HQ Rabbit

A merge gate for pull request reviews. Each developer's own local agent reviews the diff and submits the result; this server checks the submission and turns the hq-rabbit check green. The server runs no AI model itself.

How it works

  1. You open or push to a pull request. HQ Rabbit marks it with a blocking hq-rabbit check and posts a comment explaining what to do.
  2. You run the review skill in Claude Code (/hq-rabbit:review). A fresh subagent, with none of your session's context, reviews the change following the review instructions.
  3. The skill posts the result as one PR comment: a marker and a JSON payload that follows the review contract.
  4. HQ Rabbit validates the payload: it must fit the contract, name the PR's current head commit, and cover every changed file.
  5. If it is valid, the bot reposts the findings as its own review, removes your comment and turns the check green. If not, it replies with the exact errors so the agent can fix them and resubmit. A new push resets the check.

Where to go next

  • Install: set up the Claude Code plugin and run a review, or submit one without Claude Code.
  • Review skill: what the skill does and the full review instructions.
  • Contract: the submission format, limits and error codes.
  • Admin: for maintainers (requires a passkey).