docs(plans): refresh plan index and status for 0.51.0

.memory/aiden-cli-standalone.md

@@ -38,3 +38,5 @@
 - Merging #71 (Linux desktop) took main's native C ports and `native-c-build-core.mjs` build scripts. The PR's OpenSSL/renameat2 shims were dropped because `native/shared/aiden-platform.h` carries its own SHA-256. `nativeHelperSourceHash` now also hashes `native/shared/*.h`, so a shared-header edit marks prebuilts stale. The PR's speech core carries main's "desktop" wording.
 - Merging #246 (root pi 0.87.1) bumped the CLI to `@earendil-works/pi-coding-agent` 0.87.1 and added a direct `@earendil-works/chord` 0.87.1 dependency. btw wraps its context in `normalizeContext`, and the child beforeTool context drops `systemPrompt`. The re-vendored advisor uses `normalizeContext`. The build's external check falls back to ESM resolution for chord's import-only subpaths and allowlists the optional `kerberos`. Session import accepts both `v: 4` (storage v1) and `version: 4` (0.84.4) journal headers.
 - Merging #251 (main's Remote contract revision 14) renumbered this PR's `GET /scheduled-tasks/notifications` addition to contract revision 15. The shared fixture and the Android copy are byte-identical at 15. The TS, iOS and Android fixture assertions expect 15, and the iOS fixture CodingKeys keep both main's streamInput/question/chatSkills and this PR's scheduleRunNotification.
+
+- 2026-09-27 review: plan index now describes Phase 6 as implemented; physical iPhone acceptance remains pending.

.memory/chat-pull-requests.md

@@ -4,7 +4,7 @@ Work continues on existing PR #184 (`devin/1789864248-chat-pull-requests`), isol
 
 Never normalize a supplied malformed expectedHeadSha to omission. Unknown creates remain pending even after an empty lookup; absence from a list does not establish that a mutation failed. Save link + settle intent atomically. Recovery uses the durable host/repository, independent of chat workspace changes. Same-target pending intents block repeat creates. Post-push identity comes from the frozen endpoint used by Git, never gh's inferred default repository. Unlink dismissal is durable and suppresses current-PR rediscovery until explicit relink.
 
-Two requested Sol medium reviews identified empty-list retries, split link/intent publication, selected-remote routing, and source IPC mismatch; all remediated with focused coverage. Final verification tracked in docs/plans/chat-pull-requests-plan.md and PR #184.
+Two requested Sol medium reviews identified empty-list retries, split link/intent publication, selected-remote routing, and source IPC mismatch; all remediated with focused coverage. Final verification tracked in docs/plans/completed/chat-pull-requests-plan.md and PR #184.
 
 Follow-up review fixes: store settlement checks that the operation is still pending inside the serialized write, preventing stale completions from undoing unlink. Unknown/retargeted/advanced-head attempts can be explicitly cleared only after a confirmation to check GitHub first. Within-repository create destination is named in the dialog; cross-fork upstream creation remains manual.
 
@@ -19,3 +19,5 @@ Failed-recovery deletion follow-up: after fencing admission and draining loads/w
 Dotted-ID recovery isolation follow-up: DataStore recovery and PR deletion share a complete-basename matcher with the fixed hash/operation/suffix fields, preventing chat-1 from consuming chat-1.json recovery files. Cached and cold deletion regressions cover both directions and both artifact suffixes; startup recovery also preserves sibling bytes. Validation: 98 focused PR tests, 255 portable-config/storage tests, 103 service-boundary tests, TypeScript and Electron build pass. Both independent Sol reviews are clean; current-head hosted checks remain pending.
 
 Recovery-token follow-up: artifact ownership also requires the lowercase UUIDv4 operation token emitted by randomUUID. Non-UUID, wrong-version and wrong-variant lookalikes remain untouched during recovery and deletion. Existing recovery fixtures now use generated-format UUIDs. Validation: 99 focused PR tests, 255 portable-config/storage tests, TypeScript and Electron build pass.
+
+Status (2026-09-27 plan refresh): PR #184 merged 2026-09-23 and shipped in 0.43.0; plan moved to docs/plans/completed/.

