31 @@
"Original typed source inventory is not normalized; retain source identities and limits in the procedure/readsets."
],
"provenance_readsets": [],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "applicability": {
+ "scope": "Build, measure and diagnose one Ascend operator in isolation using msprof op, before interpreting system-level or multi-kernel performance.",
+ "prerequisites": "A built executable or test harness and access to real NPU hardware with msprof op for collection.",
+ "input_contract": "Operator executable/harness, collection settings and OPPROF output directory, plus the operator directory used to archive each measurement round.",
+ "produced_artifact": "Per-round raw CSV archives and observational min/average/max summaries, followed by a bottleneck interpretation and comparison of operator duration and core balance.",
+ "failure_modes": "Unwarmed or reduced-frequency runs distort timing; separately budgeting simultaneous GM reads/writes misses shared bandwidth, and tiny workloads can be dominated by launch overhead."
+ },
+ "recommendations": [
+ "Collect warmup and repeated measurements into a named round, preserve the raw CSVs, and distinguish the summary statistics from the subsequent bottleneck judgment.",
+ "Compare task duration, dominant pipeline ratios and per-core balance together; change one optimization point, rebuild as needed and archive the new round before deciding to keep it.",
+ "Check current versus rated frequency and shared read/write bandwidth before blaming the kernel; aggregate msprof ratios do not provide an instruction-by-instruction timeline."
+ ]
+ },
+ "applicability": {
+ "scope": "Build, measure and diagnose one Ascend operator in isolation using msprof op, before interpreting system-level or multi-kernel performance.",
+ "prerequisites": "A built executable or test harness and access to real NPU hardware with msprof op for collection.",
+ "input_contract": "Operator executable/harness, collection settings and OPPROF output directory, plus the operator directory used to archive each measurement round.",
+ "produced_artifact": "Per-round raw CSV archives and observational min/average/max summaries, followed by a bottleneck interpretation and comparison of operator duration and core balance.",
+ "failure_modes": "Unwarmed or reduced-frequency runs distort timing; separately budgeting simultaneous GM reads/writes misses shared bandwidth, and tiny workloads can be dominated by launch overhead."
},
- "applicability": {},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Collect warmup and repeated measurements into a named round, preserve the raw CSVs, and distinguish the summary statistics from the subsequent bottleneck judgment.",
+ "Compare task duration, dominant pipeline ratios and per-core balance together; change one optimization point, rebuild as needed and archive the new round before deciding to keep it.",
+ "Check current versus rated frequency and shared read/write bandwidth before blaming the kernel; aggregate msprof ratios do not provide an instruction-by-instruction timeline."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-tuning.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-tuning.json
index 5d00e5016..e0aa8ff4f 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-tuning.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-operator-tuning.json
@@ -24,8 +24,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-tuning/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-tuning/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-tuning/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-tuning/metadata.json",
@@ -41,7 +41,7 @@
"checkout_ref": "checkout:ascend-operator-tuning",
"source_files": {
"src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-tuning/SKILL.md": "38b151bc6b4e9734c6fa41efb7a30343559400c0c75bbdd26bd8a968b8ea4592",
- "src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-tuning/metadata.json": "04d2002fc8a0a7becf86b0b73daaf2da8716ab466d7f988da43d41a10f26680e",
+ "src/inferrsi/data/skill_library/hardware/ascend/ascend-operator-tuning/metadata.json": "776fe88247ddacd3cf245920c580a9ff8f2a80fbc0ad1fa7d8be6d63b15a66f7",
"src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
@@ -69,12 +69,33 @@
"Original typed source inventory is not normalized; retain source identities and limits in the procedure/readsets."
],
"provenance_readsets": [],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "applicability": {
+ "scope": "Run a bounded, accuracy-gated tuning campaign for a single Ascend operator already identified as the bottleneck, escalating from code-level to scheduling and instruction-level changes.",
+ "prerequisites": "A measured operator baseline, an explicit numeric target and a logged on-device accuracy pass before tuning; performance comparisons use real hardware.",
+ "input_contract": "Operator implementation, fixed shape/dtype/card and warmup policy, accuracy result, per-core work/wait measurements and round logs.",
+ "produced_artifact": "A bounded campaign record of each change, accuracy and measured timing delta, kept/reverted rounds, current stage and documented stop condition.",
+ "refusal_point": "Do not start without the baseline and accuracy gate; revert accuracy failures, and stop/escalate repeated failures, timeouts, exhausted stages or the campaign budget.",
+ "failure_modes": "Scheduler-idle and predecessor-wait bottlenecks need different fixes; output inplace-reshape can corrupt slice writes, and a changed analyzer log format can produce a misleading zero-core report."
+ },
+ "recommendations": [
+ "Change one parameter and recheck accuracy each round; keep the failure record when reverting and compare the final configuration against the same baseline workload and warmup policy.",
+ "Route scheduler bubbles separately from predecessor dependencies and enter instruction-level tuning only after isolating a specific long task.",
+ "Verify the analyzer parsed real cores before trusting its diagnosis, and rederive PyPTO tile/buffer/scheduling settings for the target SoC rather than copying numeric examples."
+ ]
+ },
+ "applicability": {
+ "scope": "Run a bounded, accuracy-gated tuning campaign for a single Ascend operator already identified as the bottleneck, escalating from code-level to scheduling and instruction-level changes.",
+ "prerequisites": "A measured operator baseline, an explicit numeric target and a logged on-device accuracy pass before tuning; performance comparisons use real hardware.",
+ "input_contract": "Operator implementation, fixed shape/dtype/card and warmup policy, accuracy result, per-core work/wait measurements and round logs.",
+ "produced_artifact": "A bounded campaign record of each change, accuracy and measured timing delta, kept/reverted rounds, current stage and documented stop condition.",
+ "refusal_point": "Do not start without the baseline and accuracy gate; revert accuracy failures, and stop/escalate repeated failures, timeouts, exhausted stages or the campaign budget.",
+ "failure_modes": "Scheduler-idle and predecessor-wait bottlenecks need different fixes; output inplace-reshape can corrupt slice writes, and a changed analyzer log format can produce a misleading zero-core report."
},
- "applicability": {},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Change one parameter and recheck accuracy each round; keep the failure record when reverting and compare the final configuration against the same baseline workload and warmup policy.",
+ "Route scheduler bubbles separately from predecessor dependencies and enter instruction-level tuning only after isolating a specific long task.",
+ "Verify the analyzer parsed real cores before trusting its diagnosis, and rederive PyPTO tile/buffer/scheduling settings for the target SoC rather than copying numeric examples."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-optimization-casebook.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-optimization-casebook.json
index 639dfd7d5..bc74ae8f6 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-optimization-casebook.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-optimization-casebook.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/hardware/ascend/ascend-optimization-casebook/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/hardware/ascend/ascend-optimization-casebook/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-optimization-casebook/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-optimization-casebook/metadata.json",
@@ -39,7 +39,7 @@
"checkout_ref": "checkout:ascend-optimization-casebook",
"source_files": {
"src/inferrsi/data/skill_library/hardware/ascend/ascend-optimization-casebook/SKILL.md": "647f5e9a288f2c4e1617ba53df59d9d827df749eb4fe86182d21a9c41175caf8",
- "src/inferrsi/data/skill_library/hardware/ascend/ascend-optimization-casebook/metadata.json": "f425f7c1abbd616387273e8932c6bd02f4b1da012f099286ad4171e0f6037b22",
+ "src/inferrsi/data/skill_library/hardware/ascend/ascend-optimization-casebook/metadata.json": "e5e49bf7b24fbc32ddaf1aa8ec14051ef6393533b12bb2c55fff2fb0da927541",
"src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
@@ -65,12 +65,29 @@
"Original typed source inventory is not normalized; retain source identities and limits in the procedure/readsets."
],
"provenance_readsets": [],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "applicability": {
+ "scope": "Find source-derived prior cases for an already classified Ascend inference bottleneck, including fallback, memory traffic, exposed communication, graph startup and host scheduling gaps.",
+ "prerequisites": "A bottleneck classification from profiling and a check of the selected case preconditions/mechanism before applying it.",
+ "input_contract": "Bottleneck symptoms and matching profiler evidence, model/hardware/CANN/configuration context and, for a new case, measured before/after metrics.",
+ "produced_artifact": "A matching prior-art case and evidence requirements for a later measured application; a new case record requires a verified win or clear failure with a metric comparison.",
+ "refusal_point": "Do not add an INCONCLUSIVE case or one without a measured metric comparison; a subjective direction is insufficient evidence."
+ },
+ "recommendations": [
+ "Classify first, then check the matching case preconditions and use its owning procedure to apply and measure the candidate; the casebook itself does not perform profiling.",
+ "Keep profiler mechanism evidence and same-window end-to-end outcomes together, and retain locally_verified=false until the applied case has actually run."
+ ]
+ },
+ "applicability": {
+ "scope": "Find source-derived prior cases for an already classified Ascend inference bottleneck, including fallback, memory traffic, exposed communication, graph startup and host scheduling gaps.",
+ "prerequisites": "A bottleneck classification from profiling and a check of the selected case preconditions/mechanism before applying it.",
+ "input_contract": "Bottleneck symptoms and matching profiler evidence, model/hardware/CANN/configuration context and, for a new case, measured before/after metrics.",
+ "produced_artifact": "A matching prior-art case and evidence requirements for a later measured application; a new case record requires a verified win or clear failure with a metric comparison.",
+ "refusal_point": "Do not add an INCONCLUSIVE case or one without a measured metric comparison; a subjective direction is insufficient evidence."
},
- "applicability": {},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Classify first, then check the matching case preconditions and use its owning procedure to apply and measure the candidate; the casebook itself does not perform profiling.",
+ "Keep profiler mechanism evidence and same-window end-to-end outcomes together, and retain locally_verified=false until the applied case has actually run."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-performance-analysis.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-performance-analysis.json
index 9ddb517d3..a03778265 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-performance-analysis.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-performance-analysis.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/hardware/ascend/ascend-performance-analysis/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/hardware/ascend/ascend-performance-analysis/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-performance-analysis/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-performance-analysis/metadata.json",
@@ -41,7 +41,7 @@
"checkout_ref": "checkout:ascend-performance-analysis",
"source_files": {
"src/inferrsi/data/skill_library/hardware/ascend/ascend-performance-analysis/SKILL.md": "a057b7f86a24f34a8f84abbbb4910d58d4e50ff47060e1c404600017e3febe89",
- "src/inferrsi/data/skill_library/hardware/ascend/ascend-performance-analysis/metadata.json": "9e705625c8308013c287576b0247da6f6116d817a4a535cbecff80abd6ea35ee",
+ "src/inferrsi/data/skill_library/hardware/ascend/ascend-performance-analysis/metadata.json": "f3fb35b14b5b1054a9192459862451149395368f4a75f430caef2f41ad050296",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-performance-analysis/references/cannbot-CANN-2.0-LICENSE.md": "89b1c114df62cfd183004f38d18d07f186436c3b37203dd69c3e84becf9d2282",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-performance-analysis/references/cannbot-stage-evidence.md": "f56e4125325c9c1b9235a1d64fd732861c7686444492422daff44ba8dd648be4",
"src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
@@ -102,12 +102,31 @@
"provenance_readsets": [
"references/cannbot-stage-evidence.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "applicability": {
+ "scope": "Interpret Ascend CANN profiling evidence from step breakdown through kernel/utilization/communication analysis, and assess execution/validator evidence in automated optimization-stage reports.",
+ "prerequisites": "Usable warm-state captures with representative steps, required files and non-null utilization columns; healthy cards and a consistent valid topology for multi-card comparisons.",
+ "input_contract": "CANN step/kernel/utilization/communication outputs with verified column semantics; for stage acceptance, the loaded candidate, baseline, quality-validator output and comparable measurements.",
+ "produced_artifact": "An evidence-qualified bottleneck classification and optimization direction; stage conclusions retain missing-execution/validator INCONCLUSIVE or observed gate-failure FAIL.",
+ "failure_modes": "Summing concurrent communication durations double-counts overlap; misidentified columns can drop communication, and a successful export can still lack usable utilization data."
