Files
claude-code/docs/agent/sur-skill-overflow-bugs.md
unraid f2e9af4927 feat: harden autonomy lifecycle, OOM bounds, and provider-boundary finalization
This PR consolidates a coordinated batch of fixes around autonomy run/flow lifecycle, scheduled task deduplication, provider-boundary state finalization, and matching memory-bound treatments for adjacent long-running subsystems (REPL fullscreen scrollback, skill-search/skill-learning runtime activation). All changes were developed and reviewed together because they touched the same lifecycle invariants and were uncovered by the same long-running session reproductions.

## Lifecycle correctness

- Queued autonomy prompts are not injected unless the persisted run was successfully claimed; queued run claiming is now terminal-safe so a once-consumed/cancelled/failed run can not slip back into `queued`.
- Autonomy run/flow finalization happens on completion, provider error, generator close, and cancellation — not just the happy path. New `src/__tests__/queryAutonomyProviderBoundary.test.ts` covers these provider-boundary transitions.
- `requestManagedAutonomyFlowCancel` and `resumeManagedAutonomyFlowPrompt` carry `rootDir` and `currentDir` explicitly across detached async boundaries (proactive-tick, cron, daemon restart) instead of inferring from process state.
- Active runs/flows are protected from janitor pruning so a running step can not be garbage-collected mid-flight (`src/utils/autonomyAuthority.ts`).
- Heartbeat parser now ignores fenced code blocks; the two-phase commit window for autonomy state transitions is documented in `docs/internals/autonomy-jira.md`.

## Ownership and dedup

- `src/utils/autonomyRuns.ts`: ownership stamping (run id + rootDir carried end-to-end), source-based dedup against active runs.
- `src/hooks/useScheduledTasks.ts`: scheduled ticks deduplicate against runs already active on the same source label.
- `src/utils/processUserInput/processSlashCommand.tsx`: forked slash commands now thread the autonomy `runId` so completion finalizers can find the originating run for deferred completion.
- New `src/utils/autonomyQueueLifecycle.ts` and tests collect the queue-side lifecycle invariants in one place.

## Memory bounds (related, same review pass)

- `src/screens/REPL.tsx`: caps fullscreen scrollback after the compact boundary and updates trailing progress rows in place. Long-running fullscreen sessions could otherwise retain thousands of post-compaction messages and duplicate progress rows, keeping Ink trees alive long after their useful context had moved on.
- `src/services/skillSearch/*` and `src/services/skillLearning/*`: runtime activation is strictly opt-in via existing env toggles; session caches are capped so long-running processes can not grow them forever. Build presence is preserved so operators can still discover and opt into the slash commands.

## CI / test contract

- `tests/integration/dependency-overrides.test.ts`: smoke test no longer drives Mermaid's browser renderer; it validates the package-resolution contract directly so CI does not regress on unrelated browser timing.
- New `tests/integration/autonomy-lifecycle-user-flow.test.ts`: end-to-end CLI subprocess flow exercising `status --deep`, `flows`, `flow <id>`, `flow resume`, `flow cancel` against persisted state.
- `src/entrypoints/cli.tsx`: `claude autonomy …` routes through an entrypoint fast path that reuses the slash-command formatter without booting the full interactive CLI. Stdout is flushed before forced exit so coverage subprocesses do not terminate with empty stdout.
- `packages/builtin-tools/src/tools/RemoteTriggerTool/__tests__/RemoteTriggerTool.test.ts`: stabilized to prevent audit flake under coverage.

## Tests added

- `src/__tests__/queryAutonomyProviderBoundary.test.ts`
- `src/hooks/__tests__/useScheduledTasks.test.ts`
- `src/utils/__tests__/autonomyAuthority.test.ts`
- `src/utils/__tests__/autonomyFlows.test.ts` (extended)
- `src/utils/__tests__/autonomyPersistence.test.ts` (extended)
- `src/utils/__tests__/autonomyQueueLifecycle.test.ts`
- `src/utils/__tests__/autonomyRuns.test.ts` (extended)
- `src/utils/processUserInput/__tests__/processSlashCommand.test.ts`
- `tests/integration/autonomy-lifecycle-user-flow.test.ts`

## Docs

- `docs/agent/sur-loop-scheduled-oom.md`: System Understanding Report covering the scheduled/loop OOM problem, the call graphs investigated, and the lifecycle invariants this PR establishes.
- `docs/agent/sur-skill-overflow-bugs.md`: SUR for the related skill-overflow context.
- `docs/internals/autonomy-jira.md`: documents the two-phase commit window and ownership stamping invariants.
- `docs/memory-leak-audit.md`: audit notes covering the REPL/scrollback and skill-search bounds.