.memory/dictation-parakeet-modes.md

@@ -1,6 +1,6 @@
 # Dictation Parakeet lifecycle, modes, and dictionary — 2026-09-27
 
-- Branch `feature/dictation-parakeet-modes`. Plan: `docs/plans/dictation-parakeet-modes-plan.md`.
+- Branch `feature/dictation-parakeet-modes`. Plan: `docs/plans/completed/dictation-parakeet-modes-plan.md`.
 - **Shared, pure modules:**
   - `renderer/shared/dictation-preferences.ts` holds the mode resolution, idle-minute validation, and `parseDictationPreferencePatch`, which `settings:set` uses.
   - `renderer/shared/dictation-dictionary.ts` holds parse, apply, and add-entry.
@@ -34,3 +34,5 @@
 - Added shortcut/dictionary regressions; 44 focused tests, CI policy suite, desktop typecheck, and CLI build/typecheck pass. New suites are assigned to CI lanes.
 
 Independent review corrected the dictionary settings row keys to use the same locale-independent lowercase identity as parser deduplication; a Turkish-casing regression verifies distinct I/dotless-ı entries keep distinct React keys.
+
+Status (2026-10-01): merged in PR #279 (on main after 0.51.0); plan moved to `docs/plans/completed/`.

.memory/dictation-secure-input.md

@@ -1,6 +1,6 @@
 # Dictation Secure Input warning — 2026-09-27
 
-Branch `feature/dictation-secure-input`; plan `docs/plans/dictation-secure-input-plan.md`.
+Branch `feature/dictation-secure-input`; plan `docs/plans/completed/dictation-secure-input-plan.md`.
 
 - `pasteTranscript` (main/services/dictation-paste.ts) takes an injectable
   `isSecureInputActive`. Order: Accessibility check → Secure Input probe → atomic
@@ -23,3 +23,5 @@ Review validation: 15 focused paste/pill tests pass, including a process-owned e
 The JXA probe explicitly binds `IsSecureEventInputEnabled` as a no-argument boolean function, avoiding reliance on OS BridgeSupport metadata. The plan index and PR description now match the Carbon detector and conservative copy fallback.
 
 Independent review: reading the original AXValue is optional, so text controls without an accessible value still receive a guarded paste attempt. An unconfirmed result or transport error says “Check the field — transcript copied.” It never instructs a second paste after a possibly successful attempt; the prior clipboard is restored only after confirmed delivery.
+
+Status (2026-10-01): merged in PR #267 (on main after 0.51.0); plan moved to `docs/plans/completed/`.

.memory/durable-tool-outputs.md

@@ -1,7 +1,7 @@
 # Durable tool output / file provenance — September 22, 2026
 
 Active implementation: `feature/durable-produced-tool-output`, baseline `c8c09e0d`.
-Plan: `docs/plans/durable-tool-output-plan.md`. MCP was already bounded on current
+Plan: `docs/plans/completed/durable-tool-output-plan.md`. MCP was already bounded on current
 main, despite the September 15 Notion audit. Adapt the successful-mutation principle
 from verified `deepseek-ai/deepseek-harness` deliverables documentation; no code copied.
 
@@ -20,3 +20,5 @@ Validation: 125 focused tests and 10 inventory-fence tests pass; TypeScript and
 PR #219 at 58abfb02 passed hosted CI/Android/Electron gates. Pullfrog follow-up fixes include all three encrypted credential stores even in no-MCP workspaces, POSIX colon path parity, 240 Unicode code-point limits across clients, ASCII-only drive prefixes, and AJV-tested normative path/tool/status constraints. Follow-up validation: 128/128 focused tests, TypeScript, lint, Android chat/contract tests, and unsigned generic iOS test build pass; both independent reviewers clear. Follow-up hosted CI remains pending; physical XCTest remains blocked by device lock.
 
 Pullfrog incremental follow-up: ProducedFile schema lookaheads now scan all characters, including U+2028/U+2029; parity vectors cover valid paths, traversal, dot/empty segments and trailing separators for both. Protocol suite passes via node --import tsx (CLI IPC blocked by sandbox). Runtime/native validators unchanged. Hosted validation pending; physical device still locked.
+
+Status (2026-09-27 plan refresh): PR #219 merged 2026-09-23 and shipped in 0.43.0; plan moved to docs/plans/completed/. Physical iOS execution is still unverified.

