fix(runtime): land conversation affinity and effective capabilities

[lidge-jun]

Summary

• Give Command Code requests a stable conversation cache affinity and evaluate effective provider capabilities against the resolved, available route. Carries fix(command-code): preserve conversation cache affinity #3581 through fix(command-code): preserve conversation affinity across recovery #3692 and fix(routing): evaluate effective provider capabilities #3671 through fix(routing): evaluate effective provider capabilities and availability #3694 with all original contributor commits preserved.
• Include fix(codex): bound probe cleanup and stabilize verification fixtures #3708's bounded Unix probe cleanup and deterministic replay/verification fixtures. The final transition fixture covers permitted Windows identity startup latency and reports early child exits. Quota fixtures join observation and forget work before asserting or resetting the sink. Neither repair changes production deadlines or event/transition assertions.
• Integrate the independently reviewed stack through this final descendant. Lower review PRs receive the actual dev landing evidence after ancestry verification. Related unreadable_encrypted_agent_task intermittently fails routed V2 subagent dispatch (strict recovery envelope; no retry during chatgpt.com 5xx windows) #3661 remains open for its residual scope.

Verification

• Exact head 5097e66faa031d4e11b09c8ad79ec94ecb3f713d: all 24 actual Linux, macOS and Windows CI producers passed in full cross-platform CI. Historical failed predecessor runs remain recorded; this is the tested integration candidate.
• Remote pinned Bun1.4.0: focused transition tests4pass/0fail and typecheck passed. Controlled10.5s startup passes the new watchdog and fails the old10s guard; an injected startup exception is reported before the barrier. Quota-focused tests16pass/0fail and typecheck passed; delayed-queue and suppressed-delivery controls preserve the exact event assertions. Independent code, security and topology reviews passed.
• No local tests, typechecks or builds were run, as requested by the owner. Final dev integration CI and docs-only closeout follow the verified merge.

Checklist

• Scope stays focused and avoids unrelated cleanup.
• Docs or release notes were updated when needed.
• Security-sensitive changes were reviewed for secrets, auth, and unsafe defaults.

Co-authored-by: SB Yoon 44089734+yansigit@users.noreply.github.com
Co-authored-by: Hako 25837994+devswha@users.noreply.github.com

[coderabbitai]

Important

Draft PR not reviewed

Draft PRs are not automatically reviewed by default.

• Trigger a manual review

To automatically review draft PRs, update your CodeRabbit configuration:

reviews:
  auto_review:
    drafts: true

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[github-actions]

✅ Deterministic PR hygiene checks passed.

[lidge-jun]

리뷰 · 우선순위 46 / 80

이 PR은 제품 타이밍이 아니라 Windows transition probe 테스트 하네스의 예산 결함만 고칩니다. 손대는 파일은 tests/codex-integration/codex-transition-state-race.test.ts 와 plan 노트뿐입니다. 지금 실패 로그(foundation run 33988432596)는 두 자식이 ready barrier를 올리기 전에 바깥 10초 데드라인에 걸렸고, 같은 픽스처는 스택 tip(33988434944)에선 통과했습니다. Windows 자식은 ready 전에 SID·known folder를 각각 최대 WINDOWS_PRINCIPAL_LOOKUP_TIMEOUT_MS (30초) PowerShell로 풀 수 있어, 10초 바깥 예산은 정상 동작도 거절할 수 있습니다.

고침은 세 가지입니다. (1) CHILD_TIMEOUT_MS 를 win32에서 2 * WINDOWS_PRINCIPAL_LOOKUP_TIMEOUT_MS + 5_000 으로 잡고 watchdogMs 로 감쌉니다. (2) waitForFiles 가 barrier 기다리다 자식이 먼저 죽으면 stderr와 함께 바로 실패합니다. (3) 테스트 finally에서 child.exited 를 await한 뒤 sandbox를 지워, 삭제 레이스를 줄입니다. 바깥 test timeout도 N * CHILD_TIMEOUT_MS 로 키웠습니다. winner/generation/shared-db/busy/unsafe-path 단언은 그대로입니다.

베이스는 dev가 아니라 codex/a-05-capabilities (#3694 capabilities 스택)입니다. A-lane은 이미 ancestry에 SSE, WS, recovery가 있고, 다음이 affinity/capabilities입니다. 이 테스트 후속은 그 tip 검증용이라, 지금 HEAD 6dd23d631 대비 단독 머지 가치는 낮고 스택 순서 의존이 큽니다. 우선순위 46은 실프로세스 레이스 픽스처를 살리는 유용한 chore지만 제품 경로가 아니고 스택이 깊기 때문입니다.

경로/심볼 - CHILD_TIMEOUT_MS - win32 identity 두 번 + 5초 headroom, 타 플랫폼 10초 + watchdogMs.
경로/심볼 - waitForFiles(paths, children) - 조기 exit 진단.
라인 - finally await Promise.all(children.map(c => c.exited)) - sandbox 삭제 전 회수.
경로 - 각 test timeout을 N * CHILD_TIMEOUT_MS 로 조정 - 단계 수에 비례.
경로 - src/codex/transition-state.ts 제품 코드 - 변경 없음.
경로 - 아래층 #3694 / affinity 스택 - 이 PR 단독 dev 랜딩 금지.

메인테이너의 판단이 필요한 지점

• Windows CI에서 새 예산으로 delayed-ready 컨트롤이 구 10초에선 실패·신 예산에선 통과하는지 원격으로 확인했는지
• A-lane 전체를 한 트레인으로 올릴 때 이 chore를 tip에 붙일지, capabilities 직후에 짧게 넣을지
• 30초×2 예산이 CI wall-time에 과도한지, watchdog 배율만으로 충분한지

너의 추천
스택 대기. #3694(및 affinity 아래층)가 dev에 들어간 뒤 이 PR을 retarget하고 Windows exact-head 포커스 테스트로 검증한 다음 머지하세요. 제품 타임아웃을 건드리지 않은 점은 유지. types/config 분할 무관. 아래층이 방향을 바꾸면 close-don't-rebase.

이 댓글은 grok-bot이 작성했습니다

[lidge-jun]

Owner-authorized admin integration of the independently reviewed A stack through #3716.

Exact head: 5097e66. All 24 actual full cross-platform producers succeeded in https://github.com/lidge-jun/opencodex/actions/runs/33991642514. Any queued aggregation-only job is accepted only after directly checking its exact allowlist. No pending or failed functional producer of this integration candidate is waived.

Carries #3708 bounded Unix probe cleanup and deterministic verification, #3692 Command Code conversation affinity (source #3581), and #3694 effective provider capability evaluation (source #3671). Earlier failing foundation jobs remain recorded and are superseded by this fully tested descendant. All lower-layer and contributor commits remain ancestors; original author identities are preserved. After dev ancestry verification, originals and superseded review PRs are closed with the actual landing evidence. Related #3661 remains open.

Co-authored-by: SB Yoon 44089734+yansigit@users.noreply.github.com
Co-authored-by: Hako 25837994+devswha@users.noreply.github.com

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-09-05T21:46:52.996684Z	5097e66	Draft marked ready

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.