tests/skills/ce-work-cross-model-integration.test.ts flakes on main — which test fails varies by environment, so this looks like a shared timing/state dependency across the serial-transaction tests rather than one bad assertion.
Reproduction (pristine main, commit at the time of writing, no local changes):
- Linux, bun 1.x:
bun test tests/skills/ce-work-cross-model-integration.test.ts fails 2/2 runs in same-base disjoint wave terminalizes together and lands as separate controlled host commits (peer_job_runner wait returns failed where done is expected, the same expect(...).toBe("done") shape as the CI failure below).
- GitHub Actions
test job (ubuntu) on a PR whose diff touches only ce-plan markdown and its contract test failed instead in ce-work serial cross-model transaction > scope expansion remains a successful detached result for host inspection — same file, same wait -> "failed" signature: https://github.com/EveryInc/compound-engineering-plugin/actions/runs/33956247463/job/101280011018
Since the failing PR diff cannot influence ce-work integration machinery (docs-only change), and main fails locally on a sibling test in the same file, the flake lives on main. I have not root-caused it; happy to dig in and send a fix if the maintainers want one.
tests/skills/ce-work-cross-model-integration.test.tsflakes onmain— which test fails varies by environment, so this looks like a shared timing/state dependency across the serial-transaction tests rather than one bad assertion.Reproduction (pristine
main, commit at the time of writing, no local changes):bun test tests/skills/ce-work-cross-model-integration.test.tsfails 2/2 runs insame-base disjoint wave terminalizes together and lands as separate controlled host commits(peer_job_runner waitreturnsfailedwheredoneis expected, the sameexpect(...).toBe("done")shape as the CI failure below).testjob (ubuntu) on a PR whose diff touches only ce-plan markdown and its contract test failed instead ince-work serial cross-model transaction > scope expansion remains a successful detached result for host inspection— same file, samewait -> "failed"signature: https://github.com/EveryInc/compound-engineering-plugin/actions/runs/33956247463/job/101280011018Since the failing PR diff cannot influence ce-work integration machinery (docs-only change), and main fails locally on a sibling test in the same file, the flake lives on main. I have not root-caused it; happy to dig in and send a fix if the maintainers want one.