+ },
+ "recommendations": [
+ "Start from the step-level computing/communication/overlap/free breakdown and stop deepening the analysis once the bottleneck is located; do not sum concurrent streams as wall time.",
+ "Inspect actual column headers and enum values on the installed CANN version, then verify files and non-null metrics before interpreting utilization heuristics.",
+ "Use same-window interleaved baseline/candidate measurements on shared machines; progress markers, hashes and source-derived thresholds do not substitute for quality and execution evidence."
+ ]
+ },
+ "applicability": {
+ "scope": "Interpret Ascend CANN profiling evidence from step breakdown through kernel/utilization/communication analysis, and assess execution/validator evidence in automated optimization-stage reports.",
+ "prerequisites": "Usable warm-state captures with representative steps, required files and non-null utilization columns; healthy cards and a consistent valid topology for multi-card comparisons.",
+ "input_contract": "CANN step/kernel/utilization/communication outputs with verified column semantics; for stage acceptance, the loaded candidate, baseline, quality-validator output and comparable measurements.",
+ "produced_artifact": "An evidence-qualified bottleneck classification and optimization direction; stage conclusions retain missing-execution/validator INCONCLUSIVE or observed gate-failure FAIL.",
+ "failure_modes": "Summing concurrent communication durations double-counts overlap; misidentified columns can drop communication, and a successful export can still lack usable utilization data."
},
- "applicability": {},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Start from the step-level computing/communication/overlap/free breakdown and stop deepening the analysis once the bottleneck is located; do not sum concurrent streams as wall time.",
+ "Inspect actual column headers and enum values on the installed CANN version, then verify files and non-null metrics before interpreting utilization heuristics.",
+ "Use same-window interleaved baseline/candidate measurements on shared machines; progress markers, hashes and source-derived thresholds do not substitute for quality and execution evidence."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-video-diffusion-optimization.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-video-diffusion-optimization.json
index 5863ada55..937d6f5d9 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-video-diffusion-optimization.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascend-video-diffusion-optimization.json
@@ -26,8 +26,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/hardware/ascend/ascend-video-diffusion-optimization/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/hardware/ascend/ascend-video-diffusion-optimization/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascend-video-diffusion-optimization/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-video-diffusion-optimization/metadata.json",
@@ -44,7 +44,7 @@
"checkout_ref": "checkout:ascend-video-diffusion-optimization",
"source_files": {
"src/inferrsi/data/skill_library/hardware/ascend/ascend-video-diffusion-optimization/SKILL.md": "53c21e64f0dd47f9510faef1f2c01e1283899787ba7ebfe70cdabd8e57a76f00",
- "src/inferrsi/data/skill_library/hardware/ascend/ascend-video-diffusion-optimization/metadata.json": "8c2b364298330751fd1252516159b16c27e90354e2a724782d2851e68e43cdb3",
+ "src/inferrsi/data/skill_library/hardware/ascend/ascend-video-diffusion-optimization/metadata.json": "35f725c9d9e0a8638b234da798232d2ae49a37b70f0b0360710bce333f9265ca",
"src/inferrsi/data/skill_library/hardware/ascend/ascend-video-diffusion-optimization/references/field-notes-video-diffusion.md": "dadac7b23462fdc325218c72ddf8a2ac553151761a5d3e45b854b53d2c8c6908",
"src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
@@ -78,7 +78,10 @@
"applicability": {
"scope": "Ascend NPU video diffusion inference and serving optimization",
"hardware": "Ascend NPU; the measure-change-verify workflow skeleton is backend-agnostic and lives in methods/video-diffusion-optimization, while the operator, layout and profiler references here are Ascend-specific",
- "prerequisites": "A reproducible benchmark, a frozen quality check, and numeric gates defined in the governing acceptance issue"
+ "prerequisites": "A reproducible benchmark, a frozen quality check, and numeric gates defined in the governing acceptance issue",
+ "input_contract": "Issue-scoped performance and quality gates, timing boundary, representative and held-out workloads, frozen model/runtime/input settings and an authorized optimization budget.",
+ "produced_artifact": "An experiment ledger and final report of the frozen goal, baseline/final samples, quality/resource checks, accepted or reverted changes and remaining gap; verdicts require the acceptance issue evidence.",
+ "failure_modes": "Packed-layout conversion from noncontiguous weights can corrupt outputs; missing side-stream lifetime/order handling or omitted layout-copy costs can invalidate an optimization comparison."
},
"limitations": [
"Ascend specialization of the backend-agnostic methods/video-diffusion-optimization skeleton; every operator, layout and profiler reference (TransData, FRACTAL_NZ, CANN) is Ascend-specific and must be re-derived per backend.",
@@ -95,16 +98,25 @@
"../../../methods/video-diffusion-optimization/scripts/evaluate_target.py",
"../../../methods/video-diffusion-optimization/assets/campaign.template.json"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "Follow the common goal-contract and measure-change-verify loop, including final restarted-artifact acceptance; an operator win does not establish end-to-end improvement.",
+ "Materialize contiguous weights before FRACTAL_NZ conversion when strides differ from physical order, and attribute TransData/copy costs before proposing fusion.",
+ "Use field notes only to form hypotheses; derive thresholds, chunking, precision and sparsity from the target model measurements, and preserve stream dependencies and tensor lifetimes."
+ ]
},
"applicability": {
"scope": "Ascend NPU video diffusion inference and serving optimization",
"hardware": "Ascend NPU; the measure-change-verify workflow skeleton is backend-agnostic and lives in methods/video-diffusion-optimization, while the operator, layout and profiler references here are Ascend-specific",
- "prerequisites": "A reproducible benchmark, a frozen quality check, and numeric gates defined in the governing acceptance issue"
+ "prerequisites": "A reproducible benchmark, a frozen quality check, and numeric gates defined in the governing acceptance issue",
+ "input_contract": "Issue-scoped performance and quality gates, timing boundary, representative and held-out workloads, frozen model/runtime/input settings and an authorized optimization budget.",
+ "produced_artifact": "An experiment ledger and final report of the frozen goal, baseline/final samples, quality/resource checks, accepted or reverted changes and remaining gap; verdicts require the acceptance issue evidence.",
+ "failure_modes": "Packed-layout conversion from noncontiguous weights can corrupt outputs; missing side-stream lifetime/order handling or omitted layout-copy costs can invalidate an optimization comparison."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Follow the common goal-contract and measure-change-verify loop, including final restarted-artifact acceptance; an operator win does not establish end-to-end improvement.",
+ "Materialize contiguous weights before FRACTAL_NZ conversion when strides differ from physical order, and attribute TransData/copy costs before proposing fusion.",
+ "Use field notes only to form hypotheses; derive thresholds, chunking, precision and sparsity from the target model measurements, and preserve stream dependencies and tensor lifetimes."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascendc-tiling-design.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascendc-tiling-design.json
index 53eca492c..2a0c44b1c 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascendc-tiling-design.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/ascendc-tiling-design.json
@@ -24,8 +24,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/hardware/ascend/ascendc-tiling-design/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/hardware/ascend/ascendc-tiling-design/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/ascendc-tiling-design/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/ascendc-tiling-design/metadata.json",
@@ -41,7 +41,7 @@
"checkout_ref": "checkout:ascendc-tiling-design",
"source_files": {
"src/inferrsi/data/skill_library/hardware/ascend/ascendc-tiling-design/SKILL.md": "93aa1b775ced304cebfe6c4dec0503ddb2260ecb3cb01edc17ce329446c5ce2b",
- "src/inferrsi/data/skill_library/hardware/ascend/ascendc-tiling-design/metadata.json": "f6d94f104cb70969024b2f4542078a240e200712c66279eca8b94f60bee1dd71",
+ "src/inferrsi/data/skill_library/hardware/ascend/ascendc-tiling-design/metadata.json": "03cfd09faa3edf42f9d08d0d8878fe5cb7c7a376abacf211dd8f500c698f3a8f",
"src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
@@ -69,12 +69,29 @@
"Original typed source inventory is not normalized; retain source identities and limits in the procedure/readsets."
],
"provenance_readsets": [],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "applicability": {
+ "scope": "Design or restructure AscendC tiling for the covered reduction, broadcast and elementwise classes, plus the bounded mxfp8 fusion and small-channel transpose cases.",
+ "input_contract": "Operator class and semantics, normalized shapes/axes, dtype, target chip generation, valid/aligned lengths and intermediate-buffer requirements.",
+ "produced_artifact": "A multi-core split, UB chunk formula, buffer-size/total-usage plan and branch matrix with edge tests; these are design outputs, not benchmark results.",
+ "failure_modes": "Budgeting only inputs omits intermediate UB storage; mixing valid and aligned lengths or copy units corrupts boundaries, and choosing unsupported generation-specific primitives invalidates the design."
+ },
+ "recommendations": [
+ "Normalize axes first, then decide full-load versus chunking from the entire live working set; derive tile size from buffer, axis, API and load-balance constraints.",
+ "Keep valid compute/copy lengths separate from aligned storage offsets and sizes, and complete dtype, tail/alignment and boundary branches before tuning.",
+ "Select primitives by the target generation and preserve class-specific coverage limits; re-evaluate sample constants and API semantics instead of treating formulas as tested performance evidence."
+ ]
+ },
+ "applicability": {
+ "scope": "Design or restructure AscendC tiling for the covered reduction, broadcast and elementwise classes, plus the bounded mxfp8 fusion and small-channel transpose cases.",
+ "input_contract": "Operator class and semantics, normalized shapes/axes, dtype, target chip generation, valid/aligned lengths and intermediate-buffer requirements.",
+ "produced_artifact": "A multi-core split, UB chunk formula, buffer-size/total-usage plan and branch matrix with edge tests; these are design outputs, not benchmark results.",
+ "failure_modes": "Budgeting only inputs omits intermediate UB storage; mixing valid and aligned lengths or copy units corrupts boundaries, and choosing unsupported generation-specific primitives invalidates the design."
