docs: describe conditions and advice for 22 hardware skills

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: [8115da15b9ef34978de500b08aa9f14ea790c35b](https://github.com/InferMatrixLab/InferRSI/tree/8115da15b9ef34978de500b08aa9f14ea790c35b/src/inferrsi/data/skill_library).
+Source snapshot: [fb4171e87b4c78d4d62c927f297374ed917949f5](https://github.com/InferMatrixLab/InferRSI/tree/fb4171e87b4c78d4d62c927f297374ed917949f5/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 <committed-source-revision>
-python tools/build_skill_index.py --source-revision 8115da15b9ef34978de500b08aa9f14ea790c35b --check
+python tools/build_skill_index.py --source-revision fb4171e87b4c78d4d62c927f297374ed917949f5 --check
 ```
 
 `--check` detects missing, stale and unexpected generated records without writing.

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/8115da15b9ef34978de500b08aa9f14ea790c35b/src/inferrsi/data/skill_library/hardware/ascend/ascend-aclgraph-capture-replay/SKILL.md",
-      "commit": "8115da15b9ef34978de500b08aa9f14ea790c35b",
+      "url": "https://github.com/InferMatrixLab/InferRSI/blob/fb4171e87b4c78d4d62c927f297374ed917949f5/src/inferrsi/data/skill_library/hardware/ascend/ascend-aclgraph-capture-replay/SKILL.md",
+      "commit": "fb4171e87b4c78d4d62c927f297374ed917949f5",
       "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.",

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/8115da15b9ef34978de500b08aa9f14ea790c35b/src/inferrsi/data/skill_library/hardware/ascend/ascend-device-health/SKILL.md",
-      "commit": "8115da15b9ef34978de500b08aa9f14ea790c35b",
+      "url": "https://github.com/InferMatrixLab/InferRSI/blob/fb4171e87b4c78d4d62c927f297374ed917949f5/src/inferrsi/data/skill_library/hardware/ascend/ascend-device-health/SKILL.md",
+      "commit": "fb4171e87b4c78d4d62c927f297374ed917949f5",
       "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.",

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/8115da15b9ef34978de500b08aa9f14ea790c35b/src/inferrsi/data/skill_library/hardware/ascend/ascend-distributed-parallel/SKILL.md",
-      "commit": "8115da15b9ef34978de500b08aa9f14ea790c35b",
+      "url": "https://github.com/InferMatrixLab/InferRSI/blob/fb4171e87b4c78d4d62c927f297374ed917949f5/src/inferrsi/data/skill_library/hardware/ascend/ascend-distributed-parallel/SKILL.md",
+      "commit": "fb4171e87b4c78d4d62c927f297374ed917949f5",
       "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.",