Code Review

Code Review is an official Anthropic plugin for Claude Code from the curated anthropics/claude-plugins-official marketplace. Per plugin.json it is published by Anthropic; the README and LICENSE name Boris Cherny (boris@anthropic.com) as author, version 1.0.0, license Apache-2.0.

The plugin provides the /code-review command. On a pull request it launches four parallel agents: two check compliance with CLAUDE.md guidelines, one specifically scans the changed lines for bugs, and one analyses git history and git blame for context.

Every finding is scored for confidence from 0 to 100; only findings scoring 80 or above are posted as a review comment on the pull request, linked to the exact code line via commit SHA. Closed, draft, trivial, or already-reviewed pull requests are skipped automatically.

Requirements per the README: a git repository with GitHub integration, and the GitHub CLI (gh) installed and authenticated. CLAUDE.md files are optional but improve guideline-compliance checking.

Installation: the official Anthropic marketplace is added automatically the first time Claude Code starts interactively; to add it manually:

/plugin marketplace add anthropics/claude-plugins-official

Then install the plugin with:

/plugin install code-review@claude-plugins-official

How four-agent review differs from a single reviewer

Instead of a single AI going through a pull request top to bottom, Code Review runs four specialized agents in parallel, each taking its own angle: two check exclusively against the CLAUDE.md guidelines stored in the repository, one focuses specifically on possible bugs in the actually changed lines, and one draws on git history and git blame to understand why a piece of code was originally written that way. Per the provider, this multi-perspective approach is meant to surface both surface-level style violations and deeper structural problems that a single pass tends to miss.

Why the confidence threshold matters

Automated code reviews often have the problem of flooding a pull request with too many comments, most of which turn out to be irrelevant. Code Review addresses this with a confidence score from 0 to 100 per finding: only findings that reach at least 80 are actually posted as a comment on the pull request. This significantly reduces noise for the development team, but it also means the plugin deliberately suppresses some less certain findings instead of surfacing them for discussion.

Who the plugin fits

Teams that already work with pull-request workflows on GitHub and want an initial automated pass before human review benefit the most. It's particularly useful for repositories with a maintained CLAUDE.md file, since guideline checking can then evaluate concrete, project-specific rules instead of generic best practices. For projects without GitHub integration or without the GitHub CLI installed, the plugin isn't usable.

In short, Code Review is meant as a first automated pass, not a replacement for human judgment on architectural or design-level decisions.

/plugin install code-review@claude-plugins-official