Skip to content

Recurring bot quota/trial notices (Codex usage limits, Qodo trial, CodeRabbit review-limit) still block approval — add info_status_patterns; fix stale maintainer-resolve refusal message #1932

Open

Description

@rachel-petry

Summary

With #1918 (PR #1920) live, a clean SonarCloud pass no longer blocks approval. But the other recurring operational notices from registered review bots still do. They repeat on every push while a bot is out of quota, and nothing can clear them without dev-lead:

scripts/maintainer-resolve-comment.sh deliberately refuses these, because its bot path is restricted to sources with an info_status_pattern and only sonarqubecloud has one. So a maintainer has to hide each notice by hand in the GitHub UI. On PR #1907 (2026-09-25) I did exactly that for four Codex notices, posting a maintainer-resolve-marked reply first.

Acceptance criteria

  1. Add info_status_pattern rows for the exact, anchored text of each registered bot's quota, rate-limit or trial notice, where that notice carries no code finding. Start with Codex's usage-limit notice, then Qodo's trial-ended notice and CodeRabbit's review-limit notice. Use fixtures from real comments. Coordinate with cubic refusal detection: tighten to the real trial-ended notice and classify refusals in formal reviews (after trial ends ~2026-09-29) #1930 (cubic).
  2. Each pattern matches only the notice, never a genuine review from the same bot, with tests for both.
  3. Fix the refusal message in maintainer-resolve-comment.sh. It still says the path accepts "a REGISTERED REVIEWER BOT's comment", but after the feat: implement issue #1918 — Approval gate can never clear bot status comments — SonarCloud re-posts 'Quality Gate passed' on every push and #1911's resolve script is self-scoped, so the #1894 queue is still deadlocked #1920 hardening it accepts only registered info-status bots. A registered finding bot such as Codex is refused while the message implies it is eligible.
  4. Remember the rollout: the gate runs inside pr-review, so a pr-review/v1-next cut (and ring promotion) is needed before this takes effect.

Activity

  1. github-actions commented on Sep 25, 2026

    @github-actions
    Contributor

    Thanks for the detailed write-up and for including the real-PR evidence — that context makes the acceptance criteria very actionable. One thing that would unblock the pattern work fastest: could you paste the exact, full verbatim text of the Qodo trial-ended notice and the CodeRabbit review-limit notice as they appeared in those real comments? The Codex text is partially quoted above, but having all three anchored strings up-front means the patterns (and the matching tests) can be written against real fixtures without anyone needing to reproduce a quota event.

  2. don-petry commented on Sep 27, 2026

    @don-petry
    Collaborator

    Still happening as of 2026-09-27, on #1958. Codex (usage limit) and Qodo (trial ended) posted their notices on the first push. CodeRabbit then hit its review limit after one review ("0 remain", next in about 53 minutes). pr-review posted pr-review-agent rate-limited … auto-approval is withheld on every push (01:45, 01:50, 01:53, 02:00, 02:04 UTC), each time moving the re-review about an hour out. The PR was green and fully reviewed at 1ad231f by 02:09, but it only got approved at 02:14, once the sweep happened to run. Five fix pushes in 20 minutes each reset the approval clock. Restoring or retiring the bots is tracked in #1914.


    Generated by Claude Code

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions