Overview
- Available to Claude for Teams and Enterprise users in research preview, the feature integrates with GitHub and auto-runs on new pull requests once admins enable it per repository.
- A team of specialized agents analyzes changes in parallel, verifies suspected bugs to reduce false positives, ranks severity, and posts inline comments, while final approvals remain with human reviewers.
- Reviews prioritize code correctness over style and can read across the broader codebase for cross-file issues, assigning more agents for larger or more complex pull requests.
- Anthropic cites an average review time of about 20 minutes and token-based pricing that typically totals $15–$25 per review, with admin controls for monthly caps, repo-level settings, and cost analytics.
- Internal metrics show substantive PR comments rising from 16% to 54% and 84% of large PRs surfacing issues with under 1% of findings marked incorrect; light security checks are included, with deeper scanning handled by Claude Code Security.