CodeRabbit reviews via comenq

Updated Oct 17, 2026

The comenq-coderabbit skill requests CodeRabbit reviews through the managed comenq queue and carries the review-response loop through evidence-backed convergence. install-skills copies every immediate skill directory into the agent skill paths, so the skill installs with the rest of the repository's skills; keep approved hosts, credentials, and posting identities in deployment configuration rather than in the skill itself.

Verify that the PR is ready before requesting a formal review. Inspect the pending queue, current CodeRabbit check, and recent activity first. Await an active or automatic review; use the queue for a current review rate limit or separately justified recovery, not an unanswered assessment conversation. Route new full reviews and retries through the queue rather than a direct GitHub comment or a personal bot identity:

comenq list
comenq hist -n 20
comenq put OWNER/REPO PR_NUMBER "@coderabbitai review"

Focused replies to existing findings or pre-merge rows are a different operation and use only the project's already authorized reply route. A posted request, a completed review, a resolved thread, an approval, green checks, and a successful integration are separate facts; the skill never substitutes one for another, and it stops at an andon rather than guessing a repair. See failure modes and recovery for the incident-derived cases and their exit evidence, and evidence and rehearsal for the handoff and reconciliation templates and the offline rehearsal scenarios.

comenq put

existing activity

current candidate and scope verified

fix, rebut, or scope findings

threads and rows read back

approval and required checks on head

rebase or later commit

CandidatePrepared

RequestPending

ReviewActivityFound

ReviewPosted

RequestFailed

InspectionIncomplete

InspectionComplete

FindingsPending

DispositionsComplete

ReviewStateReconciled

MergeEligible

CandidateStale

Merged

IntegrationVerified

IntegrationFailed

Andon

Figure 1: the review lifecycle. A prepared candidate either receives a queued comenq put request or already has review activity. A failed request leaves inspection incomplete and stops at an andon rather than counting as a clean review. A posted review, or activity verified against the current candidate and scope, reaches inspection complete; findings then need dispositions and reconciliation, which reaches merge eligibility only when approval and required checks hold on the head. A rebase or later commit makes the candidate stale and returns it to preparation. A merge is verified separately from integration, and an integration failure is another andon stop.