test(service): block live scheduler mutation under the test guard

[luvs01]

Summary

The repository test runner isolates HOME, USERPROFILE, and OPENCODEX_HOME, but Windows Task Scheduler and UAC-launched processes are machine-global.

A partially-faked service repair test crossed that boundary and executed the default schtasks /create /tn opencodex-proxy ... /f path. It replaced the live task with a launcher inside the temporary test home, the test still passed, and cleanup deleted the launcher. The already-running proxy hid the damaged registration until the next reconnect or restart.

When the explicit repository test-home guard is armed, this change:

• permits only read-only /query operations through the direct Task Scheduler runner; and
• refuses the final live Windows elevation process boundary while retaining explicitly injected test launchers; and
• blocks the default native WinSW/SCM removal boundary while the test guard is armed, while retaining explicitly injected removal operations.

This covers direct and elevated registration, create/run/rollback, delete, and other machine-global mutation paths. Production behavior is unchanged because only the repository test preload sets OCX_TEST_HOME_GUARD=1.

Verification

• Regression green: a valid fresh-registration request is rejected before even the scheduler recorder is reached.
• Safe red proof: removing the direct guard reaches the fake create recorder and fails later with “registration is absent,” proving that path was reachable without touching the real scheduler.
• bun test tests/service.test.ts tests/windows-scheduler-install-verification.test.ts tests/test-home-guard.test.ts — 172 pass, 0 fail; 3 POSIX-only permission tests skipped on Windows.
• bun test tests/windows-elevation-spawn.test.ts — 53 pass, 0 fail; the live-boundary probe contains only a fixed non-RunAs exit 0 command if its guard regresses.
• bun run typecheck, bun run privacy:scan, and git diff --check — clean.
• The scheduled task still points to the real OpenCodex service launcher, and the current proxy health readback is HTTP 200.
• Exact head 56bd40d242 is based on current dev 870a2adb6e; three current-head scheduler, native-SCM, and explicit-injection guard regressions pass 3/3 on Bun 1.4.0. The previously completed full focused suites remain 197/0, with typecheck and privacy scan clean; the live opencodex-proxy registration was unchanged afterward and /healthz returned 200.

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.

Review readiness checklist

This PR stays in draft until every box below is ticked. Tick all four boxes once the requirements are met:

• All CI tests are green on my local testing.

• I pushed my PR to the latest dev commit.

• I resolved all correct Codex and CodeRabbit findings.

• My PR is ready for review.

Summary by CodeRabbit

• Bug Fixes

  • Added safeguards that prevent test-mode processes from modifying Windows Task Scheduler tasks or native services.
  • Prevented test-mode processes from launching live Windows elevation commands.
  • Preserved supported injected test launchers and simulated service workflows.

• Tests

  • Added coverage verifying blocked scheduler, service, and elevation operations occur before any system changes.

[github-actions]

✅ Deterministic PR hygiene checks passed.

[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 2026-08-30T18:09:17.135240Z	56bd40d	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.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Pro Plus

Run ID: b5965802-5ea8-4a08-9670-ca4d47955052

📥 Commits

Reviewing files that changed from the base of the PR and between 1101aba and 56bd40d.

📒 Files selected for processing (2)

• src/service.ts
• tests/service.test.ts

Included review availability: Your plan provides up to 10 included reviews per hour; 8 remain after this review.

---

📝 Walkthrough

Walkthrough

The change adds guards for armed test processes before live Task Scheduler mutations, native WinSW service removal, and Windows elevation launches. Tests verify rejection before external interaction and rollback before native-service removal.

Changes

Windows test-process guards

Layer / File(s)	Summary
Task Scheduler mutation guard src/service.ts, tests/service.test.ts	querySchtasks allows /query but rejects other operations when the guard is armed. The registration test confirms that the scheduler recorder receives no call.
Native service removal guard src/service.ts, tests/service.test.ts	removeNativeWindowsServiceForScheduler rejects the production uninstallWinswService path when the guard is armed. The lifecycle test confirms rollback completes before native-service removal.
Windows elevation launch guard src/lib/windows-elevation.ts, tests/windows-elevation-spawn.test.ts	startPowerShellCommand rejects live default-spawn launches when the guard is armed. Injected test launchers remain available, and the test verifies the rejected completion and null launcher PID.

