coderabbit-cli

name

coderabbit-cli

description

Use CodeRabbit CLI to perform automated code review and iterative improvement in an AI agent workflow. Use when: (1) After generating non-trivial code (new features, refactors, algorithms), (2) Improving code quality, maintainability, or readability before submission, (3) Validating code changes against best practices, (4) Exploring unfamiliar languages, patterns, or domains, or (5) Creating a self-reviewing coding loop. Requires CodeRabbit CLI installed and authenticated. Not for trivial changes (typos, formatting-only) or rapid prototyping without quality constraints.

CodeRabbit CLI

Use CodeRabbit CLI to perform structured, automated code review and iterative improvement, enabling AI agents to write, review, and refine code in a tight feedback loop.

Prerequisites

• CodeRabbit CLI installed: npm install -g @coderabbitai/cli or see docs.coderabbit.ai
• Authenticated: coderabbit auth login (one-time setup)
• Git repository: Run commands from within a git repository. CodeRabbit reviews: unstaged working-tree changes, staged-but-uncommitted changes, and local commits not yet pushed. Does not run on a clean working tree with no local changes.
• Repository context: CodeRabbit needs access to repository metadata (can be configured via .coderabbit.yaml)

Quick Start

Run review on current changes:

coderabbit review --plain

Get token-efficient summary:

coderabbit review --prompt-only

Limit scope to local changes:

coderabbit review --plain --type uncommitted

Limit scope by base branch or commit:

coderabbit review --plain --base main

AI Agent Review Workflow

1. Implement Code

Write the requested code or changes following project conventions and requirements.

2. Run CodeRabbit Review

Choose the appropriate review mode based on context:

Detailed feedback mode (recommended for active development):

coderabbit review --plain

Token-efficient mode (for tight token budgets):

coderabbit review --prompt-only

Limit scope (when focusing on particular changes):

• Use --type uncommitted to review only local changes
• Use --base or --base-commit to compare against a specific baseline

3. Analyze Feedback

CodeRabbit provides feedback in several categories:

• Correctness issues: Bugs, logic errors, type safety problems
• Readability improvements: Code clarity, naming, structure
• Maintainability suggestions: Best practices, patterns, technical debt
• Security concerns: Vulnerabilities, unsafe patterns
• Performance optimizations: Efficiency improvements

Key principles for analyzing feedback:

• Treat CodeRabbit as a senior reviewer: reason about suggestions, don't blindly apply them
• Prioritize correctness and security issues first
• Consider maintainability and readability improvements
• Evaluate performance suggestions in context of actual requirements
• Some suggestions may be stylistic or context-dependent

4. Revise Code

Apply meaningful improvements based on CodeRabbit's feedback:

• Fix correctness issues immediately
• Address security concerns
• Improve readability where it adds value
• Apply maintainability suggestions that align with project patterns
• Consider performance optimizations if they're relevant

Document rationale for significant changes or when choosing not to apply suggestions.

5. Re-review (Optional)

For significant changes or when addressing critical issues, re-run CodeRabbit to validate improvements:

coderabbit review --plain

This creates an iterative improvement loop until code quality meets standards.

Usage Patterns

After Feature Implementation

When implementing new features:

1. Complete the feature implementation
2. Run coderabbit review --plain for comprehensive feedback
3. Address critical and major issues
4. Re-review if significant changes were made
5. Proceed with submission when quality gates pass

Before PR Submission

When preparing code for human review:

1. Stage all changes: git add .
2. Run coderabbit review --plain to catch issues early
3. Fix all actionable feedback
4. Re-run review to confirm fixes
5. Submit PR with confidence that basic quality checks pass

Exploring Unfamiliar Domains

When working with new languages, frameworks, or patterns:

1. Implement initial solution
2. Run coderabbit review --plain to learn best practices
3. Study feedback to understand domain conventions
4. Revise code applying learned patterns
5. Use as learning tool to understand idiomatic code

Refactoring Existing Code

When improving existing code:

1. Make refactoring changes
2. Run coderabbit review --plain to ensure no regressions
3. Verify feedback aligns with refactoring goals
4. Address any new issues introduced
5. Confirm code quality improved or maintained

Command Reference

Basic Review Commands

Review all uncommitted changes:

coderabbit review

Plain text output (detailed):

coderabbit review --plain

Prompt-only output (token-efficient):

coderabbit review --prompt-only

Review only uncommitted changes:

coderabbit review --type uncommitted

Review changes against a base:

coderabbit review --base main

Review staged changes:

git add .
coderabbit review

CodeRabbit automatically detects and reviews staged changes when they exist. The coderabbit review command will review all uncommitted changes (both staged and unstaged) by default.

Note: the current CodeRabbit CLI does not support a --files option. To limit scope (or avoid file-count limits), rely on --type, --base, or --base-commit, or use git to stage only the changes you want reviewed.

Authentication

Login to CodeRabbit:

coderabbit auth login

Check authentication status:

coderabbit auth status

Configuration

CodeRabbit can be configured via .coderabbit.yaml in the repository root:

language: "en-US"

reviews:
  review_status: false  # Suppress auto-generated status comments
  pre_merge_checks:
    docstrings:
      mode: "off"  # Disable docstring coverage checks

See CodeRabbit Configuration for full options.

Integration with Development Workflow

With Git Workflow

1. Make code changes
2. Stage changes: git add .
3. Run CodeRabbit review
4. Fix issues
5. Commit with confidence: git commit -m "feat: implement feature"
6. Push and create PR

With AI Agent Workflows

1. Agent implements code based on requirements
2. Agent runs coderabbit review --plain
3. Agent analyzes feedback and identifies actionable issues
4. Agent revises code addressing feedback
5. Agent optionally re-runs review to validate fixes
6. Agent documents changes and rationale
7. Agent proceeds with next steps (tests, documentation, etc.)

Quality Bar

When using CodeRabbit in an agent workflow:

• Address correctness issues: All bugs and logic errors must be fixed
• Consider security concerns: Security issues should be addressed or documented
• Evaluate maintainability: Apply suggestions that align with project patterns
• Reason about feedback: Don't blindly apply all suggestions; understand intent
• Document decisions: When choosing not to apply suggestions, note rationale

When Not to Use

• Trivial changes: Typos, formatting-only edits, simple renames
• Rapid prototyping: When speed is more important than quality
• Repository not initialized: CodeRabbit needs git context
• No local changes: Nothing to review if working tree is clean (no unstaged, staged, or uncommitted local changes)

Best Practices

Token Management

• Use --prompt-only when operating under tight token budgets
• Use --plain during active development for detailed feedback
• Focus on actionable feedback rather than reading all suggestions

Feedback Analysis

• Prioritize critical and major issues
• Group similar suggestions for efficient addressing
• Consider context when evaluating stylistic suggestions
• Some suggestions may conflict with project conventions

Iterative Improvement

• Don't try to address all feedback in one pass
• Focus on correctness and security first
• Re-review after significant changes
• Use feedback as learning opportunity

Reference Documentation

• CodeRabbit CLI Docs: docs.coderabbit.ai/cli
• Configuration Reference: docs.coderabbit.ai/configuration
• Review Guidelines: See references/cli-commands.md for complete command reference