},
- "applicability": {},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Normalize axes first, then decide full-load versus chunking from the entire live working set; derive tile size from buffer, axis, API and load-balance constraints.",
+ "Keep valid compute/copy lengths separate from aligned storage offsets and sizes, and complete dtype, tail/alignment and boundary branches before tuning.",
+ "Select primitives by the target generation and preserve class-specific coverage limits; re-evaluate sample constants and API semantics instead of treating formulas as tested performance evidence."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/cann-attribution-overlap.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/cann-attribution-overlap.json
index 92e1ceb1d..94e4d504a 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/cann-attribution-overlap.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/cann-attribution-overlap.json
@@ -24,8 +24,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/hardware/ascend/cann-attribution-overlap/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/hardware/ascend/cann-attribution-overlap/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/cann-attribution-overlap/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/cann-attribution-overlap/metadata.json",
@@ -41,7 +41,7 @@
"checkout_ref": "checkout:cann-attribution-overlap",
"source_files": {
"src/inferrsi/data/skill_library/hardware/ascend/cann-attribution-overlap/SKILL.md": "0e6f094dc554abea12e07def50f20b233b47127779a8aefad595b4ab1efd24c2",
- "src/inferrsi/data/skill_library/hardware/ascend/cann-attribution-overlap/metadata.json": "acccc8eb86de0df6caf78a9b3c6db76a2a686adc4a5ba0029a4d62d50ff9a8f8",
+ "src/inferrsi/data/skill_library/hardware/ascend/cann-attribution-overlap/metadata.json": "7ad137076afb55fbea74c6c5c0c0ca8eb3a8e615a765ec23199e8df75a8abd4f",
"src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
@@ -69,12 +69,33 @@
"Original typed source inventory is not normalized; retain source identities and limits in the procedure/readsets."
],
"provenance_readsets": [],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "applicability": {
+ "scope": "Attribute representative CANN trace kernels to source owners and plan bounded multistream overlap using data, state, event and collective-order dependencies.",
+ "prerequisites": "Pin the installed CANN/TorchNPU/TorchAir versions before choosing one execution-mode synchronization API family; communication proposals require communication evidence.",
+ "input_contract": "Representative-step selection and shapes, kernel/stream inventory with code_ref, adjacent wait windows and dependency edges; communication analysis also needs rank/transport evidence.",
+ "produced_artifact": "An attribution map, serial/candidate/pending dependency groups and a minimal overlap plan with joins, enable/fallback path, resource tradeoffs and acceptance checks.",
+ "failure_modes": "Kernel names and mechanical coverage can misidentify ownership; logical independence can hide resource contention, and overlap ratios or summed durations can hide unchanged wall time or new tails."
+ },
+ "recommendations": [
+ "Join hotspots to source owners and adjacent wait windows before selecting a target; keep compute cost, waiting and host gaps separate.",
+ "Build explicit data/state/event/collective-order edges, then verify one correct overlap point before increasing overlap or partitioning resources.",
+ "Require observed overlap, preserved state/order and fallback behavior plus latency and outside-module checks; report wall time, interval unions and duration sums separately.",
+ "Carry disputed formulas and source-revision distinctions forward; device free time alone does not identify Python, GIL or OS causes without matching host evidence."
+ ]
+ },
+ "applicability": {
+ "scope": "Attribute representative CANN trace kernels to source owners and plan bounded multistream overlap using data, state, event and collective-order dependencies.",
+ "prerequisites": "Pin the installed CANN/TorchNPU/TorchAir versions before choosing one execution-mode synchronization API family; communication proposals require communication evidence.",
+ "input_contract": "Representative-step selection and shapes, kernel/stream inventory with code_ref, adjacent wait windows and dependency edges; communication analysis also needs rank/transport evidence.",
+ "produced_artifact": "An attribution map, serial/candidate/pending dependency groups and a minimal overlap plan with joins, enable/fallback path, resource tradeoffs and acceptance checks.",
+ "failure_modes": "Kernel names and mechanical coverage can misidentify ownership; logical independence can hide resource contention, and overlap ratios or summed durations can hide unchanged wall time or new tails."
},
- "applicability": {},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Join hotspots to source owners and adjacent wait windows before selecting a target; keep compute cost, waiting and host gaps separate.",
+ "Build explicit data/state/event/collective-order edges, then verify one correct overlap point before increasing overlap or partitioning resources.",
+ "Require observed overlap, preserved state/order and fallback behavior plus latency and outside-module checks; report wall time, interval unions and duration sums separately.",
+ "Carry disputed formulas and source-revision distinctions forward; device free time alone does not identify Python, GIL or OS causes without matching host evidence."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torch-npu.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torch-npu.json
index b7b8dc41e..82dbdabf1 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torch-npu.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torch-npu.json
@@ -20,8 +20,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/frameworks/torch-npu/torch-npu/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/frameworks/torch-npu/torch-npu/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/frameworks/torch-npu/torch-npu/SKILL.md",
"src/inferrsi/data/skill_library/frameworks/torch-npu/torch-npu/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-device-memory-lifecycle.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-device-memory-lifecycle.json
index 9ac5c3784..69d216b0e 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-device-memory-lifecycle.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-device-memory-lifecycle.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/hardware/ascend/torchnpu-device-memory-lifecycle/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/hardware/ascend/torchnpu-device-memory-lifecycle/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/torchnpu-device-memory-lifecycle/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/torchnpu-device-memory-lifecycle/metadata.json",
@@ -40,7 +40,7 @@
"checkout_ref": "checkout:torchnpu-device-memory-lifecycle",
"source_files": {
"src/inferrsi/data/skill_library/hardware/ascend/torchnpu-device-memory-lifecycle/SKILL.md": "8428df7157af024fa17cfe0b3818ed54bd9f468306dda46eb8d4f0edf4f3f230",
- "src/inferrsi/data/skill_library/hardware/ascend/torchnpu-device-memory-lifecycle/metadata.json": "c650dce88c07b636d68e063dbda4d319f6f7f3ecbcdd5e47f393cbfc3af7c71d",
+ "src/inferrsi/data/skill_library/hardware/ascend/torchnpu-device-memory-lifecycle/metadata.json": "02de593e638e71c6debb763b6b0d5a5596e0b810bbe48450ead28dc43de83f99",
"src/inferrsi/data/skill_library/hardware/ascend/torchnpu-device-memory-lifecycle/references/source-notes.md": "80cba4d3a884e7ee2618326cf74b9f9c3a34e4d9cf443707f6bbe673f2d5ba65",
"src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
@@ -69,12 +69,29 @@
"provenance_readsets": [
"references/source-notes.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "applicability": {
+ "scope": "Diagnose TorchNPU device-HBM reuse, stream lifetimes, fragmentation and graph/private-pool retention before proposing allocator changes.",
+ "input_contract": "Allocation failure/size sequence, device and allocator path, installed versions/settings, stream and graph-pool timeline, retained tensor references and allocator statistics with synchronization boundaries.",
+ "produced_artifact": "A suspected allocator state transition with source evidence IDs, one eligible action, resource cost, counter-signals and a correctness-first verification/rollback plan.",
+ "refusal_point": "Missing evidence supports only a conditional proposal, not a diagnosed leak; stop when external/custom/shared allocation ownership is untraced and keep baseline settings for unsupported versions or live-tensor capacity pressure."
+ },
+ "recommendations": [
+ "Separate stream lifetime protection from producer/consumer ordering, and distinguish live tensors, pending events, inactive splits and private-pool retention before tuning.",
+ "Match split or expandable-segment proposals to the observed allocation pattern and actual feature gates; do not combine incompatible controls or import comment thresholds as defaults.",
+ "Treat cache release as synchronization/allocation work at a quiescent boundary; compare identical allocation sequences in fresh processes at the same synchronization points."
+ ]
+ },
+ "applicability": {
+ "scope": "Diagnose TorchNPU device-HBM reuse, stream lifetimes, fragmentation and graph/private-pool retention before proposing allocator changes.",
+ "input_contract": "Allocation failure/size sequence, device and allocator path, installed versions/settings, stream and graph-pool timeline, retained tensor references and allocator statistics with synchronization boundaries.",
+ "produced_artifact": "A suspected allocator state transition with source evidence IDs, one eligible action, resource cost, counter-signals and a correctness-first verification/rollback plan.",
+ "refusal_point": "Missing evidence supports only a conditional proposal, not a diagnosed leak; stop when external/custom/shared allocation ownership is untraced and keep baseline settings for unsupported versions or live-tensor capacity pressure."
},
- "applicability": {},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Separate stream lifetime protection from producer/consumer ordering, and distinguish live tensors, pending events, inactive splits and private-pool retention before tuning.",
+ "Match split or expandable-segment proposals to the observed allocation pattern and actual feature gates; do not combine incompatible controls or import comment thresholds as defaults.",
+ "Treat cache release as synchronization/allocation work at a quiescent boundary; compare identical allocation sequences in fresh processes at the same synchronization points."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-pinned-host-memory-lifecycle.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-pinned-host-memory-lifecycle.json
index 5f4f82ec1..a61864e58 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-pinned-host-memory-lifecycle.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/ascend/skills/torchnpu-pinned-host-memory-lifecycle.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/hardware/ascend/torchnpu-pinned-host-memory-lifecycle/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/hardware/ascend/torchnpu-pinned-host-memory-lifecycle/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/hardware/ascend/torchnpu-pinned-host-memory-lifecycle/SKILL.md",
"src/inferrsi/data/skill_library/hardware/ascend/torchnpu-pinned-host-memory-lifecycle/metadata.json",
@@ -40,7 +40,7 @@
"checkout_ref": "checkout:torchnpu-pinned-host-memory-lifecycle",
"source_files": {
"src/inferrsi/data/skill_library/hardware/ascend/torchnpu-pinned-host-memory-lifecycle/SKILL.md": "23e93ea7c98f99743509fdc04139d78ed2e7123c63e80b366c6beaba7eb70d0d",
- "src/inferrsi/data/skill_library/hardware/ascend/torchnpu-pinned-host-memory-lifecycle/metadata.json": "97836a38465106d816a9ace894a7b277d6d330f9accade8f8493ce6f0fb57ef9",
+ "src/inferrsi/data/skill_library/hardware/ascend/torchnpu-pinned-host-memory-lifecycle/metadata.json": "0fa51de079663043fd5785beffb153d989b1cac5833dbd2c99535f6b782a5e95",
"src/inferrsi/data/skill_library/hardware/ascend/torchnpu-pinned-host-memory-lifecycle/references/source-notes.md": "435193590b4a2ba32e2e4c6fa9d7e8f889528d412cbbb48e35ec5433f7dab384",
"src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
@@ -69,12 +69,29 @@
"provenance_readsets": [
"references/source-notes.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "applicability": {
+ "scope": "Diagnose TorchNPU pinned-host RAM allocation, asynchronous-copy ownership/readiness, reuse and path-specific cache controls.",
+ "input_contract": "Installed versions/settings, host pressure and allocation/reuse distribution, stream timeline, owned versus registered/pageable buffers, selected host allocator path and host statistics/allocation timings.",
+ "produced_artifact": "An ownership classification, source evidence IDs, one eligible allocator proposal, benefit/resource cost and rollback criteria for a later fresh-process comparison.",
+ "refusal_point": "Keep baseline settings for live-buffer growth, pageable copies, stable reuse, unverified versions or unsupported paths; stop tuning on event/runtime errors and escalate external ownership or gate conflicts."
