Skip to content

coderabbit.md

Latest commit

 

History

History
66 lines (51 loc) · 3.53 KB

coderabbit.md

File metadata and controls

66 lines (51 loc) · 3.53 KB
Raw

CodeRabbit Integration

CodeRabbit is an external AI code-review service. aidd integrates with it through the coderabbit skill, which runs the CodeRabbit CLI against a project's pending changes and triages every finding before anything is pushed. Whether the fixes are applied depends on the run's execution intent: apply-changes edits the working tree, review-only reports the changes it would have made. The integration is opt-in and requires tools you install and authenticate — aidd never holds CodeRabbit credentials.

Privacy note: the CLI uploads the diff to CodeRabbit's cloud for analysis. Only enable it for projects where that is acceptable.

Prerequisites

  • The CodeRabbit CLI, installed and logged in by you:

    curl -fsSL https://cli.coderabbit.ai/install.sh | sh
coderabbit auth login

  • Windows: the official CLI supports macOS, Linux, and WSL only — there is no native Windows build. Install and log in inside WSL; the skill probes native coderabbit first and falls back to running it through WSL against the project's /mnt/<drive>/... path automatically. If neither is available, the run stops with setup guidance rather than failing cryptically.

What the skill does

Run the coderabbit skill as a one-shot — coderabbit [app] [base-ref] — with the target application and an optional base ref. It:

  1. Resolves the base ref (the [base-ref] argument, else main locally or as origin/main, else master; with none of those it reviews only uncommitted changes) and confirms there is something to review.
  2. Runs coderabbit review --plain --base <base> when a base ref resolves, or coderabbit review --plain when only uncommitted changes are available. The review stays in the foreground and can take several minutes.
  3. Triages every finding against the actual code: fixes the ones judged real, dismisses false positives with a one-line justification each. Under review-only, the fixes are described rather than applied.
  4. Runs the project's quality gate (bun run smoke:qc when present, else the relevant build, typecheck, lint, format-validation, and test scripts).
  5. Reports fixed findings, dismissed findings with justifications, and the gate result.

It never pushes, opens a pull request, or posts review comments. Apart from the disclosed diff upload to CodeRabbit, its only possible project mutation is editing the working tree.

Costs and limits

CodeRabbit's free tier has daily review limits; paid plans meter usage per reviewed file beyond plan quotas. When the CLI reports a rate limit, the skill stops and surfaces it instead of retrying.

Where it fits

Use the skill for pre-push review inside pipelines — e.g. after a coding step, the way deepreview is used, but with CodeRabbit as an independent second reviewer. The bundled coding-spirit-coderabbit-document-changes recipe does exactly that: coding, a spirit review and its remediation, then the coderabbit skill review-only followed by a remediation step that consumes its findings, then document-changes.

The CodeRabbit GitHub App is a separate product that reviews pull requests in the cloud with no aidd involvement at all. It remains a reasonable choice for PR-based workflows, but aidd ships no skill for pulling its feedback back down; remediate those comments the way you would any other PR review.