## Invariants this PR establishes

1. Queued autonomy prompts are not injected unless the persisted run was successfully claimed.
2. Terminal run/flow states are terminal — completion, failure, and cancellation all finalize state regardless of which provider/error path triggered them.
3. Autonomy run/flow `rootDir` is carried explicitly across detached async boundaries instead of inferred from a shared singleton.
4. State-only CLI subcommands (`autonomy status|runs|flows|flow …`) bypass full interactive bootstrap so they do not hold unrelated handles open.
5. REPL fullscreen scrollback and skill-search/skill-learning session caches are explicitly bounded.

## Validation

```bash
bun run typecheck
CI=true GITHUB_ACTIONS=true bun test            # 3996 pass / 0 fail across 305 files
bun test src/__tests__/queryAutonomyProviderBoundary.test.ts \
         src/hooks/__tests__/useScheduledTasks.test.ts \
         src/utils/__tests__/autonomy{Runs,Flows,Authority,QueueLifecycle,Persistence}.test.ts \
         src/utils/processUserInput/__tests__/processSlashCommand.test.ts \
         tests/integration/autonomy-lifecycle-user-flow.test.ts
```

## Origin

This PR is the consolidated, upstream-targeted version of two fork-side review PRs (fix/loop-scheduled-autonomy-oom and fix/autonomy-lifecycle). The fork-side review history is preserved at https://github.com/amDosion/claude-code-bast/pull/7 . The fork's own internal `chore: keep fork current with upstream` sync commits and the `docs: update contributors` automation are intentionally not included in this PR.

The autonomy CLI handler `rootDir` threading that the fork added (78f64d8a, 98d04ddb) is intentionally omitted here because upstream `a2cfaf91` (fix: 修复 RemoteTriggerTool 和 autonomy 测试的全量运行失败) already performed the equivalent change with an additional `currentDir` option. Keeping the upstream version avoids regressing that improvement.
2026-04-29 14:04:27 +08:00

8.0 KiB
Raw Blame History

System Understanding Report — Skill Search / Skill Learning Overflow Bugs

  • Flow id: recurring-bug-skill-overflow (sibling pilot to recurring-bug-loop-oom)
  • Branch: fix/loop-scheduled-autonomy-oom (folded into the OOM PR — same audit-and-cap pattern)
  • Trigger: post-merge review of the autonomy OOM fix surfaced unbounded module-level state in adjacent EXPERIMENTAL_SKILL_SEARCH and SKILL_LEARNING subsystems. The user explicitly asked for a 肯定也有同类溢出 audit.

1. Problem

The autonomy OOM bug came from unbounded module-level state (run records, scheduler queues, heartbeat timestamps) growing for the lifetime of the process. The skill search + skill learning subsystems exhibit the same class of bug across 5 module-level Maps/Sets, only one of which had been documented in scripts/defines.ts ("projectContext cache 无淘汰机制(非 GB 级主因)").

These bugs were latent because:

  • EXPERIMENTAL_SKILL_SEARCH / SKILL_LEARNING were enabled-by-default in DEFAULT_BUILD_FEATURES, but tests pass because they exercise short paths.
  • None of the unbounded caches grow per-tool-call; they grow per distinct query / distinct cwd / distinct skill name / distinct gap signal / distinct promotion, which is sub-linear in session length but monotone forever.
  • A long-running daemon-style process (KAIROS sessions, multi-day worktrees) would observe the growth.

2. Module-level state audit

File:Line Symbol Pre-fix bound Pre-fix evict
intentNormalize.ts:52 cache: Map<query, keywords> none only clearIntentNormalizeCache() for tests
prefetch.ts:17 discoveredThisSession: Set<skillName> none none
prefetch.ts:18 recordedGapSignals: Set<gapKey> none none
projectContext.ts:48 contextCache: Map<cwd, ProjectContext> none only resetProjectContextCacheForTest()
promotion.ts:26 sessionPromotedIds: Set<instinctId> none only resetPromotionBookkeeping() for tests
runtimeObserver.ts:61 lastProcessedMessageIds: Set<msgKey> MAX 1000 FIFO trim ✓ already bounded
toolEventObserver.ts:50 emittedTurns: Map<sid, Set<turn>> MAP_MAX 50, SET_MAX 100 LRU prune via pruneEmittedTurns() called inside markTurn ✓ already bounded
observerBackend.ts:21 registry: Map<name, Backend> fixed N n/a — registry pattern, finite ✓

5 unbounded out of 8 module-level mutables. All 5 are addressed in this PR.

3. Severity rationale

Per-entry cost is small (key strings + small objects), so OOM in days is unlikely on a normal workstation. But the canary scenarios:

  • intentNormalize.cache: every distinct Chinese query → Haiku call → cached. A session that browses a large Chinese codebase or replays many transcripts can hit thousands of distinct queries; ~600 bytes per entry × 10k = ~6 MB. Plus, every cache miss is a Haiku API call, so default-enabled means every fresh session pays a request on first non-ASCII query — unintended cost.
  • projectContext.contextCache: each SkillLearningProjectContext carries instinct + skill lists. Multi-worktree orchestrators (this very repo!) blow past the typical "1 cwd per session" assumption.
  • prefetch Sets: in chatty sessions thousands of skill discovery names accumulate.
  • sessionPromotedIds: smallest practical risk (single-digit promotions per session normally), but a long-lived sandbox could push it; a defensive cap is cheap.