.memory/live-activity-freshness-bot-deeplinks.md

@@ -1,6 +1,6 @@
 # Live Activity freshness chips and Bot deep links (2026-09-27)
 
-Branch: `feature/live-activity-freshness`. Plan: `docs/plans/live-activity-freshness-bot-deeplinks-plan.md`.
+Branch: `feature/live-activity-freshness`. Plan: `docs/plans/completed/live-activity-freshness-bot-deeplinks-plan.md`.
 
 ## Live Activity state and chips
 
@@ -24,3 +24,5 @@ Branch: `feature/live-activity-freshness`. Plan: `docs/plans/live-activity-fresh
 ## Coordination
 
 PR #119 adds `AidenBotLiveActivityStateTests.swift` and edits the pbxproj. This branch adds no new iOS files, to avoid conflicting with it.
+
+Status (2026-10-01): merged in PR #276 (on main after 0.51.0); plan moved to `docs/plans/completed/`.

.memory/mcp-advertised-status.md

@@ -8,4 +8,6 @@ Shared TypeScript status definition is desktop-only. Confirmed no McpStatus/mcp:
 
 Validation: registered `test:mcp` includes six new status tests; 88 total pass. Real in-memory SDK servers cover success, discovery failure, opaque extension metadata, empty/resource-only capability sets and independent clients. Controlled barriers cover document/connection revocation and generation-bound replacement. Full type-check, ESLint and whitespace checks pass. Independent GPT-5.6 Sol medium blast-radius/adversarial reviews requested.
 
-Original assignment completion remains partial: docs/plans/mcp-session-context-audit.md classifies every item. Resources, scoped server instructions and per-model-request AGENTS refresh are concrete remaining deliverables, not external blockers. Pi compaction owner confirms no live AGENTS/MCP hook overlap. No native test blocker is claimed for this desktop-only diagnostic slice; the prior #214 physical iOS lock remains that slice's limitation.
+Original assignment completion remains partial: docs/plans/completed/mcp-session-context-audit.md classifies every item. Resources, scoped server instructions and per-model-request AGENTS refresh are concrete remaining deliverables, not external blockers. Pi compaction owner confirms no live AGENTS/MCP hook overlap. No native test blocker is claimed for this desktop-only diagnostic slice; the prior #214 physical iOS lock remains that slice's limitation.
+
+Status (2026-09-27 plan refresh): PR #226 merged 2026-09-23 (0.43.0). Resources (#230), server instructions (#229) and trusted AGENTS refresh (#232) merged the same day, so the audit is complete.

.memory/mcp-numeric-schema-formats.md

@@ -17,3 +17,5 @@ Tests: `main/services/mcp-tool-schema.test.ts` (in `test:mcp`) uses realistic sc
 Draft-07 schema-valued dependencies now normalize too; property-name dependency lists remain unchanged. Added strict-Ajv regression and registered the suite in the core-git CI lane. Focused schema tests (7) and CI policy tests (46) pass.
 
 Independent review identified a silent 64-level cutoff that could leave numeric formats unnormalized. Traversal now uses an iterative worklist and weak object-copy map, preserving raw identity and schema-keyword boundaries while normalizing deep schemas fully; a 2,000-level regression covers this case.
+
+Status (2026-10-01): merged in PR #264 (on main after 0.51.0); plan moved to `docs/plans/completed/`.

.memory/mcp-scoped-resources.md

@@ -9,3 +9,5 @@ Pullfrog identified that pinned SDK1.30.0 concatenates multi-variable expression
 
 ## Constrained template syntax correction
 A follow-up review found SDK1.30 also misinterprets single-variable operators/modifiers. Discovery now rejects every form except plain {name} expressions with ASCII identifier names, before any inventory handles are published. Operators, prefix/explode modifiers, comma-separated names, dotted/percent-encoded names and malformed braces fail closed. Accepted plain scalar values use strict RFC6570 percent encoding (including !'()*), without SDK expansion. Repeated separate expressions still work. An unsupported template makes that inventory fail closed; no partial inventory is presented as complete. MCP97 covers the requested single-variable forms, unpublished-handle denial and Unicode/reserved encoding.
+
+Status (2026-09-27 plan refresh): PR #230 merged 2026-09-23 and shipped in 0.43.0; plan moved to docs/plans/completed/.

