Claude Code Review is a new feature that dispatches a team of AI agents to perform deep, thorough code reviews on every pull request (PR). Modeled on the system used internally at Anthropic, it addresses the bottleneck of human code review by catching bugs that often slip through quick skims. The system scales with PR complexity—large changes get more agents and deeper analysis, while trivial ones receive a lightweight pass. Reviews typically take around 20 minutes and are billed on token usage, averaging $15–25 per review. Admins have control via monthly spend caps, repository-level enablement, and analytics dashboards. Key features include:
- Multi-agent parallel review: Agents work in parallel to identify bugs, verify findings to reduce false positives, and rank issues by severity.
- High-signal output: Results are presented as a single overview comment on the PR plus in-line comments for specific bugs.
- Internal validation: Used on nearly every PR at Anthropic, increasing substantive review comments from 16% to 54% of PRs.
- Cost and control: Admins can set monthly organization caps, enable reviews only on selected repositories, and track usage via analytics.
- Research preview: Available for Team and Enterprise plans, with no configuration needed for developers once enabled.
Use Cases:
- Development teams: Catch critical bugs in production services, such as authentication failures, before merge.
- Open-source projects: Identify latent issues in adjacent code touched by PRs, like type mismatches in encryption key caches.
- Enterprise codebases: Scale code review processes as developer output increases, ensuring thorough coverage of shipped code.
