Claude Code Adds Multi-Agent Code Review System
Code Review for Claude Code
Anthropic has introduced Code Review for Claude Code, a multi-agent review system modeled on their internal processes. Available in research preview for Team and Enterprise plans, it's designed to address the code review bottleneck as code output per Anthropic engineer has grown 200% in the last year.
How It Works
When a PR is opened, Code Review dispatches a team of agents that look for bugs in parallel, verify bugs to filter out false positives, and rank bugs by severity. The results appear as a single high-signal overview comment plus in-line comments for specific bugs. Reviews scale with PR size: large or complex changes get more agents and deeper reads, while trivial ones get lightweight passes.
The average review takes around 20 minutes based on Anthropic's testing. The system won't approve PRs—that remains a human decision—but aims to close the gap so reviewers can better cover what's shipping.
Performance Metrics
- Before implementation: 16% of PRs got substantive review comments
- After implementation: 54% of PRs get substantive review comments
- Large PRs (over 1,000 lines changed): 84% get findings, averaging 7.5 issues
- Small PRs (under 50 lines): 31% get findings, averaging 0.5 issues
- Engineer agreement rate: less than 1% of findings marked incorrect
Real-World Examples
In one case, Code Review flagged a one-line change to a production service as critical—it would have broken authentication for the service. The engineer shared they wouldn't have caught it on their own.
Early access customers have seen similar patterns. On a ZFS encryption refactor in TrueNAS's open-source middleware, Code Review surfaced a pre-existing bug in adjacent code: a type mismatch that was silently wiping the encryption key cache on every sync.
Cost and Control
Code Review is more expensive than the existing Claude Code GitHub Action (which remains open source). Reviews are billed on token usage and generally average $15–25, scaling with PR size and complexity. Admins have several control options:
- Monthly organization caps for total spend
- Repository-level control to enable reviews only on selected repositories
- Analytics dashboard to track PRs reviewed, acceptance rate, and total review costs
Getting Started
For admins: Enable Code Review in Claude Code settings, install the GitHub App, and select repositories. For developers: Once enabled, reviews run automatically on new PRs with no configuration needed.
📖 Read the full source: HN AI Agents
👀 See Also
OpenClaw Skill Reduces Accessibility Tree Tokens from 600K to 1.3K
A developer built an OpenClaw skill that uses ML-based element ranking to prune accessibility trees, cutting slickdeals.com from ~598K tokens to ~1.3K tokens by keeping only the top ~50 actionable elements.
Building a Self-Updating Writing Style Guide for AI-Assisted Content
A team building a voice extraction platform called Noren has developed a 117-line Markdown style guide that rewrites itself after every published piece, using Claude to enforce rules and banning AI-sounding words like 'cadence' and 'optimize'.
Local Trello-Style Project Manager for OpenClaw Agents
A developer built a local Trello-like project management tool that runs on the same machine as their OpenClaw agent, storing cards as markdown files with YAML frontmatter. The system uses Node.js/Express for the API, React for the UI, and allows the AI agent to read/write files directly on the filesystem.
120 Prompt Patterns Tested: 8 That Actually Work for Claude Code
A 3-month empirical test of 120 prompt patterns for Claude Code yields 8 actionable commands and 5 validation prompts. Key patterns: L99 (cuts hedging), /ghost (removes AI voice), OODA (structured reasoning), ULTRATHINK (deep reasoning), HARDMODE (constraint debugging).