fix: stop replay retries after cleanup failures

[thymikee]

Summary

Replay test retries now stop when the preceding attempt cannot close its session. A rejected close response is promoted to the same cleanup failure path, and the failed attempt is classified as infrastructure so a later attempt cannot contend with its retained device claim.

Adds regression coverage for the macOS open "System Settings" shape: a failed attempt whose close is rejected runs exactly once instead of reaching DEVICE_IN_USE on the next attempt.

Rebased onto current origin/main: this PR now contains one cleanup-retry commit and four targeted source/test files only.

Motivating CI evidence: merged PR #1952 macOS run 32734629087, job 97454414587; and PR #1994 macOS run 32734431023, job 97453800202.

Validation

• Regression was observed red before the production change, then green.
• Focused replay-test and daemon suite tests passed (23 tests).
• pnpm test:replay:macos passed against the local macOS System Settings replay after the rebase.
• pnpm check:affected --run passed after the rebase (GitHub-authoritative lanes remain pending on this head).

Docs and skills unchanged: the retry/session lifecycle contract is internal and existing guidance already covers it.

[thymikee]

Review of exact head 8b9cfb0807b6be4637321ddebc62d9dfa449008c:

The functional fix is sound and sits at the right ownership seam: a rejected close response becomes a cleanup error, the attempt is classified as infrastructure, and the existing scheduler stops before retrying into a retained device claim. Both regression tests are non-vacuous.

The PR is not ready because its branch topology is wrong: it conflicts with main and carries 13 unrelated iOS snapshot commits (31 files, +939/-148) ahead of the one cleanup-retry commit. Rebuild/rebase it onto current main as only the cleanup-retry change and its four targeted source/test files, then rerun exact-head CI and re-stamp the macOS replay evidence. Do not apply labels.

[github-actions]

Size Report

Metric	Base	Current	Diff
JS raw	2.40 MB	2.40 MB	+268 B
JS gzip	805.1 kB	805.3 kB	+106 B
npm tarball	929.5 kB	929.6 kB	+119 B
npm unpacked	3.22 MB	3.22 MB	+268 B

npm unpacked components

Component	Base	Current	Diff
JS / dist source	2.55 MB	2.55 MB	+268 B
Apple runner source/project	570.6 kB	570.6 kB	0 B
macOS helper source	54.5 kB	54.5 kB	0 B
Android helper artifacts	0 B	0 B	0 B
Other package files	44.5 kB	44.5 kB	0 B

Startup median (7 runs, lower is better):

Scenario	Base	Current	Diff
CLI --version	21.5 ms	21.4 ms	-0.1 ms
CLI --help	61.4 ms	61.7 ms	+0.4 ms

Top changed chunks:

Chunk	Raw diff	Gzip diff
dist/src/session2.js	+268 B	+106 B

Top changed packed files

Packed file	Base	Current	Diff
dist/src/session2.js	233.2 kB	233.5 kB	+268 B

[thymikee]

Re-review of exact head 469455f83d34387341becdd49c0c4777dbdc0b5a: clean and ready for human review. The prior topology blocker is resolved: one commit, four targeted files, mergeable, with the cleanup failure classified at the owning attempt seam and non-vacuous regression coverage. Exact-head static, coverage, integration, and macOS smoke are green; the macOS lane ran the System Settings replay successfully. iOS smoke is still in progress and is the only pending check.

[github-actions]

PR Preview Action v1.8.1
Preview removed because the pull request was closed.
2026-08-24 14:41 UTC