[BUG] Per-PR "Auto-fix CI & address comments" toggle acts as a one-shot: stays visually checked but stops re-dispatching after the first red run; only a manual uncheck/recheck re-arms it
Preflight Checklist
- [x] I have searched existing issues and this is not a duplicate. Closest matches, all distinct from this report:
- #68083 — the global "Auto-fix CI and address comments" desktop toggle never applies to gh-created local-session PRs and doesn't persist. This report is about the per-PR CI-monitoring popup toggle, which did dispatch once, then went inert while still showing checked.
- #69161 — the per-branch CI-monitoring panel's "Auto-fix CI & address comments" toggle is global/shared across branches (a scoping bug). Same control as this report, different defect (re-arming vs. scoping).
- #52474 (closed) — "doesn't dispatch to local session."
- [x] This is a single bug report.
- [x] I am using the latest version available to me.
Environment
- Surface: Claude Code desktop app — FleetView PR chip → "CI monitoring" popup (screenshots below).
- Desktop app version: 1.20186.1 (df1d8a), built 2026-07-10T21:55:12Z.
- OS: macOS 26 (Tahoe), 26.4.1 (build 25E253).
- Exact control: the per-PR "CI monitoring" popup opened from the PR chip for PR #2 (branch
autofix-global-test). The checked box is "Auto-fix CI & address comments"; "Auto-merge when ready" was left unchecked. The popup header showed "✕ Failed 2".
What's Wrong?
The per-PR "Auto-fix CI & address comments" toggle presents as a persistent on/off setting but behaves like a one-shot trigger. After it dispatches once, a subsequent failing (red) CI run on the same PR produces no dispatch at all, even though the toggle still shows as checked. The only way to make it fire again is to manually uncheck and re-check the box.
The checked state is therefore a false affordance: it signals "armed and watching," while the feature is inert until re-toggled. A user who sees the box still ticked has no reason to suspect their next red push will be silently ignored.
Steps to Reproduce
On a disposable fixture repo with a single unit test that fails on an obvious bug:
- Open a PR whose CI check is red. (Here the PR was created locally via
gh pr create.) - From the PR chip, open the "CI monitoring" popup and check "Auto-fix CI & address comments." → A
ci-monitor-eventis dispatched to the session ("CI check failed … fix the failing check"). ✅ Works. - Push a new commit to the same PR that is still red with a different failure. New CI run completes, red.
- Observe: no new dispatch arrives, though the toggle still shows checked. ❌
- Manually uncheck then re-check the toggle. → The dispatch fires again. ✅ (Confirms the arm is consumed on first fire and only a re-toggle re-arms it.)
Expected
While the toggle is checked, every subsequent red CI run on the monitored PR should re-dispatch (or the docs/UI should make explicit that arming is single-shot and consumed per fire). The displayed checked state should correspond to an actually-armed state.
Actual
The arm is consumed after the first dispatch. Later red runs do not re-dispatch. The box remains visually checked throughout, misrepresenting the feature as active.
Evidence
Fixture PR #2 (butterworth-advisors/ba-toy-test-exercises-autofix), two red pushes, identical CI workflow:
| Commit | Bug | Failure signature | Push run | PR-check run |
|---------|----------------|---------------------|-------------|--------------|
| 72ad3a8 | a - b | assert -1 == 5 | 29153020961 | 29153026061 |
| 842ddb8 | a + b + 1 | assert 6 == 5 | 29154245323 | 29154246735 |
- First dispatch arrived only after the toggle was checked.
- The second red push (842ddb8) produced no dispatch on its own.
- A dispatch fired again only after a manual uncheck/recheck.
Impact
Silent gap. A user relying on the visible checked state pushes broken code, sees the box still ticked, assumes Auto-fix will handle it — and gets nothing. This is arguably worse than a toggle that plainly doesn't work (#68083), because the UI actively signals that it does.
<!-- Screenshots (checked "CI monitoring" popup, and the PR chip it was opened from) can be attached in the GitHub web UI; gh cannot upload inline images. -->
(authored by my Claude of course)
This issue has 1 comment on GitHub. Read the full discussion on GitHub ↗