+ },
+ "recommendations": [
+ "Do not mutate an H2D source or read a D2H destination before completion; non_blocking=True alone proves neither overlap nor external-buffer lifetime tracking.",
+ "Interpret host allocated/active bytes using the selected host path, and test round/cache controls only in supported default-path fresh processes; a per-buffer cache threshold is not a total-cache cap.",
+ "Keep host and device allocator options separate and use an explicit quiescent boundary for host cache release; cache clearing is not a copy-completion barrier."
+ ]
+ },
+ "applicability": {
+ "scope": "Diagnose TorchNPU pinned-host RAM allocation, asynchronous-copy ownership/readiness, reuse and path-specific cache controls.",
+ "input_contract": "Installed versions/settings, host pressure and allocation/reuse distribution, stream timeline, owned versus registered/pageable buffers, selected host allocator path and host statistics/allocation timings.",
+ "produced_artifact": "An ownership classification, source evidence IDs, one eligible allocator proposal, benefit/resource cost and rollback criteria for a later fresh-process comparison.",
+ "refusal_point": "Keep baseline settings for live-buffer growth, pageable copies, stable reuse, unverified versions or unsupported paths; stop tuning on event/runtime errors and escalate external ownership or gate conflicts."
},
- "applicability": {},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Do not mutate an H2D source or read a D2H destination before completion; non_blocking=True alone proves neither overlap nor external-buffer lifetime tracking.",
+ "Interpret host allocated/active bytes using the selected host path, and test round/cache controls only in supported default-path fresh processes; a per-buffer cache threshold is not a total-cache cap.",
+ "Keep host and device allocator options separate and use an explicit quiescent boundary for host cache release; cache clearing is not a copy-completion barrier."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-serving-preemption.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-serving-preemption.json
index 6469ebff5..ad0808449 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-serving-preemption.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-serving-preemption.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/methods/assess-serving-preemption/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/methods/assess-serving-preemption/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/methods/assess-serving-preemption/SKILL.md",
"src/inferrsi/data/skill_library/methods/assess-serving-preemption/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-split-backward-schedule.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-split-backward-schedule.json
index 816d251e3..da9ae7462 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-split-backward-schedule.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/assess-split-backward-schedule.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/methods/assess-split-backward-schedule/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/methods/assess-split-backward-schedule/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/methods/assess-split-backward-schedule/SKILL.md",
"src/inferrsi/data/skill_library/methods/assess-split-backward-schedule/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-fixed-workload-claim.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-fixed-workload-claim.json
index d719fe7fa..7ca5cad92 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-fixed-workload-claim.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-fixed-workload-claim.json
@@ -20,8 +20,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/methods/audit-fixed-workload-claim/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/methods/audit-fixed-workload-claim/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/methods/audit-fixed-workload-claim/SKILL.md",
"src/inferrsi/data/skill_library/methods/audit-fixed-workload-claim/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-metal-fasth3-semantics.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-metal-fasth3-semantics.json
index 04c533d97..6e4cc46a6 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-metal-fasth3-semantics.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-metal-fasth3-semantics.json
@@ -29,8 +29,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/LICENSE",
"src/inferrsi/data/skill_library/repositories/h3-metal/audit-h3-metal-fasth3-semantics/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-runtime-residency.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-runtime-residency.json
index b9973b506..3cab986c6 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-runtime-residency.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-h3-runtime-residency.json
@@ -30,8 +30,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/LICENSE",
"src/inferrsi/data/skill_library/repositories/h3-c/audit-h3-runtime-residency/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-llama-runtime-state.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-llama-runtime-state.json
index 047037bcb..5ddfce7b9 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-llama-runtime-state.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-llama-runtime-state.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/SKILL.md",
"src/inferrsi/data/skill_library/repositories/llama-cpp/audit-llama-runtime-state/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-recurrent-kv-finalization.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-recurrent-kv-finalization.json
index 1ec0620cf..df6dc6e37 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-recurrent-kv-finalization.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-recurrent-kv-finalization.json
@@ -26,8 +26,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/LICENSE",
"src/inferrsi/data/skill_library/repositories/vllm-rlt/audit-recurrent-kv-finalization/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sd-cpp-residency.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sd-cpp-residency.json
index 5d2b6cef5..1d8f61df4 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sd-cpp-residency.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sd-cpp-residency.json
@@ -25,8 +25,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/LICENSE",
"src/inferrsi/data/skill_library/repositories/stable-diffusion-cpp/audit-sd-cpp-residency/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sglang-diffusion-lifetimes.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sglang-diffusion-lifetimes.json
index 984809c22..f809d4a52 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sglang-diffusion-lifetimes.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-sglang-diffusion-lifetimes.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/LICENSE",
"src/inferrsi/data/skill_library/repositories/sglang/audit-sglang-diffusion-lifetimes/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-vllm-kv-lifetimes.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-vllm-kv-lifetimes.json
index 3199dbc0f..0f3f76d17 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-vllm-kv-lifetimes.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/audit-vllm-kv-lifetimes.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/SKILL.md",
"src/inferrsi/data/skill_library/repositories/vllm/audit-vllm-kv-lifetimes/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/bound-windowed-attention-replay.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/bound-windowed-attention-replay.json
index 7d41bea5d..bc36aa315 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/bound-windowed-attention-replay.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/bound-windowed-attention-replay.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/SKILL.md",
"src/inferrsi/data/skill_library/repositories/deepinfer/bound-windowed-attention-replay/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/budget-speculative-verification.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/budget-speculative-verification.json
index 85695fd40..96920f691 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/budget-speculative-verification.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/budget-speculative-verification.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/SKILL.md",
"src/inferrsi/data/skill_library/repositories/deepinfer/budget-speculative-verification/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/calibrate-serving-simulation.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/calibrate-serving-simulation.json
index 07b837586..465d050c7 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/calibrate-serving-simulation.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/calibrate-serving-simulation.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/methods/calibrate-serving-simulation/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/methods/calibrate-serving-simulation/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/methods/calibrate-serving-simulation/SKILL.md",
"src/inferrsi/data/skill_library/methods/calibrate-serving-simulation/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cgroup-resource-controls.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cgroup-resource-controls.json
index 8e05d49ee..8df5ed768 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cgroup-resource-controls.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cgroup-resource-controls.json
@@ -25,8 +25,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/hardware/host/cgroup-resource-controls/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/hardware/host/cgroup-resource-controls/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/hardware/host/cgroup-resource-controls/SKILL.md",
"src/inferrsi/data/skill_library/hardware/host/cgroup-resource-controls/metadata.json",
@@ -42,7 +42,7 @@
"checkout_ref": "checkout:cgroup-resource-controls",
"source_files": {
"src/inferrsi/data/skill_library/hardware/host/cgroup-resource-controls/SKILL.md": "fd1f5a7e6a55714730044155f5b38c1d700383d4e16c3e91963bbfbe1c773af5",
- "src/inferrsi/data/skill_library/hardware/host/cgroup-resource-controls/metadata.json": "19e40763fedc20990651bb989c32a7693306c5c54af46c3d4be77fd4f87b6c03",
+ "src/inferrsi/data/skill_library/hardware/host/cgroup-resource-controls/metadata.json": "155aa391e573fa5bd90a089f4e3d06f357795f5d89c8cf9260e3e8ec3636b243",
"src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
@@ -71,12 +71,33 @@
"Original typed source inventory is not normalized; retain source identities and limits in the procedure/readsets."
],
"provenance_readsets": [],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "applicability": {
+ "scope": "Audit Linux cgroup v2 CPU quota/shares, effective placement, memory pressure/protection and conditional block-device IO controls for serving or training workloads.",
+ "prerequisites": "Identify workload cgroup membership, kernel/controller support, ancestor limits and the authorized file owner; prepare previous values and stop conditions before a separately authorized trial.",
+ "input_contract": "Target PID and hierarchy/namespace view, requested/effective cpusets, time-aligned CPU/memory/IO counters and application/sibling observations; IO proposals require resolved block-backed traffic.",
+ "produced_artifact": "An evidence-qualified control review/proposal with capacity tradeoff, configuration provenance, rollback manifest and matched-window application/sibling checks.",
+ "failure_modes": "Ancestor constraints can invalidate requested placement; pressure movement alone is not application causality, and unsupported IO accounting or shared-inode attribution can misassign traffic."
+ },
+ "recommendations": [
+ "Separate own-quota throttling, ancestor limits and sibling contention before choosing a cpu.max or cpu.weight review.",
+ "Compare requested with effective CPU/node sets and budget cpuset.mems migration before proposing placement; align memory PSI with local versus descendant event deltas.",
+ "Treat increased headroom as a shared-capacity spend and verify application outcomes and sibling impact, not just counter improvements.",
+ "Resolve actual block-backed traffic and supported IO controls before proposing caps; preserve inaccessible-ancestor uncertainty and do not infer container write authority."
+ ]
+ },
+ "applicability": {
+ "scope": "Audit Linux cgroup v2 CPU quota/shares, effective placement, memory pressure/protection and conditional block-device IO controls for serving or training workloads.",
+ "prerequisites": "Identify workload cgroup membership, kernel/controller support, ancestor limits and the authorized file owner; prepare previous values and stop conditions before a separately authorized trial.",
+ "input_contract": "Target PID and hierarchy/namespace view, requested/effective cpusets, time-aligned CPU/memory/IO counters and application/sibling observations; IO proposals require resolved block-backed traffic.",
+ "produced_artifact": "An evidence-qualified control review/proposal with capacity tradeoff, configuration provenance, rollback manifest and matched-window application/sibling checks.",
+ "failure_modes": "Ancestor constraints can invalidate requested placement; pressure movement alone is not application causality, and unsupported IO accounting or shared-inode attribution can misassign traffic."
