Code Review Assistant

Why it's ranked here

A strong choice for teams willing to encode their standards and ownership knowledge. It connects analysis to useful GitHub actions while keeping those actions behind confirmation. The verdict is tempered by several service credentials, incomplete handling of very large pull requests, and brittle reviewer parsing.

What's good

Generated files, lockfiles, build output, and coverage artifacts are excluded to reduce token use. Teams can customize expertise, standards, and output structure. The CLI accepts several common pull request identifier formats, supports unattended CI output, and separates recommendations from assignment and comment posting.

Tradeoffs

Basic operation requires Anthropic and GitHub credentials. Diagram generation adds Gemini and Cloudinary accounts. Pull request file retrieval requests only 100 files without pagination, so larger changes may be incomplete. Reviewer parsing only accepts word characters, which can lose hyphenated GitHub handles.

How to use it well

Use it on teams with documented ownership areas and recurring coding rules. Run it interactively for reviewer preparation or non-interactively as an advisory CI step. Keep human approval and automated tests alongside it. It is not a reliable exhaustive check for pull requests exceeding 100 changed files.

Observed

License

Apache License 2.0

Primary language

TypeScript

Packaging

npm package using ECMAScript modules, with an agent executable

Install surfaces

npm install for the CLI, or copy the bundled skill into a Claude Code project

Interfaces

Interactive CLI, non-interactive CLI for CI, Claude Code skill, and Agent SDK usage

External integrations

Anthropic Agent SDK, GitHub Octokit and CLI, optional Gemini diagram generation, and optional Cloudinary uploads

Tests

No test directory or test script appears in the supplied repository tree