Estimated code review effort: 2 (Simple) | ~10 minutes

Merge Risk: 🔵 Low · up to 56bd4

The test guard now prevents most live scheduler and elevation mutations, but some native Windows service lifecycle paths can still change machine-global service state when an armed test reaches them. The PR is mergeable with explicit owner awareness and follow-up to enforce the guard at the common service-management boundary.

Suggested reviewers: lidge-j, ingwannu

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 25.00% which is insufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 4 functions across 4 files.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly identifies the main change: blocking live Windows Task Scheduler mutations when the test guard is active. It is concise and specific, although it does not mention the related Windows…
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

Full details: Title check

Explanation

The title clearly identifies the main change: blocking live Windows Task Scheduler mutations when the test guard is active. It is concise and specific, although it does not mention the related Windows elevation safeguard.

• Fix all pre-merge checks with AI

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

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]

✅ READY

• all PR quality gates passed; the review readiness checklist is complete.

Review readiness checklist

• ✅ All CI tests are green on my local testing.
• ✅ I pushed my PR to the latest dev commit.
• ✅ I resolved all correct Codex and CodeRabbit findings.
• ✅ My PR is ready for review.

✅ 4/4 boxes ticked.

This pull request is already Ready for Review.
The review-ready label marks this PR as ready; review automation runs independently.
Maintainers: @lidge-jun @Ingwannu

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 97f499c1bc

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".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[lidge-jun]

리뷰 · 우선순위 69 / 80

설명

지금 dev HEAD는 f4d28626d 이다. 마지막 머지는 #2987 슬러그 셀프네임스페이스 네이티브 id 해석이고, 패키지 줄은 2.36.0이다. Windows 백그라운드 서비스는 src/service.ts 의 querySchtasks → schtasks 경로로 Task Scheduler 를 부른다. /query 뿐 아니라 /create, /delete, /run, /end 도 같은 함수를 탄다.

저장소 테스트 러너는 HOME / USERPROFILE / OPENCODEX_HOME 을 임시 집으로 바꾼다. 그런데 Windows Task Scheduler 는 기계 전역이다. 집이 바뀌어도 opencodex-proxy 태스크 이름은 같다. 그래서 테스트가 가짜 스케줄러를 일부만 끼우고, 나머지 호출이 기본 querySchtasks 로 떨어지면 살아 있는 사용자 태스크를 /create /f 로 덮어쓸 수 있다. 본문이 적은 사고도 그 모양이다. 부분 페이크 repair 테스트가 라이브 등록을 테스트 홈 런처로 바꾸고, 테스트는 초록, 정리 단계에서 그 런처를 지웠다. 이미 떠 있던 프록시가 망가진 등록을 가렸다가, 다음 재연결·재시작에서야 드러난다.

지금 HEAD에는 이미 src/lib/test-home-guard.ts 의 isTestHomeGuardArmed 이 있다. tests/preload.ts 만 OCX_TEST_HOME_GUARD=1 을 켠다. 프로덕션 ocx 에서는 가드가 꺼진 채다. 가드가 막던 것은 주로 진짜 ~/.opencodex / ~/.codex 쓰기였다. 스케줄러 변이까지는 막지 않았다. 이 PR은 무장된 테스트 프로세스에서 querySchtasks 가 첫 인자가 /query 가 아니면 바로 거절한다. 거절은 테스트 심 setQuerySchtasksForTests 보다 앞이다. 그래서 녹화용 페이크를 끼워도 /create 는 녹화기에조차 안 간다. 의도된 변이 테스트는 registerFreshWindowsSchedulerTask(..., { create }) 처럼 deps 로 자기 연산을 주입해야 한다. 프로덕션 동작은 가드가 꺼져 있어서 그대로다.

#2975 가 tests/service.test.ts 에 넣은 크로스 플랫폼 서비스 심 픽스처와는 같은 파일이지만 다른 구멍이다. 저쪽은 Windows 샤드 실패를 줄이는 픽스처이고, 이쪽은 라이브 스케줄러 변이 차단이다. 서로 싸우지 않고 위에 쌓인다. #2857 windowsSchedulerHomeScoped 는 inspectWindows 가 따옴표 친 런처가 지금 configDir 밖이면 다른 홈 주장으로 보는 소유권 규칙이다. 이 PR은 검사 규칙을 안 바꾸고, 테스트가 라이브 schtasks 변이를 못 하게만 막는다. types.ts / config.ts 분할 캠페인과도 안 겹친다. 닫고 리베이스할 대상이 아니다. 미리보기 배포도 이 패치와 무관하다.