},
- "applicability": {},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Separate own-quota throttling, ancestor limits and sibling contention before choosing a cpu.max or cpu.weight review.",
+ "Compare requested with effective CPU/node sets and budget cpuset.mems migration before proposing placement; align memory PSI with local versus descendant event deltas.",
+ "Treat increased headroom as a shared-capacity spend and verify application outcomes and sibling impact, not just counter improvements.",
+ "Resolve actual block-backed traffic and supported IO controls before proposing caps; preserve inaccessible-ancestor uncertainty and do not infer container write authority."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/checkpoint-recovery-consistency.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/checkpoint-recovery-consistency.json
index 36d983000..f6591279e 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/checkpoint-recovery-consistency.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/checkpoint-recovery-consistency.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/methods/checkpoint-recovery-consistency/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/methods/checkpoint-recovery-consistency/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/methods/checkpoint-recovery-consistency/SKILL.md",
"src/inferrsi/data/skill_library/methods/checkpoint-recovery-consistency/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/choose-parallel-training-plan.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/choose-parallel-training-plan.json
index b1445e424..ab404a363 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/choose-parallel-training-plan.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/choose-parallel-training-plan.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/methods/choose-parallel-training-plan/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/methods/choose-parallel-training-plan/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/methods/choose-parallel-training-plan/SKILL.md",
"src/inferrsi/data/skill_library/methods/choose-parallel-training-plan/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cite-check-downstream-pr.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cite-check-downstream-pr.json
index 4a0e3343e..e6af78776 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cite-check-downstream-pr.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/cite-check-downstream-pr.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/workflows/cite-check-downstream-pr/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/workflows/cite-check-downstream-pr/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/skill-index.json",
"src/inferrsi/data/skill_library/workflows/cite-check-downstream-pr/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/claim-contract.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/claim-contract.json
index b2d17c5e1..c0c19ea67 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/claim-contract.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/claim-contract.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/workflows/claim-contract/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/workflows/claim-contract/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/skill-index.json",
"src/inferrsi/data/skill_library/workflows/claim-contract/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/comfy-ui.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/comfy-ui.json
index fb4006d68..965358804 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/comfy-ui.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/comfy-ui.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/repositories/comfy-ui/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/repositories/comfy-ui/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/repositories/comfy-ui/LICENSE",
"src/inferrsi/data/skill_library/repositories/comfy-ui/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/competitor-scan.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/competitor-scan.json
index ed41c713f..1599d7dfe 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/competitor-scan.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/competitor-scan.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/workflows/competitor-scan/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/workflows/competitor-scan/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/skill-index.json",
"src/inferrsi/data/skill_library/workflows/competitor-scan/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dcp-async-save-performance.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dcp-async-save-performance.json
index b0a6f8e10..ecd40cf06 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dcp-async-save-performance.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dcp-async-save-performance.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/SKILL.md",
"src/inferrsi/data/skill_library/frameworks/pytorch/dcp-async-save-performance/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dense-performance-evidence.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dense-performance-evidence.json
index 72bc12779..f5f7f9e6d 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dense-performance-evidence.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dense-performance-evidence.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/methods/dense-performance-evidence/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/methods/dense-performance-evidence/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/methods/dense-performance-evidence/SKILL.md",
"src/inferrsi/data/skill_library/methods/dense-performance-evidence/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diagnose-vllm-admission-pressure.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diagnose-vllm-admission-pressure.json
index 1a4341c68..f4cbdd57d 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diagnose-vllm-admission-pressure.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diagnose-vllm-admission-pressure.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/SKILL.md",
"src/inferrsi/data/skill_library/repositories/vllm/diagnose-vllm-admission-pressure/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diffusers-invocation-state.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diffusers-invocation-state.json
index f30022b8e..5a8863730 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diffusers-invocation-state.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/diffusers-invocation-state.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/SKILL.md",
"src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/draft-acceptance-issue.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/draft-acceptance-issue.json
index a5ace976e..ecc817755 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/draft-acceptance-issue.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/draft-acceptance-issue.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/workflows/draft-acceptance-issue/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/workflows/draft-acceptance-issue/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/skill-index.json",
"src/inferrsi/data/skill_library/workflows/draft-acceptance-issue/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dwarfstar.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dwarfstar.json
index 5e0da828f..00163960e 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dwarfstar.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/dwarfstar.json
@@ -25,8 +25,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/repositories/dwarfstar/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/repositories/dwarfstar/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/repositories/dwarfstar/LICENSE",
"src/inferrsi/data/skill_library/repositories/dwarfstar/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/evaluate-elastic-serving-placement.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/evaluate-elastic-serving-placement.json
index 0b6140435..9ba2db480 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/evaluate-elastic-serving-placement.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/evaluate-elastic-serving-placement.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/methods/evaluate-elastic-serving-placement/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/methods/evaluate-elastic-serving-placement/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/methods/evaluate-elastic-serving-placement/SKILL.md",
"src/inferrsi/data/skill_library/methods/evaluate-elastic-serving-placement/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/generation-cache-lifecycle.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/generation-cache-lifecycle.json
index 99de5ec47..a3299719e 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/generation-cache-lifecycle.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/generation-cache-lifecycle.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/SKILL.md",
"src/inferrsi/data/skill_library/repositories/mlx-lm/generation-cache-lifecycle/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-memory-management.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-memory-management.json
index a097c6a6a..bb514d1d8 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-memory-management.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-memory-management.json
@@ -30,8 +30,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/hardware/host/host-memory-management/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/hardware/host/host-memory-management/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/hardware/host/host-memory-management/SKILL.md",
"src/inferrsi/data/skill_library/hardware/host/host-memory-management/metadata.json",
@@ -56,7 +56,7 @@
"checkout_ref": "checkout:host-memory-management",
"source_files": {
"src/inferrsi/data/skill_library/hardware/host/host-memory-management/SKILL.md": "4c8db9f603cd581d002e8f41da65af9cbd03699be05c4965a55de47e044e4dd8",
- "src/inferrsi/data/skill_library/hardware/host/host-memory-management/metadata.json": "0d1fe7238bf3e744b463df5f7dc683110468ae266709e63f0d3edfaa0e2f110c",
+ "src/inferrsi/data/skill_library/hardware/host/host-memory-management/metadata.json": "9fe5988b136da9aeff9d00b0c6c607cac00b4071efaf637d287a350e5d4fcafb",
"src/inferrsi/data/skill_library/hardware/host/host-memory-management/references/cpython-PSF-LICENSE.md": "33f1a470702f459c160751bf3188b3fa7462fc761a006e03e80ef3f0a7a01ed6",
"src/inferrsi/data/skill_library/hardware/host/host-memory-management/references/cpython-free-threading-readset.json": "46336623b3c573fad97d4c686d9836baab807adf10f4bea05352d7a46236e004",
"src/inferrsi/data/skill_library/hardware/host/host-memory-management/references/legacy-source-notes.md": "bb1c0fb7a4d1a0301ca9e699e733224f3dcf0ded3cfe13e612e5c04d48a8cab5",
@@ -168,12 +168,29 @@
"references/cpython-PSF-LICENSE.md",
"references/legacy-source-notes.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "applicability": {
+ "scope": "Audit host preparation, pinned-host pressure and asynchronous CUDA-copy ownership/order/lifetime, with separately scoped routing for Ascend descriptor retention, Python preprocessing and NUMA placement.",
+ "input_contract": "Host buffer ownership and alias/mutation behavior, copy direction/dtype/layout, producer/consumer streams, live/in-flight/cache budget and the actual interpreter/placement context.",
+ "produced_artifact": "An ownership and dependency audit with bounded transfer candidates and a correctness-first validation plan separating enqueue, pinning, transfer, model and end-to-end cost.",
+ "failure_modes": "A stream wait does not prevent CPU mutation and record_stream does not create a data dependency; cached host storage may outlive Python scope, and the eager reusable-slot example is not capture-safe."
+ },
+ "recommendations": [
+ "Track copy-before-consumer order, CPU payload reuse and device storage lifetime as three independent obligations; wait for D2H completion before CPU reads.",
+ "Compare pageable, per-request pinning and bounded reuse on the real workload while budgeting all live and in-flight slots, cache and preprocessing memory.",
+ "Keep the eager staging-slot preconditions explicit and validate ordering/lifetime under stress before performance; a small passing sample does not establish a hazardous pattern is safe."
+ ]
+ },
+ "applicability": {
+ "scope": "Audit host preparation, pinned-host pressure and asynchronous CUDA-copy ownership/order/lifetime, with separately scoped routing for Ascend descriptor retention, Python preprocessing and NUMA placement.",
+ "input_contract": "Host buffer ownership and alias/mutation behavior, copy direction/dtype/layout, producer/consumer streams, live/in-flight/cache budget and the actual interpreter/placement context.",
+ "produced_artifact": "An ownership and dependency audit with bounded transfer candidates and a correctness-first validation plan separating enqueue, pinning, transfer, model and end-to-end cost.",
+ "failure_modes": "A stream wait does not prevent CPU mutation and record_stream does not create a data dependency; cached host storage may outlive Python scope, and the eager reusable-slot example is not capture-safe."
},
- "applicability": {},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Track copy-before-consumer order, CPU payload reuse and device storage lifetime as three independent obligations; wait for D2H completion before CPU reads.",
+ "Compare pageable, per-request pinning and bounded reuse on the real workload while budgeting all live and in-flight slots, cache and preprocessing memory.",
+ "Keep the eager staging-slot preconditions explicit and validate ordering/lifetime under stress before performance; a small passing sample does not establish a hazardous pattern is safe."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-overhead-optimization.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-overhead-optimization.json
index 329ff605d..f88703d25 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-overhead-optimization.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/host-overhead-optimization.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/hardware/host/host-overhead-optimization/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/hardware/host/host-overhead-optimization/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/hardware/host/host-overhead-optimization/SKILL.md",
"src/inferrsi/data/skill_library/hardware/host/host-overhead-optimization/metadata.json",
@@ -40,7 +40,7 @@
"checkout_ref": "checkout:host-overhead-optimization",
"source_files": {
"src/inferrsi/data/skill_library/hardware/host/host-overhead-optimization/SKILL.md": "36ef88826a32ac3147b869001004ded17a8d5a491e25bfe105005ba39ea9d1eb",
- "src/inferrsi/data/skill_library/hardware/host/host-overhead-optimization/metadata.json": "ae71c389dfdb1a19d267feecb4c6f1fbb4c9afad501ff6d8fab7e38b29bcc37a",
+ "src/inferrsi/data/skill_library/hardware/host/host-overhead-optimization/metadata.json": "db3c63ca77a46e647a806e094ae662504768a958d537630a4d7bcabde35d6da3",
"src/inferrsi/data/skill_library/hardware/host/host-overhead-optimization/references/deepinfer-zero-overhead-scheduling.md": "f452bf40132c9195d739f1c86b3aa18857138c7ed0f1861af5853d93b9955924",
"src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
@@ -67,12 +67,31 @@
"Original typed source inventory is not normalized; retain source identities and limits in the procedure/readsets."
],
"provenance_readsets": [],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "applicability": {
+ "scope": "Use line-level profiling to reduce exposed Python request-preparation, scheduling and execution-coordination overhead in a bounded serving optimization loop.",
+ "prerequisites": "Confirm host overhead is the limiting factor with the detection procedure before tuning; choose a representative workload and consistent CPU affinity for comparisons.",
+ "input_contract": "Detection/NVTX triage, qualified target function paths and line-profiler output, workload/affinity settings, related correctness tests and serving benchmark metrics.",
+ "produced_artifact": "Per-round hotspot/change/rollback and correctness records plus a final before/after benchmark report, failed attempts, remaining hotspots and scoped follow-up recommendations.",
+ "failure_modes": "Optimizing the default preparation function can miss a different regressing owner; caching/hoisting/preallocation can violate lifetime or invalidation, and a line-level speedup can leave serving throughput unchanged."
+ },
+ "recommendations": [
+ "Profile the actual top regressing function by absolute exposed time, drilling into dominant unprofiled subcalls instead of assuming _prepare_tp_inputs is the owner.",
+ "Prefer one small evidence-supported change with explicit ownership/lifetime/invalidation checks, run relevant correctness tests and revert failed or regressing candidates.",
+ "Re-profile identical workloads after each accepted change and compare hotspot time, total function time and serving metrics; reassess the next bottleneck and stay within the authorized budget."
