Code & Git
AI review & rule sets
A review pipeline that applies your team's rules, fact-checks its findings and leaves them as drafts on the exact lines.
Run a review
In a review, open the AI tab and press Run AI review (also in ⇧⌘P). Choose depth — quick, normal or deep — and add an optional focus ("check the transaction boundary"). Findings land on the diff lines as draft comments (merge requests) or notes (local changes), with severity and category. Discard what you disagree with; post the rest with your review.
The pipeline
A review is a flow of steps, each reported with its timing:
- Select files — skips generated, vendored and secret files.
- Resolve rules — the rule set and
rule.jsonthat apply. - Group related files — so the reviewer sees what belongs together.
- Review — Claude Code reviews each group, two at a time, read-only.
- Fact-check — drops findings the diff disproves.
- Drafts — on the exact lines; nothing is posted.
| Flow | When |
|---|---|
| Open Code Review process · Claude Code | Default; the steps above |
| Quick · Claude Code | One session, no fact-check; fastest |
| Open Code Review CLI · own LLM | Runs the ocr CLI with its own LLM configuration |
The file selection and rules of Open Code Review run inside Beetool, so the default flows need only Claude Code.
Rule sets
Settings › AI review › Rule sets. Collect rules from .md and .txt files (or a folder) and pasted notes into named sets, scoped by glob, and choose which set each repository uses. Linked files are re-read on every run: edit the document and the next review follows it. Which rules apply to a file? checks a path.
The assistant and Fix with AI use the same set.
Drafts in merge requests
Settings › AI review › Drafts in MRs: minimum severity, maximum number of findings, an "AI · severity · category" tag, and fixes formatted as GitLab or GitHub suggestion blocks the author applies with one click.