.memory/mcp-scoped-server-instructions.md

@@ -11,3 +11,5 @@ Onboarding's existing MCP tile now discloses service-provided tool guidance with
 Validation: MCP90, onboarding56, Bot447, scheduled151 passed. Full type-check, ESLint and whitespace checks passed after test-fixture fixes. New mcp-server-instructions.test.ts is registered in test:mcp. Both requested independent GPT-5.6 Sol medium reviews (blast radius; adversarial/edge cases) clear. Reviewers confirmed exact tool/server identity, final filtering, immutable lifetime, context budget and fail-closed bounds. Hosted checks and review remain PR follow-through gates.
 
 Original scope remains partial: scoped MCP resource operations and trusted AGENTS loading/request-boundary prompt refresh remain concrete deliverables. Capability status is separate #226, skill policy #214, result spills another owner's work. This change claims no resource access, provider upgrade or native acceptance.
+
+Status (2026-09-27 plan refresh): PR #229 merged 2026-09-23 (0.43.0). The resources and trusted AGENTS deliverables listed above as remaining also merged (#230, #232), so the original assignment is complete; plan and audit are in docs/plans/completed/.

.memory/mobile-transcript-polish.md

@@ -1,6 +1,6 @@
 # Mobile transcript polish (Hermex Tier 1)
 
-Branch `feature/mobile-transcript-polish`. Plan: `docs/plans/mobile-transcript-polish-plan.md`.
+Branch `feature/mobile-transcript-polish`. Plan: `docs/plans/completed/mobile-transcript-polish-plan.md`.
 
 - There was no protocol change. "Worked for" reads `Message.timeline`, but only when the status is completed and `finishedAt >= startedAt`. The live timer uses the running timeline's `startedAt` and otherwise falls back to the newest user message's `createdAt`. Timestamps use `createdAt`.
 - iOS code lives in `Features/Chat/AidenTranscriptPolish.swift`. It is a new file because `AidenChatFeature.swift` is far over the 500-line Swift warning, and it is registered manually in the pbxproj with IDs `A714739233F64F4FA4C2B700` and `...710`. The footer replaced the old safeAreaInset copy/read-aloud row. `AidenSettledMessageRows` computes `showsFooter`, so Bot clusters show it only on the last joined bubble.
@@ -9,3 +9,5 @@ Branch `feature/mobile-transcript-polish`. Plan: `docs/plans/mobile-transcript-p
 - SwiftUI `.textSelection` and Compose `SelectionContainer` cannot add custom menu items, so selection uses a native select-text sheet/dialog. On iOS this is a `UITextView` with `editMenuForTextIn`; on Android it is a `TextView` with `customSelectionActionModeCallback`.
 - User messages are now copyable on iOS. `copyText` returns nil only for empty text.
 - Open items: flatten Markdown in the Android select-text dialog, and focus the Android composer after Ask.
+
+Status (2026-10-01): merged in PR #277 and shipped in 0.51.0; plan moved to `docs/plans/completed/`. Physical-device visual acceptance remains.

