CodeRabbit reviews pull requests with AI. It runs automatically on GitHub PRs and can also run locally via CLI before you push.

Install

CLI:

curl -fsSL https://cli.coderabbit.ai/install.sh | sh
coderabbit auth login

Claude Code plugin:

claude plugin install coderabbit

Local review workflow

Run a review inside Claude Code before opening a PR:

/coderabbit:review uncommitted

Or from the terminal:

coderabbit review --plain

Fix issues, re-run, iterate until clean. Then push knowing the automated PR review will be quiet.

Configuration

.coderabbit.yaml at the repo root controls review behavior.

Key settings:

• reviews.profile: chill reduces comment volume, default is more aggressive
• reviews.path_filters: glob patterns to skip (lock files, build artifacts, node_modules)
• reviews.path_instructions: per-path AI instructions that tell CodeRabbit what to focus on or ignore

CodeRabbit also reads CLAUDE.md files automatically and treats them as code guidelines during review.

Handling false positives

Two approaches:

1. Inline on the PR: reply @coderabbitai on the comment explaining why it's wrong. CodeRabbit stores these as "learnings" and applies them to future reviews.

2. path_instructions in .coderabbit.yaml: add permanent instructions for specific paths. More durable since it's checked into the repo.

Example from this repo: blog frontmatter uses comma-separated tags on one line. CodeRabbit kept suggesting YAML list format. Fixed by adding to the blog posts path instruction:

- path: "apps/blog/blog/markdown/posts/**"
  instructions: >
    Frontmatter tags use comma-separated values on a single
    line. Do not suggest converting to YAML list format.

Related Blog Posts

• Claude Code + CodeRabbit: full setup walkthrough and .coderabbit.yaml config
• Safety Hooks for Claude Code: hooks that run before/after tool calls, reviewed by CodeRabbit