[marcusrbrown]

The createSpawnRunner timeout starts counting at spawn. On a loaded host it could fire before a freshly started Node child had registered its SIGTERM handler or written its pid file. The grace-period test then failed on a missing marker, and the grandchild tests failed with ENOENT on the pid file.

Seven tests need their child to be running before the timeout fires. They now use a 4 s timeout (CHILD_STARTUP_TIMEOUT_MS), and their elapsed bounds are derived from it. The grace-period test also checks that the child wrote a ready file, and the grandchild test waits for the pid file before reading it. Test files only.

Measured on the file:

• Alone: 18/20 passing before, 20/20 after.
• Under load (8 busy loops plus a full suite): 10/10 after.

The file now takes about 57 s per run, up from about 36 s, because five tests wait out the 4 s timeout.

[fro-bot]

Verdict: PASS

Test-only change to .opencode/impeccable/hook-bridge.integration.test.ts. I read the diff and the surrounding helpers (waitUntil, readPid, expectProcessGone, trackedFile). I did not run the suite.

Blocking issues

None

Non-blocking concerns

• The affected tests now rely on the child starting inside 4 s, not 0.4–1 s. This is a better margin, but still a fixed wall-clock bound. If the flakes come back on a heavily loaded host, a startup-gated approach would be sturdier. For example, the fixture could signal readiness and the test could start the clock from that signal. The ready-file assertion in the grace-period test already does part of this.
• waitUntil (line 120) does not throw on budget exhaustion. In the exited-child grandchild test, the expect(existsSync(grandchildPidFile)).toBe(true) that follows it (line 445) is what fails, so the assertion is correct. It is also an easy line to delete as redundant later. The comment there already explains why it stays.
• Suite runtime goes up by about 20 s (about 36 s to about 57 s), as the PR description says. Per-test 15_000 budgets still cover the worst case of 4000 + grace + 2000 slack. If this file gets slower, consider a shared constant for the 2000 slack.

Missing tests

None. The change fixes test reliability only. It adds the ready-file check, which stops the grace-period test from passing or failing on a missing marker alone, and the pid-file wait. No production code or security boundary is touched. There are no dependency, workflow, or secret-handling changes.

Risk assessment (LOW)

Test-only change with no runtime, API, auth, or dependency impact. The worst outcome is slightly slower CI, and the timing bounds are derived from CHILD_STARTUP_TIMEOUT_MS so they stay consistent if it is tuned.

---

Run Summary

Field	Value
Event	pull_request
Repository	fro-bot/dashboard
Run ID	40984858305
Cache	hit
Session	ses_e759e4ddaffeJl23tGE05y6a24