[kixelated]

Completes quest/m0/lifecycle-flakes.md: the lifecycle CI jobs no longer fail for reasons outside the code under test.

Windows EBUSY

Cause. On windows-11-arm hosted runners, GitHub's hosted-compute-agent starts an unsigned helper,
%TEMP%\provjobd.exe<N>. About every 15–35 s it opens the image of a running process. It first opens it
without delete sharing for 300–450 ms (RW-), then with delete sharing for about 3 s (RWD). If that scan
caught the check's last quest.exe run, rmSync(work) hit the RW- window while removing the check's copy at
work\tool bin\quest.exe, and failed with EBUSY. windows-2022 has no failures, and its Defender real-time
protection is off. The JS build in Check never failed because bun.exe lives outside the temp directory.

Reproduction. I used a throwaway diag/lifecycle-ebusy branch, now deleted. It looped the check on native
runners, running Sysinternals Handle when the check failed and polling it in the background:

• Before: 8 EBUSY in 390 runs on windows-11-arm (0 in 80 on windows-2022). Each time, the only thing left
  was tool bin\quest.exe, held by provjobd.exe. The failing rm started 0 ms after the last quest uninstall
  exited, inside a polled RW- window (06:22:23.003–23.461, rm at 23.067). It was removable a moment later.
• After: 0 failures in 300 runs of lifecycle-check.mjs and 120 runs of install-check.ps1 -Method archive on
  windows-11-arm.

Fix. Nothing deletes an executable the check just ran. There are no retries, sleeps, or ignored errors.

• lifecycle-check.mjs runs the caller's binary in place instead of copying it into the directory it deletes.
• lifecycle-check.sh copies the binary into "$work/tool bin/" first, which keeps the Linux/macOS coverage of a
  binary path with spaces.
• install-check.ps1 puts its scratch under RUNNER_TEMP, which the runner empties after every job, and no
  longer removes it itself. It already installs to a path with spaces (quest install <guid>). The script is
  CI-only (Release and Install check), so it now refuses to run without RUNNER_TEMP.

macOS queue timeout

When hosted macOS runners are short, the lifecycle (macos-14) job waits in the queue until timeout-minutes
cancels it. That happened on kixelated/quest#191 (queued 15:40:13,
cancelled 15:55:14, no runner). The job then shows as failed. Pull requests now run lifecycle on
ubuntu-latest only. macos-14 still runs on pushes to main and on manual runs.

macOS stays covered on pull requests by Release. Its native check (aarch64-apple-darwin) and
check (x86_64-apple-darwin) jobs run the lifecycle check on any PR that touches src/, scripts/, or the
lockfiles. The only macOS check missing from PRs is the flake's own Nix build on Darwin, which now runs on main.

Choices made without asking

• Of the three macOS options, I picked "drop it from pull request runs". A shorter queue limit still shows a
  failure. A cancelled job can't be made to report as skipped.
• install-check.ps1 now needs RUNNER_TEMP, so running it locally is unsupported. It never had any local use.
• Check's native Windows job still runs dist/quest.js from the checkout. That path has no spaces, so the
  spaces coverage on Windows now comes from the native binary that Release and Install check run.

Validation

• just check and just test passed locally (both lifecycle builds included), as did quest check --base origin/main.
• Release passed on this branch, including native windows-11-arm and macOS lifecycle checks.
• The diagnostic loops are described above.

Suggested follow-ups

• Release's macOS check jobs can hit the same queue timeout. On a PR, that shows as a failed (non-required)
  Release check.
• Native Windows unit-test flakes, unrelated to the lifecycle, have caused most recent Check failures:
  test/gate.test.ts "concurrent processes never admit past the headroom" (5 of the last 15 failed Check runs),
  test/move.test.ts "refuses a claimed quest" timing out at 5 s, and test/start.test.ts timeouts.
• quest start failed once with "git push to origin failed" even though the remote printed the "Create a pull
  request" hint. A retry worked. The error dropped the porcelain stdout, which would have explained it.

🤖 Generated with Claude Code

(Written by Claude Opus 5.5)

[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 2026-10-24T07:23:23.686645Z	7c8f9c2	Draft marked ready
🔒 Security Review	✅ Completed 2026-10-24T07:23:57.021077Z	7c8f9c2	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.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 7c8f9c206f

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".

[kixelated]

Merging as a squash at 7c8f9c2.

• Windows EBUSY: the root cause is fixed. Nothing deletes an executable the check just ran, and there are no retries or sleeps.
• macOS: the lifecycle job runs on pushes to main and on manual runs. On PRs, Release covers macOS for changes under src/, scripts/, and the lockfiles.
• Maintainer decision on the Codex thread: keep this as is. PRs that touch only Nix files get no pre-merge macOS check. Thread resolved.
• CI passed (16 checks). quest check --base origin/main passed on the head.

(Written by Claude Opus 5.5)