+ ]
+ },
+ "applicability": {
+ "scope": "Use line-level profiling to reduce exposed Python request-preparation, scheduling and execution-coordination overhead in a bounded serving optimization loop.",
+ "prerequisites": "Confirm host overhead is the limiting factor with the detection procedure before tuning; choose a representative workload and consistent CPU affinity for comparisons.",
+ "input_contract": "Detection/NVTX triage, qualified target function paths and line-profiler output, workload/affinity settings, related correctness tests and serving benchmark metrics.",
+ "produced_artifact": "Per-round hotspot/change/rollback and correctness records plus a final before/after benchmark report, failed attempts, remaining hotspots and scoped follow-up recommendations.",
+ "failure_modes": "Optimizing the default preparation function can miss a different regressing owner; caching/hoisting/preallocation can violate lifetime or invalidation, and a line-level speedup can leave serving throughput unchanged."
},
- "applicability": {},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Profile the actual top regressing function by absolute exposed time, drilling into dominant unprofiled subcalls instead of assuming _prepare_tp_inputs is the owner.",
+ "Prefer one small evidence-supported change with explicit ownership/lifetime/invalidation checks, run relevant correctness tests and revert failed or regressing candidates.",
+ "Re-profile identical workloads after each accepted change and compare hotspot time, total function time and serving metrics; reassess the next bottleneck and stay within the authorized budget."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/husky.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/husky.json
index 37e502f7f..8ae51b6fa 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/husky.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/husky.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/repositories/husky/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/repositories/husky/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/repositories/husky/SKILL.md",
"src/inferrsi/data/skill_library/repositories/husky/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-benchmark-design.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-benchmark-design.json
index e8b219314..950b70b56 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-benchmark-design.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-benchmark-design.json
@@ -35,8 +35,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/methods/inference-benchmark-design/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/methods/inference-benchmark-design/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/methods/inference-benchmark-design/SKILL.md",
"src/inferrsi/data/skill_library/methods/inference-benchmark-design/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-optimization-casebook.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-optimization-casebook.json
index 08b281116..d85803a4f 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-optimization-casebook.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/inference-optimization-casebook.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/methods/inference-optimization-casebook/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/methods/inference-optimization-casebook/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/methods/inference-optimization-casebook/SKILL.md",
"src/inferrsi/data/skill_library/methods/inference-optimization-casebook/cases/admission-drain-occupancy.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kernel-writing-guide.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kernel-writing-guide.json
index 301c5980e..bcc6daa2f 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kernel-writing-guide.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kernel-writing-guide.json
@@ -39,8 +39,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/methods/kernel-writing-guide/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/methods/kernel-writing-guide/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/methods/kernel-writing-guide/SKILL.md",
"src/inferrsi/data/skill_library/methods/kernel-writing-guide/cases/causal-conv1d-dispatch-state.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kv-tier-management.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kv-tier-management.json
index 3abfdb434..f7d9b52c5 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kv-tier-management.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/kv-tier-management.json
@@ -24,8 +24,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/repositories/lmcache/kv-tier-management/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/repositories/lmcache/kv-tier-management/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/repositories/lmcache/kv-tier-management/SKILL.md",
"src/inferrsi/data/skill_library/repositories/lmcache/kv-tier-management/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lazy-evaluation-stream-lifetimes.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lazy-evaluation-stream-lifetimes.json
index 24f59676b..07875bd1b 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lazy-evaluation-stream-lifetimes.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lazy-evaluation-stream-lifetimes.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/SKILL.md",
"src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lily.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lily.json
index 3f384c085..2d9e49eb6 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lily.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lily.json
@@ -25,8 +25,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/repositories/lily/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/repositories/lily/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/repositories/lily/LICENSE",
"src/inferrsi/data/skill_library/repositories/lily/NOTICE.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lithos-metal.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lithos-metal.json
index bdf86884d..588b3f204 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lithos-metal.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/lithos-metal.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/repositories/lithos-metal/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/repositories/lithos-metal/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/repositories/lithos-metal/LICENSE",
"src/inferrsi/data/skill_library/repositories/lithos-metal/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/local-dream.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/local-dream.json
index 22a714f55..52de5913c 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/local-dream.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/local-dream.json
@@ -29,8 +29,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/repositories/local-dream/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/repositories/local-dream/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/repositories/local-dream/SKILL.md",
"src/inferrsi/data/skill_library/repositories/local-dream/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mistral-rs.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mistral-rs.json
index 3e21ec778..386ee1047 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mistral-rs.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mistral-rs.json
@@ -26,8 +26,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/repositories/mistral-rs/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/repositories/mistral-rs/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/repositories/mistral-rs/LICENSE",
"src/inferrsi/data/skill_library/repositories/mistral-rs/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlc-llm.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlc-llm.json
index 7032a96d5..3d7b73082 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlc-llm.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlc-llm.json
@@ -27,8 +27,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/repositories/mlc-llm/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/repositories/mlc-llm/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/repositories/mlc-llm/LICENSE",
"src/inferrsi/data/skill_library/repositories/mlc-llm/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlx-swift-lm.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlx-swift-lm.json
index cc8e664c1..bc510c7e9 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlx-swift-lm.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/mlx-swift-lm.json
@@ -27,8 +27,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/repositories/mlx-swift-lm/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/repositories/mlx-swift-lm/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/repositories/mlx-swift-lm/LICENSE",
"src/inferrsi/data/skill_library/repositories/mlx-swift-lm/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/npu-skill-factory.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/npu-skill-factory.json
index 9cb886cf4..b8ab6fb57 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/npu-skill-factory.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/npu-skill-factory.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/workflows/npu-skill-factory/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/workflows/npu-skill-factory/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/skill-index.json",
"src/inferrsi/data/skill_library/workflows/npu-skill-factory/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/optimization-search-policy.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/optimization-search-policy.json
index bc67998a1..14c16ee29 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/optimization-search-policy.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/optimization-search-policy.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/methods/optimization-search-policy/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/methods/optimization-search-policy/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/methods/optimization-search-policy/SKILL.md",
"src/inferrsi/data/skill_library/methods/optimization-search-policy/cases/sarathi-chunk-budget.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/os-process-placement.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/os-process-placement.json
index 9c2378939..36532f246 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/os-process-placement.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/os-process-placement.json
@@ -24,8 +24,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/hardware/host/os-process-placement/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/hardware/host/os-process-placement/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/hardware/host/os-process-placement/SKILL.md",
"src/inferrsi/data/skill_library/hardware/host/os-process-placement/metadata.json",
@@ -42,7 +42,7 @@
"checkout_ref": "checkout:os-process-placement",
"source_files": {
"src/inferrsi/data/skill_library/hardware/host/os-process-placement/SKILL.md": "2003492e4d89fde0443b180d74933e764bd1979435e109c8d26e20eb8f6e5053",
- "src/inferrsi/data/skill_library/hardware/host/os-process-placement/metadata.json": "ec02995c075f0a540921f57b4846f2cafe334eb90f44389f17c3c50457c3e9c5",
+ "src/inferrsi/data/skill_library/hardware/host/os-process-placement/metadata.json": "d53e53f724c7b6f261cf6c9b579afd20b2279cb1ffd0af89f8ce9f475046fa76",
"src/inferrsi/data/skill_library/hardware/host/os-process-placement/references/llama-numa-thread-readset.md": "748b254b3b043a5ddb0f480eb4a566854d27eedd7b111499863e9bf0a5a6951e",
"src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
@@ -92,12 +92,31 @@
"provenance_readsets": [
"references/llama-numa-thread-readset.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "applicability": {
+ "scope": "Prepare and record CPU/NUMA process placement, phase-specific thread settings, device contention and environment prerequisites before shared-host performance measurements.",
+ "prerequisites": "No outstanding hard environment errors: required devices/driver, planned device count and any container prerequisites must be present before running.",
+ "input_contract": "Current affinity/NUMA policy, accelerator locality, workload phase/thread controls, device occupancy, runtime environment and relevant container/resource limits.",
+ "produced_artifact": "A recorded placement/thread/environment configuration and preflight outcome, with errors requiring remediation and warnings retained for the measurement campaign.",
+ "failure_modes": "Changing affinity or thread settings between rounds confounds timing; combining independent affinity mechanisms is order-sensitive, and small driver memory usage can be mistaken for device occupation."
+ },
+ "recommendations": [
+ "Inspect affinity and NUMA policy before placement and hold them constant across comparisons; document and requalify a round when placement changes.",
+ "Choose one affinity-mask authority and distinguish generation, batch and frontend thread controls; source template thread counts are starting points rather than tuned constants.",
+ "Record device occupancy and structured ERROR/WARNING results before measuring, using the official environment setup for derived CANN paths."
+ ]
+ },
+ "applicability": {
+ "scope": "Prepare and record CPU/NUMA process placement, phase-specific thread settings, device contention and environment prerequisites before shared-host performance measurements.",
+ "prerequisites": "No outstanding hard environment errors: required devices/driver, planned device count and any container prerequisites must be present before running.",
+ "input_contract": "Current affinity/NUMA policy, accelerator locality, workload phase/thread controls, device occupancy, runtime environment and relevant container/resource limits.",
+ "produced_artifact": "A recorded placement/thread/environment configuration and preflight outcome, with errors requiring remediation and warnings retained for the measurement campaign.",
+ "failure_modes": "Changing affinity or thread settings between rounds confounds timing; combining independent affinity mechanisms is order-sensitive, and small driver memory usage can be mistaken for device occupation."
