🔍 Anthropic has introduced Code Review for Claude Code - a multi-agent review of pull requests.
When a pull request is opened, Claude launches a team of specialized agents. Five independent reviewers analyze changes from different perspectives: compliance with CLAUDE.md, bug detection, analysis of git history, checking previous PR comments, and verification of code comments. Each finding is rated on a scale of 0-100, and only high-confidence issues (threshold 80+) are published as comments - this sharply reduces the number of false positives.
Anthropic's internal results are impressive: code output per engineer has increased by 200% over the year. Before Code Review, only 16% of PRs received substantive comments - now it's 54%. For large PRs (1000+ lines), 84% receive remarks, averaging 7.5 issues per PR.
There are three levels of severity: red (bug, fix before merge), yellow (minor, not blocking), purple (bug already in the code, not from this PR). Configurable via CLAUDE.md and REVIEW.md in the repository.
The documentation states the cost is $15-25 per review of a single PR on average. Even considering that 5 agents work in parallel for 20 minutes, it seems pricey. Available in research preview for Team and Enterprise plans.