ai-review-bot

🤖

Two bots, one deployment

A Claude bot (Anthropic) and a Codex bot (OpenAI) run from a single Vercel deployment. Each is a separate GitHub App with its own icon. Install both and every PR gets two independent expert opinions side by side.

⚡

Automatic on every PR

Reviews fire automatically when a pull request is opened or a new commit is pushed — no slash command needed. Both bots post independently after a short delay (default 7.5 min) that lets other reviewers like Code Rabbit finish first.

🔀

Five agents, one review

Bug detection, error handling, test coverage, security, and code quality agents run in parallel via Promise.allSettled(). Their findings are merged and deduplicated before posting — same line flagged twice becomes one comment.

🔇

Cross-bot deduplication

Before running, each bot reads all existing reviews on the PR — from the other bot and from external tools like Code Rabbit. Agents are instructed not to re-report findings that have already been raised, keeping reviews complementary rather than redundant.

🛡️

Conservative by design

When two agents disagree on severity, the more conservative finding wins. The final verdict is REQUEST_CHANGES if any agent surfaced a blocking issue, COMMENT otherwise. No false approvals.

📌

Diff-anchored comments

Every inline comment is validated against the actual diff before submission. Comments referencing lines not in the diff are dropped silently — the bot never errors a review over a bad anchor.

💬

Manual triggering

Comment /ai-review on any PR to request a review on demand — useful for PRs opened before the apps were installed, or to re-run with extra instructions like "/ai-review focus on the auth flow". Pass --force to re-review the same commit.

📦

CLI & npm package

Also published to npm: `npx ai-review-bot owner/repo` audits an entire codebase — no PR, no webhook needed. Posts findings as a GitHub issue. Use it on demand, on a schedule, or as a GitHub Action step in any CI workflow.

🔧

Pluggable skill frameworks

Each agent's review framework is a vendored Markdown file in skills/. Add a new framework by dropping in a .md file and adding one line to AGENT_SKILLS in review.ts.

ai-review-botAutomatic parallel AI code reviews