Language: 日本語English

code-review Plugin

Author
Anthropic
Category
Productivity
Topics
Code Quality, Review & Testing · Claude Code Customization & Workflow
First seen
2026-07-09
Last confirmed
2026-07-13
Explanation last updated
2026-07-12

The explanation below is AI-generated. Please verify it against the sources.

code-review is a Claude Code plugin that automates pull request review by launching four specialized agents in parallel and posting only high-confidence findings. Two agents check CLAUDE.md compliance, one scans for obvious bugs introduced in the change, and one analyzes git blame/history for context. Each issue is independently scored 0-100 for confidence, and only issues scoring 80 or above are posted as review comments, filtering out false positives like pre-existing issues, pedantic nitpicks, and lint-catchable problems. It automatically skips closed, draft, trivial, or already-reviewed PRs, and posted issues link directly to code using full commit SHAs and line ranges. The plugin is invoked via the /code-review command and requires a Git repository with GitHub integration and the authenticated GitHub CLI (gh).

Overview

code-review is a plugin (not a hosted SaaS) included in the Claude Code repository that adds a /code-review command for automated pull request review. It relies on the GitHub CLI (gh) for viewing PR details, diffs, git blame/history, and posting review comments, and optionally reads repository CLAUDE.md guideline files to check compliance.

What you can do with code-review

  • Run /code-review on a PR branch to launch 4 parallel review agents (2x CLAUDE.md compliance, 1x bug detector, 1x history analyzer)
  • Automatically skip review for closed, draft, trivial, or already-reviewed PRs
  • Get each detected issue scored 0-100 for confidence, with only issues ≥80 posted as a comment
  • Receive review comments with direct links to code using full SHA and line ranges (#L[start]-L[end])
  • Adjust the confidence threshold (default 80) by editing commands/code-review.md
  • Customize review focus by adding agent tasks (e.g., security, performance, accessibility, documentation checks) in commands/code-review.md
  • Improve compliance-check accuracy by maintaining specific CLAUDE.md guideline files in the repository

Sources

Original description (English)

Automated code review for pull requests using multiple specialized agents with confidence-based scoring to filter false positives

History of code-review

Back to list