},
- "applicability": {},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Inspect affinity and NUMA policy before placement and hold them constant across comparisons; document and requalify a round when placement changes.",
+ "Choose one affinity-mask authority and distinguish generation, batch and frontend thread controls; source template thread counts are starting points rather than tuned constants.",
+ "Record device occupancy and structured ERROR/WARNING results before measuring, using the official environment setup for derived CANN paths."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/performance-ceiling-analysis.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/performance-ceiling-analysis.json
index ea5027a26..b1f1ed6dd 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/performance-ceiling-analysis.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/performance-ceiling-analysis.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/SKILL.md",
"src/inferrsi/data/skill_library/methods/performance-ceiling-analysis/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/plan-activation-residency.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/plan-activation-residency.json
index 04f7cd997..362ca17db 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/plan-activation-residency.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/plan-activation-residency.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/methods/plan-activation-residency/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/methods/plan-activation-residency/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/methods/plan-activation-residency/SKILL.md",
"src/inferrsi/data/skill_library/methods/plan-activation-residency/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/project-maya.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/project-maya.json
index 20cc1edd9..da792eebe 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/project-maya.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/project-maya.json
@@ -25,8 +25,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/repositories/project-maya/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/repositories/project-maya/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/repositories/project-maya/LICENSE",
"src/inferrsi/data/skill_library/repositories/project-maya/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/pytorch.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/pytorch.json
index e5bf6c045..ca695d1f6 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/pytorch.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/pytorch.json
@@ -24,8 +24,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/frameworks/pytorch/pytorch/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/frameworks/pytorch/pytorch/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/frameworks/pytorch/pytorch/LICENSE",
"src/inferrsi/data/skill_library/frameworks/pytorch/pytorch/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/repository-evolution-analysis.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/repository-evolution-analysis.json
index d7a9ffd12..1b8ae2489 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/repository-evolution-analysis.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/repository-evolution-analysis.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/methods/repository-evolution-analysis/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/methods/repository-evolution-analysis/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/methods/repository-evolution-analysis/SKILL.md",
"src/inferrsi/data/skill_library/methods/repository-evolution-analysis/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/rsi-evidence.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/rsi-evidence.json
index 55b99bea3..5498b2196 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/rsi-evidence.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/rsi-evidence.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/workflows/rsi-evidence/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/workflows/rsi-evidence/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/skill-index.json",
"src/inferrsi/data/skill_library/workflows/rsi-evidence/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/skill-evolution-playbook.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/skill-evolution-playbook.json
index a34a155e9..996e82073 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/skill-evolution-playbook.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/skill-evolution-playbook.json
@@ -21,8 +21,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/methods/skill-evolution-playbook/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/methods/skill-evolution-playbook/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/methods/skill-evolution-playbook/SKILL.md",
"src/inferrsi/data/skill_library/methods/skill-evolution-playbook/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/splash.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/splash.json
index f738604c4..8bc62af7f 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/splash.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/splash.json
@@ -25,8 +25,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/repositories/splash/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/repositories/splash/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/repositories/splash/LICENSE",
"src/inferrsi/data/skill_library/repositories/splash/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/storage-io-measurement.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/storage-io-measurement.json
index 8463b604c..6ed92a99b 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/storage-io-measurement.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/storage-io-measurement.json
@@ -24,8 +24,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/hardware/host/storage-io-measurement/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/hardware/host/storage-io-measurement/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/hardware/host/storage-io-measurement/SKILL.md",
"src/inferrsi/data/skill_library/hardware/host/storage-io-measurement/metadata.json",
@@ -41,7 +41,7 @@
"checkout_ref": "checkout:storage-io-measurement",
"source_files": {
"src/inferrsi/data/skill_library/hardware/host/storage-io-measurement/SKILL.md": "63b883c10a8d3e7d918abfc7175b72e2b401ff1e02697f94b8382ec1c8f369b7",
- "src/inferrsi/data/skill_library/hardware/host/storage-io-measurement/metadata.json": "0a41c7b00978332b54ee1e4a7c1ade88ecde3e961128f24d1b4f7708caff5ff5",
+ "src/inferrsi/data/skill_library/hardware/host/storage-io-measurement/metadata.json": "5017aa6c04b8cc45d8549b71bc283bfced8e7e1bdb29cf0b305a0447a3c12a9d",
"src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
@@ -69,12 +69,33 @@
"Original typed source inventory is not normalized; retain source identities and limits in the procedure/readsets."
],
"provenance_readsets": [],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "applicability": {
+ "scope": "Measure and attribute storage traffic for model loading or serving while separating device/process observations from application outcomes and cache/startup effects.",
+ "prerequisites": "Resolve backing devices and participating worker PIDs first; active fio benchmarks or cache manipulation require separate authorization, with bounded observation during already-authorized runs.",
+ "input_contract": "Observed loader access pattern, tool/OS/filesystem versions, file-to-device mapping, worker PIDs, phase timestamps, cache-state method and raw IO/job/application measurements.",
+ "produced_artifact": "A workload- and cache-qualified IO evidence record with units, achieved depth, errors/bytes/completion and device/process attribution, plus a separately measured application outcome where benefit is claimed.",
+ "refusal_point": "Reject unspecified or copied-example benchmark workloads; do not name a storage bottleneck from util/iowait alone or infer application gain from synthetic bandwidth."
+ },
+ "recommendations": [
+ "Match the fio read pattern to the observed loader and verify achieved rather than requested depth; more queued work can increase latency.",
+ "Record warm/cold/unknown cache state and its method, separate startup from repeated/steady reads, and check completed bytes for truncated runs.",
+ "Align per-device and per-process observations with the application phase, retain units and latency provenance, and corroborate saturation using traffic, queues and process activity.",
+ "Only run a separately authorized bounded read-only job against an explicit existing regular file, then remeasure the real application before accepting a benefit."
+ ]
+ },
+ "applicability": {
+ "scope": "Measure and attribute storage traffic for model loading or serving while separating device/process observations from application outcomes and cache/startup effects.",
+ "prerequisites": "Resolve backing devices and participating worker PIDs first; active fio benchmarks or cache manipulation require separate authorization, with bounded observation during already-authorized runs.",
+ "input_contract": "Observed loader access pattern, tool/OS/filesystem versions, file-to-device mapping, worker PIDs, phase timestamps, cache-state method and raw IO/job/application measurements.",
+ "produced_artifact": "A workload- and cache-qualified IO evidence record with units, achieved depth, errors/bytes/completion and device/process attribution, plus a separately measured application outcome where benefit is claimed.",
+ "refusal_point": "Reject unspecified or copied-example benchmark workloads; do not name a storage bottleneck from util/iowait alone or infer application gain from synthetic bandwidth."
},
- "applicability": {},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Match the fio read pattern to the observed loader and verify achieved rather than requested depth; more queued work can increase latency.",
+ "Record warm/cold/unknown cache state and its method, separate startup from repeated/steady reads, and check completed bytes for truncated runs.",
+ "Align per-device and per-process observations with the application phase, retain units and latency provenance, and corroborate saturation using traffic, queues and process activity.",
+ "Only run a separately authorized bounded read-only job against an explicit existing regular file, then remeasure the real application before accepting a benefit."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-input-pipeline.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-input-pipeline.json
index 732366ace..72be122d4 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-input-pipeline.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-input-pipeline.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/methods/training-input-pipeline/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/methods/training-input-pipeline/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/methods/training-input-pipeline/SKILL.md",
"src/inferrsi/data/skill_library/methods/training-input-pipeline/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-resource-budget.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-resource-budget.json
index bf058278c..a2a84a989 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-resource-budget.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/training-resource-budget.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/methods/training-resource-budget/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/methods/training-resource-budget/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/methods/training-resource-budget/SKILL.md",
"src/inferrsi/data/skill_library/methods/training-resource-budget/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/video-diffusion-optimization.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/video-diffusion-optimization.json
index 05cc3a848..7772db026 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/video-diffusion-optimization.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/video-diffusion-optimization.json
@@ -26,8 +26,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/methods/video-diffusion-optimization/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/methods/video-diffusion-optimization/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/methods/video-diffusion-optimization/SKILL.md",
"src/inferrsi/data/skill_library/methods/video-diffusion-optimization/assets/campaign.template.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/vllm-metal.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/vllm-metal.json
index 4ed28de43..30c835b77 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/vllm-metal.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/vllm-metal.json
@@ -26,8 +26,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/repositories/vllm-metal/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/repositories/vllm-metal/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/repositories/vllm-metal/LICENSE",
"src/inferrsi/data/skill_library/repositories/vllm-metal/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/weight-loading-optimization.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/weight-loading-optimization.json
index b61f018de..69a7ce019 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/weight-loading-optimization.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/common/skills/weight-loading-optimization.json
@@ -29,8 +29,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/methods/weight-loading-optimization/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/methods/weight-loading-optimization/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/methods/weight-loading-optimization/SKILL.md",
"src/inferrsi/data/skill_library/methods/weight-loading-optimization/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-afd-p2p-exchanges.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-afd-p2p-exchanges.json
index ae9d02b96..bcd6e2eed 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-afd-p2p-exchanges.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-afd-p2p-exchanges.json
@@ -23,8 +23,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/SKILL.md",
"src/inferrsi/data/skill_library/repositories/afd-plugin/audit-afd-p2p-exchanges/metadata.json",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-flare-session-lifecycle.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-flare-session-lifecycle.json
index e0e77dc79..d203a8627 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-flare-session-lifecycle.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/audit-flare-session-lifecycle.json
@@ -29,8 +29,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/LICENSE",
"src/inferrsi/data/skill_library/repositories/flare-local-inference/audit-flare-session-lifecycle/SKILL.md",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/diagnose-attention-bottlenecks.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/diagnose-attention-bottlenecks.json
index 7d4d08677..9804a2927 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/diagnose-attention-bottlenecks.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/diagnose-attention-bottlenecks.json
@@ -16,8 +16,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/hardware/cuda/diagnose-attention-bottlenecks/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/hardware/cuda/diagnose-attention-bottlenecks/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/hardware/cuda/diagnose-attention-bottlenecks/SKILL.md",
"src/inferrsi/data/skill_library/hardware/cuda/diagnose-attention-bottlenecks/metadata.json",
@@ -42,7 +42,7 @@
"checkout_ref": "checkout:diagnose-attention-bottlenecks",
"source_files": {
"src/inferrsi/data/skill_library/hardware/cuda/diagnose-attention-bottlenecks/SKILL.md": "a25738dbb3b9daa15c5827cb536a20fe13f7c8b9dd14aa3fb7cfa7e132879431",
- "src/inferrsi/data/skill_library/hardware/cuda/diagnose-attention-bottlenecks/metadata.json": "3abe626a6e4c73c943429f609ce2c62aa77aae7e1726475ebbbdfbe1794f6f7d",
+ "src/inferrsi/data/skill_library/hardware/cuda/diagnose-attention-bottlenecks/metadata.json": "6239c885aa83936a7bd9c49643b1c79b3734a17234af7c03d95839efc622c79e",
"src/inferrsi/data/skill_library/hardware/cuda/diagnose-attention-bottlenecks/references/mechanisms.md": "fe08b2292fa587be3874774ddf41063b84c579a0085df422ec6710d8aa90b8b7",
"src/inferrsi/data/skill_library/hardware/cuda/diagnose-attention-bottlenecks/references/papers-20261015/INTEGRATION_NOTES.md": "371b01ee201ef51fecbfd6da892d517aaba295930b9395025f03f34e21c030a1",
"src/inferrsi/data/skill_library/hardware/cuda/diagnose-attention-bottlenecks/references/papers-20261015/NEXT_GATE_LEADS.md": "35b8b58ae5db02b2df51ccbe5cdf870a7383d17f345f9ee5eb7b0a4146dfc74b",
@@ -81,12 +81,29 @@
"references/papers-20261015/README.md",
"references/sources.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "applicability": {
+ "scope": "Diagnose attention-kernel bottlenecks after IO reduction, tile/pipeline regressions and precision or hardware-port assumptions.",
+ "input_contract": "Actual kernel/device revisions, attention shapes and semantics, operand/accumulator types, correctness/quality criteria, and existing traffic, resource, parallelism and stall observations.",
+ "produced_artifact": "A bottleneck diagnosis with confidence, one bounded candidate, resource exchange, counter-signal, missing evidence, source ID and next authorization gate.",
+ "refusal_point": "Return conditional hypotheses when observations are missing; reject or defer changes with wrong outputs/gradients, unsafe lifetimes/order, unsupported primitives, unacceptable accuracy or gains erased by omitted work."
+ },
+ "recommendations": [
+ "Preserve masks, dropout replay, online-softmax normalization and gradient obligations before comparing costs; exact algebra does not imply bitwise equality.",
+ "After fusion, reclassify the exposed limit using scalar work, independent output tiles and resource ceilings; larger tiles and deeper pipelines can reduce residency or spill.",
+ "Include precision preprocessing/conversion in the measurement boundary and check generated asynchronous scheduling; validate accelerator-specific primitives separately before a port."
