retriggering-ai-reviews
Use when a pull request has no CodeRabbit, cubic, or Qodo review after a push, a bot replies that the review was rate limited, that the review limit was reached, that the review was skipped because th
SKILL.mdplugins/react-frontend-sdlc/skills/retriggering-ai-reviews/SKILL.md
GitHubRetriggering AI reviews
Profile keys consumed
review.coderabbitmake.pr_commentsci.required_checks
review.coderabbit says whether a CodeRabbit review is expected at all — when it is disabled,
record the CodeRabbit steps SKIPPED and fall back to the reviewers that do run. Thread listing goes
through the target mapped by make.pr_comments; skip that step with a recorded note when it maps to
null and list threads with the GitHub CLI instead. ci.required_checks is the check set that must
have registered before a review is requested.
Overview
Frontend repositories in this plugin's scope rely on bot reviewers that do not always run on their own: CodeRabbit auto-review is often paused at the organization level and refuses oversized or rate-limited PRs, cubic runs only when asked, and Qodo stops entirely when its subscription lapses. The remedy is a mention comment on a PR that is already in its final shape — plus knowing which reviewers no mention can reach.
When to use
- A PR has CI running but no review from
coderabbitai,cubic-dev-ai, orqodo-code-review. - CodeRabbit replied
Review rate limited.,Review limit reached, orReview skipped: N files exceed the limit of 100. - Qodo posted that reviews are paused because the subscription is no longer active.
- Several PRs are being landed in one session and each needs a fresh AI review.
- A branch-protection rule wants a CodeRabbit approval and none has been posted.
- Not for: qlty and SonarCloud. Both post their own summary comment automatically and have no mention command; they only re-run on a new push.
Applicability by repository shape
- React SPA shape (feature modules under the source root, a bootable app, an aggregate CI
target): yes — CodeRabbit, cubic, Qodo, and qlty all comment; the target mapped by
make.pr_commentslists threads. - Next.js app shape (routed pages, no aggregate duplication gate): yes — the same four bots; the
target mapped by
make.pr_commentslists threads. - Component-library shape (Storybook-first, no bootable app, published package): yes — the same
four plus SonarCloud; there is no mapped
make.pr_commentstarget, so list threads withgh api graphqldirectly.
Quick reference
| Reviewer | Trigger comment | Notes |
|---|---|---|
| CodeRabbit | @coderabbitai review |
incremental; about one per hour per repo |
| CodeRabbit | @coderabbitai full review |
re-reviews already-reviewed commits |
| CodeRabbit | @coderabbitai approve |
resolves threads and approves; no merge |
| cubic | @cubic-dev-ai review |
replies "I have started the AI code review" |
| cubic | @cubic-dev-ai approve |
blocked while cubic threads stay unresolved |
| Qodo | none while paused | lapsed subscription; report and skip |
| qlty | none | auto-comments a summary; re-run by pushing |
| SonarCloud | none | component-library shape; comment per push |
Start the comment with the mention — text before it can stop the bot parsing it, while trailing
words on the same line are fine (@cubic-dev-ai review this PR starts a review).
@coderabbitai review skips commits it has already reviewed, so reach for
@coderabbitai full review after a long gap or a dismissed review.
Procedure
Pre-flight the PR before asking, so the bot reviews the final diff:
- not a draft (CodeRabbit auto-review skips drafts);
- at or under 100 changed files — above that CodeRabbit answers
Review skippedand no mention helps; split the PR or change its base branch; mergeableis true (gh pr view <n> --json mergeable);- earlier threads answered — CodeRabbit will not approve while threads are open;
- the full check set named by
ci.required_checkshas registered (30-plus checks in the React SPA and Next.js shapes) with no failure; pending is fine.
Post the mention as its own comment:
gh pr comment <n> --body '@coderabbitai review' gh pr comment <n> --body '@cubic-dev-ai review'Wait for the bot's acknowledgement, then poll for the review (two to five minutes). Address findings, push, and re-request only after the push.
For several PRs, space CodeRabbit requests about an hour apart. When it answers
Review limit reachedit names the wait itself (Next review available in: 20 minutes) and that figure beats the hourly rule of thumb. cubic has no such limit. Keep aPR / checks passing / reviewer statetable and update it after each bot response.When findings are addressed and CI is green, post
@coderabbitai approveif the repository requires its approval. It resolves the threads and records an approving review; the squash merge stays a separate step.When a reviewer cannot run (Qodo subscription lapsed, cubic refusing a bot-authored PR on the free plan, CodeRabbit out of credits), say so in the PR summary and fall back to the reviewers that did run plus a local review. Never treat a missing review as an implicit pass.
Common mistakes
- Requesting before CI has registered — the bot reviews a diff that then changes; wait for checks.
- Firing
@coderabbitai reviewon every PR at once — requests queue silently; space them an hour. - Putting the mention mid-sentence — the bot ignores it; start the comment with it.
- Re-mentioning after
Review skipped: N files exceed the limit of 100— no mention lifts that; reduce the PR to 100 files or fewer. - Retriggering Qodo while it reports a lapsed subscription — nothing happens; report and move on.
- Waiting for a mention-driven reply from qlty or SonarCloud — read the summary comment they already posted and their check details instead.
- Treating
@coderabbitai approveas a merge command — it records an approval; merge with squash as the repository requires.