You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Copilot's review of PR #46 at validated head 61f5039bf152f36f5548f0a80f6712e5b7269c81 reported Findings: None and confirmed all four prior concrete findings resolved, but its summary stated that five moderate concerns remain across cancellation reasons, cache invalidation, stale errors, and poll ordering. It provided no line locations, scenarios, or inline threads for those concerns.
Investigation scope
For each category, identify a concrete state transition and controlled interleaving before changing code:
preserve the original timeout/cancellation reason through coordinator, gateway, durable attempt, HTTP response, and UI;
invalidate status caches after successful and refused irreversible mutations, and reject stale in-flight cache publication;
prevent stale polling or action errors from overwriting a newer successful UI state;
prove poll ordering cannot apply an older observation after a newer action, retry, or lifecycle transition;
identify the fifth concern if it is distinct from the four categories named in the summary.
Acceptance
Add a failing-before controlled regression for every reproduced issue, then a passing-after fix.
If a category cannot be reproduced, record the examined paths and evidence rather than patching vague wording.
Re-run unit, typecheck, and applicable browser race suites.
Reopening: #171 closed this automatically through its original "Closes #48" link, but it covers only the server part. The six server-side defects landed in #171 (49f03db). Still open here: the two browser polling defects (duplicate poll chains letting an old poll's error or answer overwrite newer state in web/public/app.js), planned as a PR stacked on F6b (#161), which rewrites that polling. Follow-up #174 covers the rebase conflict resolver's trust-read timeout.
Activity
Cover parallel inspection stops; handle the already-fixed check early (…
Merge origin/main (#170) into feat/48-merge-lifecycle-concerns
Align the drain comment and replay signal docs with #48 after #170
Replay a click's saved outcome even after its own abort (#48)
Keep a runner trust timeout through shutdown; cover two traced paths (#…
Test trust-read stop reasons in merge preparation and publishing (#48)
Fix the #48 regression list punctuation and name what each test covers
Add two review lessons from #48 to AGENTS.md; link follow-up #174
Never offer an action from a merge click's pre-command readiness (#48)
Keep merge stop reasons and GitHub answers through every layer (#171)
mchwang commentedon Oct 14, 2026
Reopening: #171 closed this automatically through its original "Closes #48" link, but it covers only the server part. The six server-side defects landed in #171 (49f03db). Still open here: the two browser polling defects (duplicate poll chains letting an old poll's error or answer overwrite newer state in
web/public/app.js), planned as a PR stacked on F6b (#161), which rewrites that polling. Follow-up #174 covers the rebase conflict resolver's trust-read timeout.Merge the updated F6b (with F6a and main) into PR B
Merge F6b 79ed50c into the #48 browser fixes
Merge F6b 7fe156d into the #48 browser fixes
Merge F6b 7b1b5fa into the #48 browser fixes
Check the merge control, not the cleared banner, after the late old a…
Merge F6b 82661d6 into the #48 browser fixes
Merge F6b 0eb8a21 into the #48 browser fixes
Merge F6b 45bff7a into the #48 browser fixes
Merge F6b cffdf4c (now on main) into the #48 browser fixes
Merge F6b d1f99fc into the #48 browser fixes
Merge F6b 25444dd into the #48 browser fixes
Merge F6b fbfe4ce into the #48 browser fixes
Merge F6b 105ea74 into the #48 browser fixes
Merge F6b a76b453 into the #48 browser fixes
Merge F6b into the #48 browser fixes