| name | coderabbit-status |
| description | Report CodeRabbit's current review status for the current PR (or a given PR). Use when the user asks where CodeRabbit is, whether CodeRabbit has finished reviewing, if CodeRabbit is rate limited/paused/still running, or for a quick CodeRabbit status check. |
CodeRabbit Status
You report CodeRabbit's current review state for a pull request. This is a read-only, single-shot status check — you determine where CodeRabbit is in its review and explain it plainly. You do not resolve feedback, wait/poll, request reviews, or edit any files.
Org context (important for interpreting status): the working policy is one review per PR — the automatic review CodeRabbit posts when the PR opens. PinePeakDigital repos enforce this at config level (auto_incremental_review: false), so there a completed status reflects that single initial review and will stay completed against an older commit after later pushes rather than refreshing — that's expected, not a problem to fix, and a fresh review only happens if someone manually posts @coderabbitai review. Other orgs (narthur, seedtime, etc.) may still have incremental reviews on, so there CodeRabbit may auto-review pushes and completed can track the latest commit. When suggesting next steps, do not imply that manually re-requesting a review is routine.
What You Do
- Resolve the target PR (current branch by default, or a PR passed as an argument)
- Run this skill's
coderabbit-status.sh detector and report the result
- Translate the raw status into a plain-English explanation and a suggested next step
What You Don't Do
- Do not poll or wait for the review to finish — that's the
wait-for-review.sh script in the drive-pr skill. Report the status once and stop.
- Do not resolve threads, dismiss comments, request reviews, or post anything to the PR.
- Do not edit files or run a review loop.
- Do not switch branches unless the user passed an explicit PR argument (see Step 1).
If the user wants to act on CodeRabbit's feedback, point them at the drive-pr skill. If they want to block until the review completes, point them at wait-for-review.sh.
The Detector Script
This skill owns the status detector — the single source of truth for combining the check signal with CodeRabbit's living first-comment status document:
~/.claude/skills/coderabbit-status/coderabbit-status.sh
The drive-pr skill's wait-for-review.sh also consumes this same script (via a sibling-skill relative path), so keep its CLI ([--json]) and output shape stable. Do not reimplement the detection logic anywhere else — always call this script.
Workflow
Step 1: Resolve Target PR
If the user passed a PR number or URL (e.g. /coderabbit-status 123, /coderabbit-status #123, or a full PR URL), switch to it first — otherwise the status reflects whatever PR matches the current branch, which may not be what they meant:
- Extract the trailing integer as the PR number.
- Confirm it exists and capture its head branch:
gh pr view <number> --json number,headRefName,isDraft,url
- Check it out:
- Standard git:
gh pr checkout <number>
- GitButler workspace (current branch is
gitbutler/workspace): do not use gh pr checkout. Find the matching virtual branch with but status; if none is applied, stop and ask rather than reporting on the wrong PR.
If no argument was given, use the current branch as-is — do not change branches.
Step 2: Get the Status
Run the detector in JSON mode so you have the full picture (combined status, the two underlying signals, and any rate-limit wait time):
~/.claude/skills/coderabbit-status/coderabbit-status.sh --json
This returns:
{ "status": "...", "check_state": "...", "comment_state": "...", "wait_seconds": 0 }
If the script exits non-zero (e.g. "Could not fetch PR data"), there is no PR for the current branch — tell the user that plainly and stop. Do not guess.
Step 3: Report
Translate the status field into a one-line headline plus a short plain-English explanation, and suggest the natural next step. Use this mapping:
status | Headline | What it means / suggested next step |
|---|
not_started | ⚪ Not started | No CodeRabbit check and no comment yet. On a draft PR, CodeRabbit won't auto-review — a review must be requested. |
starting_up | 🟡 Starting up | Check is pending but CodeRabbit hasn't posted yet. Give it a moment. |
in_progress | 🟡 Reviewing | CodeRabbit is actively reviewing right now. Check back shortly, or use wait-for-review.sh to block until it's done. |
completed | 🟢 Review complete | CodeRabbit has finished its automatic review. If they want to act on findings, suggest the drive-pr skill. Note: in repos with incremental reviews off (e.g. PinePeakDigital) this stays completed on an earlier commit after pushes — expected, not a problem; elsewhere it may refresh per push. |
rate_limited | 🟠 Rate limited | CodeRabbit hit a rate limit on the initial review. Report the wait time from wait_seconds (e.g. "retry in ~Xm Ys") if it's > 0; it should resume on its own. |
timed_out | 🔴 Timed out | CodeRabbit's review timed out. Per the one-review-per-PR policy this isn't auto-retried; a fresh review is only obtained by manually posting @coderabbitai review (a deliberate exception). |
paused | ⏸️ Paused | Auto-reviews are disabled for this PR. A review would require manually posting @coderabbitai review — a deliberate exception to the one-review-per-PR default, not a routine step. |
Keep the report tight — headline, one or two sentences of explanation, and the suggested next step. Mention the underlying check_state/comment_state only if they disagree in a way worth flagging (e.g. the check passed but the comment still says reviewing) or if the user asks for detail.
For rate_limited, format wait_seconds into minutes/seconds when reporting the retry window.
Do not take any further action unless the user explicitly asks.