회귀 테스트는 service lifecycle cleanup ordering 안에 한 줄기다. 유효한 신규 등록 XML 을 만들고 setQuerySchtasksForTests 로 호출을 녹화한 뒤 registerFreshWindowsSchedulerTask 를 deps.create 없이 부른다. 거절 메시지를 기대하고, observedCalls 가 빈 배열인지 본다. 가드가 없으면 녹화가 /create /tn opencodex-proxy ... /f 를 받는다는 빨간 증명도 본문에 적혀 있다. 에러는 schtasks 래퍼를 지나 WindowsSchtasksError 로 감싸지지만, access-denied 가 아니면 원문 메시지가 남는다. 그래서 toThrow 문자열이 맞다. elevate 경로로 가지도 않는다. reason 이 other 이기 때문이다.

라인 903-917 - querySchtasks 에 가드를 심는다. 첫 인자만 /query 허용. 대소문자 trim 처리. create/delete/run/end/change 는 전부 거절. 프로덕션에서는 isTestHomeGuardArmed() 이 false 라서 비용이 거의 없다.
라인 903 가드 위치 - setQuerySchtasksForTests 보다 앞이다. 전역 심으로는 변이를 흉내 낼 수 없다. 기존 windows-scheduler-install-verification.test.ts 는 /query 만 심에 넣어서 깨지지 않는다. 변이 테스트는 이미 deps.create / deps.elevate 쪽이다.
경로 src/service.ts runWindowsElevated / startElevateExecution - querySchtasks 를 안 타고 windowsSchtasks() 를 직접 올린다. 이 PR 가드 밖이다. UAC 가 떠서 CI 가 우연히 밟기는 어렵지만, 무장 테스트가 elevate 폴스루를 타면 여전히 라이브 변이가 가능하다.
tests/service.test.ts 새 테스트 - 거절과 녹화기 미도달만 잠근다. 가드 아래에서 /query 가 여전히 심·라이브로 가는 초록 핀은 없다. deps.create 주입이 계속 통하는 핀도 이 PR 안에는 없다(기존 등록 테스트가 간접 커버).
경로 #2857 inspectWindows / windowsSchedulerHomeScoped - 소유권·다른 홈 주장 규칙과 겹치지 않는다. 이 패치가 그 판정을 바꾸지 않는다.

메인테이너의 판단이 필요한 지점

• elevate / startElevateExecution 경로에도 같은 테스트 가드를 이번 PR에 넣을지, UAC 장벽만으로 후속으로 둘지
• 가드 아래 /query 통과 핀과 deps.create 주입 핀을 머지 전에 한 줄씩 더 넣을지
• 이 사고를 이슈 번호로 남길지, 테스트·주석만으로 기록을 끝낼지

너의 추천
머지한다. 라이브 Task Scheduler 를 테스트가 덮어쓰던 실제 구멍을 막고, 프로덕션 경로는 가드 off 로 그대로다. #2975 서비스 심 픽스처 위에 쌓이고 #2857 과는 축이 다르다. types/config 분할로 닫을 패치도 아니고 중복 가드도 아니다. elevate 우회는 후속으로 적어도 된다. Windows 포커스 스위트(tests/service.test.ts, tests/windows-scheduler-install-verification.test.ts, tests/test-home-guard.test.ts)가 초록이면 dev 에 넣으면 된다.

이 댓글은 grok-bot이 작성했습니다

[Ingwannu]

Approved exact head 1b760b5baa3f33470c085dcc111d0793ba4553f7 for the scoped Windows test-boundary change.

The latest head closes the original scheduler gap and the elevation bypass: armed tests reject machine-global schtasks mutations before the recorder/live launcher, and reject the default live elevation spawn while explicit injected launchers remain testable. Isolated focused verification passed 228/228 plus typecheck; Service lifecycle and React Doctor are green.

The remaining Cross-platform CI failures are both the repository-wide post-release base condition: package.json still says 2.36.0 after tag v2.36.0 moved to the release commit, so release-version-line.test.ts fails on every non-tagged PR head. This approval does not waive required CI. Do not merge until the dedicated dev version-line repair lands, this branch is refreshed if necessary, and exact-head required CI is green.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 83b1316ca3

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".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".