+ ]
+ },
+ "applicability": {
+ "scope": "Diagnose attention-kernel bottlenecks after IO reduction, tile/pipeline regressions and precision or hardware-port assumptions.",
+ "input_contract": "Actual kernel/device revisions, attention shapes and semantics, operand/accumulator types, correctness/quality criteria, and existing traffic, resource, parallelism and stall observations.",
+ "produced_artifact": "A bottleneck diagnosis with confidence, one bounded candidate, resource exchange, counter-signal, missing evidence, source ID and next authorization gate.",
+ "refusal_point": "Return conditional hypotheses when observations are missing; reject or defer changes with wrong outputs/gradients, unsafe lifetimes/order, unsupported primitives, unacceptable accuracy or gains erased by omitted work."
},
- "applicability": {},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Preserve masks, dropout replay, online-softmax normalization and gradient obligations before comparing costs; exact algebra does not imply bitwise equality.",
+ "After fusion, reclassify the exposed limit using scalar work, independent output tiles and resource ceilings; larger tiles and deeper pipelines can reduce residency or spill.",
+ "Include precision preprocessing/conversion in the measurement boundary and check generated asynchronous scheduling; validate accelerator-specific primitives separately before a port."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/host-overhead-profiling.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/host-overhead-profiling.json
index 92ef2edb8..a36f7f2f2 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/host-overhead-profiling.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/host-overhead-profiling.json
@@ -22,8 +22,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/hardware/cuda/host-overhead-profiling/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/hardware/cuda/host-overhead-profiling/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/hardware/cuda/host-overhead-profiling/SKILL.md",
"src/inferrsi/data/skill_library/hardware/cuda/host-overhead-profiling/metadata.json",
@@ -39,7 +39,7 @@
"checkout_ref": "checkout:host-overhead-profiling",
"source_files": {
"src/inferrsi/data/skill_library/hardware/cuda/host-overhead-profiling/SKILL.md": "326e566fbe30d3ba04e4b50c3e87aad9df037f0ac4ad6fe129165ae97dc5f91d",
- "src/inferrsi/data/skill_library/hardware/cuda/host-overhead-profiling/metadata.json": "76e34c1c64aba473a9e126b032a776417caa0f913f6ac3cf4dba3a5d123b1ee0",
+ "src/inferrsi/data/skill_library/hardware/cuda/host-overhead-profiling/metadata.json": "bc4236c65e37378f4a74e0b88c47776055c186aca44c000b1f68e8c63cc4b277",
"src/inferrsi/data/skill_library/skill-index.json": "9ea8b2ef190777e683a3905f0614612a37782be4d655d230c000188b0c7073db"
},
"skill_metadata": {
@@ -65,12 +65,31 @@
"Original typed source inventory is not normalized; retain source identities and limits in the procedure/readsets."
],
"provenance_readsets": [],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "applicability": {
+ "scope": "Detect exposed host overhead and diagnose per-step host regressions in CUDA inference serving from existing nsys timelines.",
+ "prerequisites": "An existing serving nsys trace and Python with sqlite3; comparative root-cause analysis needs baseline and target traces, with nsys export available for report-format input.",
+ "input_contract": "One trace for detection/breakdown or two comparable traces for regression analysis, including the workload/phase context needed to isolate steady-state forward steps.",
+ "produced_artifact": "A detection verdict with metric/phase evidence, per-step NVTX/kernel/API breakdown and ranked host-operation deltas for a targeted optimization handoff.",
+ "failure_modes": "Warmup/loading or mismatched phase mixtures distort per-step comparisons; duplicated rank ranges, unresolved NVTX text IDs or overlapping rank end times can miscompute host gaps."
+ },
+ "recommendations": [
+ "Isolate comparable steady-state forward iterations and inspect context/generation separately; aggregate utilization can hide a phase-specific host bottleneck.",
+ "Confirm exposed preparation using both performance impact and idle attribution, retain communication as a confounder, and triage absolute NVTX time deltas before line-level optimization.",
+ "Resolve NVTX text IDs, account for TP rank duplication and use the maximum overlapping end time for gaps; do not compare full-trace totals as per-step overhead."
+ ]
+ },
+ "applicability": {
+ "scope": "Detect exposed host overhead and diagnose per-step host regressions in CUDA inference serving from existing nsys timelines.",
+ "prerequisites": "An existing serving nsys trace and Python with sqlite3; comparative root-cause analysis needs baseline and target traces, with nsys export available for report-format input.",
+ "input_contract": "One trace for detection/breakdown or two comparable traces for regression analysis, including the workload/phase context needed to isolate steady-state forward steps.",
+ "produced_artifact": "A detection verdict with metric/phase evidence, per-step NVTX/kernel/API breakdown and ranked host-operation deltas for a targeted optimization handoff.",
+ "failure_modes": "Warmup/loading or mismatched phase mixtures distort per-step comparisons; duplicated rank ranges, unresolved NVTX text IDs or overlapping rank end times can miscompute host gaps."
},
- "applicability": {},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "Isolate comparable steady-state forward iterations and inspect context/generation separately; aggregate utilization can hide a phase-specific host bottleneck.",
+ "Confirm exposed preparation using both performance impact and idle attribution, retain communication as a confounder, and triage absolute NVTX time deltas before line-level optimization.",
+ "Resolve NVTX text IDs, account for TP rank duplication and use the maximum overlapping end time for gaps; do not compare full-trace totals as per-step overhead."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",
diff --git a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/reason-hopper-kernels.json b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/reason-hopper-kernels.json
index 6bfdb950e..6ff557bab 100644
--- a/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/reason-hopper-kernels.json
+++ b/src/inferrsi/data/repo_packs/inferrsi/knowledge/cuda/skills/reason-hopper-kernels.json
@@ -36,8 +36,8 @@
"sources": [
{
"type": "git",
- "url": "https://github.com/InferMatrixLab/InferRSI/blob/ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90/src/inferrsi/data/skill_library/hardware/cuda/reason-hopper-kernels/SKILL.md",
- "commit": "ca32733bfbbaf2ff32bca3e7e4fbad41fe588a90",
+ "url": "https://github.com/InferMatrixLab/InferRSI/blob/1bada78556d695a01cf2bc3b0c117f65855cc177/src/inferrsi/data/skill_library/hardware/cuda/reason-hopper-kernels/SKILL.md",
+ "commit": "1bada78556d695a01cf2bc3b0c117f65855cc177",
"paths": [
"src/inferrsi/data/skill_library/hardware/cuda/reason-hopper-kernels/SKILL.md",
"src/inferrsi/data/skill_library/hardware/cuda/reason-hopper-kernels/metadata.json",
@@ -62,7 +62,7 @@
"checkout_ref": "checkout:reason-hopper-kernels",
"source_files": {
"src/inferrsi/data/skill_library/hardware/cuda/reason-hopper-kernels/SKILL.md": "cec4e3aa1eb64c31c54b5f5ca40af2b61b7919d94be4174968581d5f752c0538",
- "src/inferrsi/data/skill_library/hardware/cuda/reason-hopper-kernels/metadata.json": "7520da1cc97412631a600957ecb18aa5a7a4461ea21e76c237330180193ed898",
+ "src/inferrsi/data/skill_library/hardware/cuda/reason-hopper-kernels/metadata.json": "79ca64dc97ad34331036dc071e86954c98ac568e651adfae74cce74f1ca61c7b",
"src/inferrsi/data/skill_library/hardware/cuda/reason-hopper-kernels/references/fp8.md": "31d9c211c11c2224007abbdcd4059913c4a54833fe2d2f926411c5238fcd3501",
"src/inferrsi/data/skill_library/hardware/cuda/reason-hopper-kernels/references/licenses/cutlass-bsd.md": "42fec630f410aa308f70a51a89fadcd19586fa620f9831a32bee528a9a10000e",
"src/inferrsi/data/skill_library/hardware/cuda/reason-hopper-kernels/references/licenses/deepgemm-mit.md": "6c73a30c14a2938abaa6afe98ebb19e6d81dac5bd4ace0cb05c61c20afa37b85",
@@ -238,7 +238,10 @@
"CLC",
"two-SM MMA",
"native NVFP4/block-scaled tcgen05 MMA"
- ]
+ ],
+ "input_contract": "Device/compute capability, compiler target and selected implementation, kernel/library revisions, shapes/strides/alignment, numeric and grouping/epilogue semantics, correctness criterion and measurement boundary.",
+ "produced_artifact": "A source-qualified, reviewable hypothesis with one bounded change, resource/accuracy exchange, counter-signal, stop condition and next verification step.",
+ "failure_modes": "SM90 tags or import/dispatch guards can hide an unbuilt or unselected SM90a instruction path; extra stages or register caps can increase spills or lower residency."
},
"limitations": [
"Native compilation, GPU correctness, numerical accuracy, profiling, performance, model/service behavior and skill efficacy were not run.",
@@ -256,7 +259,12 @@
"../diagnose-attention-bottlenecks/SKILL.md",
"../../../methods/kernel-writing-guide/SKILL.md"
],
- "consumption_mode": "packaged"
+ "consumption_mode": "packaged",
+ "recommendations": [
+ "For WGMMA advice, confirm the compiled and selected SM90a path; do not transplant Blackwell-only instruction or memory mechanisms.",
+ "Resolve producer/consumer ownership and completion before tuning stage depth; budget live accumulators, padded shared memory, scales and epilogue storage together.",
+ "For later validation, check tails, short K, stage wraparound and grouped offsets against the frozen reference, inspect actual PTX/SASS resources, and include setup/conversion/reduction costs."
+ ]
},
"applicability": {
"scope": "Source-derived Hopper SM90/SM90a kernel reasoning for TMA/mbarrier, WGMMA, resource/layout/pipeline/software scheduling and selected FP8 GEMM. CUDA categorization does not certify other GPU architectures, library dispatch, or a combined tested stack.",
@@ -270,11 +278,15 @@
"CLC",
"two-SM MMA",
"native NVFP4/block-scaled tcgen05 MMA"
- ]
+ ],
+ "input_contract": "Device/compute capability, compiler target and selected implementation, kernel/library revisions, shapes/strides/alignment, numeric and grouping/epilogue semantics, correctness criterion and measurement boundary.",
+ "produced_artifact": "A source-qualified, reviewable hypothesis with one bounded change, resource/accuracy exchange, counter-signal, stop condition and next verification step.",
+ "failure_modes": "SM90 tags or import/dispatch guards can hide an unbuilt or unselected SM90a instruction path; extra stages or register caps can increase spills or lower residency."
},
"recommendations": [
- "Read the pinned SKILL.md and its prerequisites before choosing this procedure.",
- "Read the packaged references and owner-local cases alongside the procedure."
+ "For WGMMA advice, confirm the compiled and selected SM90a path; do not transplant Blackwell-only instruction or memory mechanisms.",
+ "Resolve producer/consumer ownership and completion before tuning stage depth; budget live accumulators, padded shared memory, scales and epilogue storage together.",
+ "For later validation, check tails, short K, stage wraparound and grouped offsets against the frozen reference, inspect actual PTX/SASS resources, and include setup/conversion/reduction costs."
],
"limitations": [
"This generated abstract points to the complete packaged skill; external evidence and upstream repositories remain source links.",