.memory/pr187-context-meter.md

@@ -18,4 +18,6 @@
 - Pi 0.87.1 merge (#246): the prompt now lives in transcript system messages and the AGENTS refresher appends a section patch (`apply(AgentContext)`). `withAgentsInstructionsEstimate` builds a one-message transcript (`createInitialSystemMessage`), applies the refresher and reads `getCurrentSystemPrompt`. `withoutAgentsInstructions` also strips a block rendered with no base prompt. llm-client registers the generation profile after the initial AGENTS apply, so the captured prompt and the tracker baseline agree. The harness emits projections from `installCompactedMessages`' result. Journal system messages count zero in `messageTokens`, so static context is counted once. Pullfrog on 5186fe54 pointed out that Pi sends later system messages in place for `supportsMidConvoSystemMessages` models, including superseded AGENTS revisions. `GenerationContextOptions.retainsSystemUpdates` (from `modelRetainsSystemUpdates(model)`; set in llm-client, the harness projection options, and `nextRequestContextOptions`, which defaults it to false) makes whole-transcript estimates (`projectNextContextUsage` and compaction) add the rendered updates beyond the replayed prompt. Compaction's replay into one head removes that extra.
 - Retained updates and usage anchors (Pullfrog on 8f06c37e): provider usage covers everything up to its anchor, but Pi's `estimateContextTokens` and `messageTokens` price system messages at 0, so for a retaining model each system message after the latest valid anchor is added in full (rendered with `renderSystemMessageUpdate`) through `retainedTailSystemTokens`. The same helper serves the projection's anchored tail and provider term, compaction's provider-aware total and candidate tails, and `PiCompactionCoordinator` (the direct previous-assistant usage in `checkContextPressure`, the stale-anchor estimate in both post-turn checks, and the whole-transcript fallback after a model switch, which adds later updates but not the leading head). A zero-usage response does not move the anchor, so revisions before and after it both count.
 - Static-inclusive totals (Pullfrog on de658be1): content and still-active sections set after the anchor are already in `options.systemPrompt`. Totals that add `staticTokens` next to the anchored tail (the projection's usage+static+trailing term and the compaction candidates) use `retainedTailSystemTokensBeyondPrompt` (rendered updates minus `postAnchorPromptChars`). Compaction's prefix ratio subtracts the anchor-time static (static minus the post-anchor share), so `static + prefix*ratio + tail` equals usage plus the full tail, and a replayed head prices the active revision. Totals without static (Pi's `estimate.tokens`, direct usage, every pi-compaction-core path) keep the full render.
-- Plan: `docs/plans/composer-context-meter-plan.md`.
+- Plan: `docs/plans/completed/composer-context-meter-plan.md`.
+
+Status (2026-09-27 plan refresh): PR #187 merged 2026-09-26 and shipped in 0.50.0; plan moved to docs/plans/completed/composer-context-meter-plan.md.

.memory/queue-while-compacting.md

@@ -1,6 +1,6 @@
 # Queue messages while compaction runs — 2026-09-27
 
-Branch: `feature/queue-while-compacting`. Plan: `docs/plans/queue-while-compacting-plan.md`.
+Branch: `feature/queue-while-compacting`. Plan: `docs/plans/completed/queue-while-compacting-plan.md`.
 
 Manual compaction keeps the composer editable and holds queued messages until compaction succeeds. `ChatMessageQueue.holdReason` is the durable per-chat state; the composer must derive its active compaction affordance, status text and Cancel action from the queue snapshot as well as its local start state, because the chat-keyed composer remounts when navigating away and back.
 
@@ -9,3 +9,5 @@ The `/compact` slash token is consumed immediately after the asynchronous comman
 Relevant validation: `npm run test:slash-commands`; `npm run type-check:e2e`; focused `chat-message-queue.spec.ts` compaction scenario.
 
 Follow-up review: attachment and skill removal now use `composerInputLocked`, preserving draft editing while manual compaction holds queued sends. The compaction E2E removes a pasted image before navigation while the original command is still pending.
+
+Status (2026-10-01): merged in PR #272 (on main after 0.51.0); plan moved to `docs/plans/completed/`.

.memory/rich-link-previews.md

@@ -9,3 +9,5 @@
 - Escape fences both the open card and Radix's delayed focus-open callback until the next pointer-enter or focus interaction, without moving focus.
 - No network, IPC, connector, persistence, schema, transcript, Bot runtime, main-process, iOS, or Android contract changed.
 - Validation: focused rich-link/transcript suite 42/42; Bot suite 451/451; Chat/slash-command suite passed; TypeScript, ESLint, CI policy, production build, and `git diff --check` passed. Packaged-app, pointer/keyboard visual, assistive-technology, and physical-device acceptance remain separate.
+
+Status (2026-09-27 plan refresh): PR #253 merged 2026-09-26 and shipped in 0.50.0; plan moved to docs/plans/completed/rich-link-previews-plan.md. The optional authenticated metadata phase is unstarted.