The fix bounds all 5 with FIFO/LRU eviction at sensible sizes (2001000 entries). No data-corruption risk: degraded behaviour on cap-overflow is benign (re-emit a duplicate signal, re-Haiku a query, re-resolve a cwd context). Same risk profile as the autonomy stale-recovery design.

4. Fix surface

File Change
src/services/skillSearch/intentNormalize.ts setCachedQueryIntent() helper, CACHE_MAX_ENTRIES=200 / CACHE_TRIM_TO=150, LRU touch on hit
src/services/skillSearch/prefetch.ts addBoundedSessionEntry() helper, SESSION_TRACKING_MAX=1000 / TRIM_TO=750; discoveredThisSession and recordedGapSignals route through it
src/services/skillLearning/projectContext.ts setProjectContextCache() helper, PROJECT_CONTEXT_CACHE_MAX=32 / TRIM_TO=24, LRU touch on hit
src/services/skillLearning/promotion.ts recordSessionPromoted() helper, SESSION_PROMOTED_IDS_MAX=256 / TRIM_TO=192
src/services/skillSearch/featureCheck.ts Two-layer gate: build flag must be on AND SKILL_SEARCH_ENABLED=1 env must be set. Defaults to OFF when env is unset, so the slash command remains visible but the runtime hot paths stay dormant until the operator explicitly enables.
src/services/skillLearning/featureCheck.ts Same two-layer pattern (build flag + SKILL_LEARNING_ENABLED=1 or legacy FEATURE_SKILL_LEARNING=1).
scripts/defines.ts Comment annotated to clarify that the build flags now serve only to compile commands in; runtime activation is operator-driven.

5. Why default-off (without removing from build)?

Three reasons aside from the unbounded-cache concern:

  1. Implicit cost: intentNormalize calls Haiku on cache miss. Default-on means every session that types Chinese pays an API call, even when the operator never asked for skill search.
  2. Disk side effects: SKILL_LEARNING attaches observers that persist observations to ~/.claude storage. Storage volume should be opt-in, not background.
  3. Experimental status: the flag is literally named EXPERIMENTAL_*. Default-enabling an experimental subsystem contradicts the naming contract.

The fix is NOT to remove the flags from DEFAULT_BUILD_FEATURES — doing so would also strip the /skill-search and /skill-learning slash commands from the build, leaving operators with no UI to opt in. Instead the activation logic in featureCheck.ts was changed to a two-layer gate:

  • Layer 1 (compile-time): feature('EXPERIMENTAL_SKILL_SEARCH') / feature('SKILL_LEARNING') must be on. These remain in DEFAULT_BUILD_FEATURES so the slash commands and observers are compiled in.
  • Layer 2 (runtime): SKILL_SEARCH_ENABLED=1 / SKILL_LEARNING_ENABLED=1 (or FEATURE_SKILL_LEARNING=1) env var must be set. Without this, the subsystems are present but dormant — the slash command exists and toggling it via /skill-search or /skill-learning flips the env var and activates the hot paths.

Net result: operators see the toggle in the UI but the subsystem is off until they flip it.

6. Out of scope (filed for follow-up)

  • Test failures on CI (prefetch.test.ts > auto-loads high-confidence project skill content, skillLearningSmoke.test.ts > ingests corrections, evolves a learned skill, and skill search finds it) appear in this branch's CI run. Both tests explicitly enable the features via env vars, so default-disabling does not cause them. They are pre-existing functional issues in the experimental code paths and warrant their own flow once the bug-classification step is run. Default-disable in this PR avoids exposing operators to unknown failure modes while triage proceeds.
  • Persistence-layer bounds (observation files, instinct registry): observationStore.ts already has 30-day purge and 1MB archive thresholds; skillGapStore.ts uses a finite-state lifecycle. Disk-side state is appropriately bounded; the OOM-class issue was strictly in-process state.

7. Verification

Local checks (full suite covers cap behaviour via existing tests; the caps degrade gracefully so no test should break):

bun run typecheck   # 0 errors
bun test src/services/skillSearch/__tests__/intentNormalize.test.ts
bun test src/services/skillSearch/__tests__/prefetch.extractQuery.test.ts
bun test src/services/skillLearning/__tests__/projectContext.test.ts
bun test src/services/skillLearning/__tests__/promotion.test.ts
bun run lint
bun run build

The new caps are observable behaviour: under sustained load the Map/Set sizes plateau at the configured maxima rather than monotone-growing.