Coderabbit Local Dev Loop

CodeRabbit Local Dev Loop

Overview

Use CodeRabbit CLI to review code locally before opening a PR. The CLI provides the same AI-powered review as the GitHub App but runs in your terminal against staged or unstaged changes. This creates a multi-layered review process: local CLI review before commit, then automated PR review after push.

Prerequisites

• CodeRabbit CLI installed (curl -fsSL https://cli.coderabbit.ai/install.sh | sh)
• Git repository with .coderabbit.yaml configuration
• CodeRabbit account (CLI uses credits: $0.25 per file reviewed)

Instructions

Step 1: Install and Verify CLI

set -euo pipefail
# Install CodeRabbit CLI
curl -fsSL https://cli.coderabbit.ai/install.sh | sh

# Verify installation
cr --version

# Authenticate (opens browser for OAuth)
cr auth login

Step 2: Local Review Workflow

set -euo pipefail
# Review all staged changes (most common workflow)
git add -A
cr review

# Review specific files only
cr review src/api/routes.ts src/middleware/auth.ts

# Interactive mode: ask follow-up questions about review feedback
cr review --interactive

# Plain output mode (pipe to other tools or AI agents)
cr review --prompt-only

Step 3: Git Hook Integration

#!/bin/bash
# .git/hooks/pre-push (make executable: chmod +x .git/hooks/pre-push)
set -euo pipefail

echo "Running CodeRabbit pre-push review..."

# Get list of changed files vs remote
CHANGED_FILES=$(git diff --name-only @{push}.. 2>/dev/null || git diff --name-only HEAD~1)

if [ -n "$CHANGED_FILES" ]; then
  echo "$CHANGED_FILES" | xargs cr review

  # Non-blocking: show review but don't prevent push
  # To make blocking, check exit code:
  # echo "$CHANGED_FILES" | xargs cr review || {
  #   echo "CodeRabbit found issues. Push anyway? (y/n)"
  #   read -r response
  #   [ "$response" != "y" ] && exit 1
  # }
fi

Step 4: Configuration for Local Development

# .coderabbit.yaml - Settings that affect both CLI and PR reviews
language: "en-US"
reviews:
  profile: "assertive"
  path_instructions:
    - path: "src/**"
      instructions: "Check for proper error handling and type safety."
    - path: "tests/**"
      instructions: "Verify edge cases and assertion completeness."
  path_filters:
    - "!**/*.lock"
    - "!dist/**"
    - "!**/*.generated.*"
  auto_review:
    enabled: true
    drafts: false
chat:
  auto_reply: true

Step 5: IDE Integration Pattern

// .vscode/tasks.json - Run CodeRabbit review from VS Code
{
  "version": "2.0.0",
  "tasks": [
    {
      "label": "CodeRabbit: Review Current File",
      "type": "shell",
      "command": "cr review ${file}",
      "presentation": { "reveal": "always", "panel": "shared" },
      "problemMatcher": []
    },
    {
      "label": "CodeRabbit: Review Staged Changes",
      "type": "shell",
      "command": "cr review",
      "presentation": { "reveal": "always", "panel": "shared" },
      "problemMatcher": []
    }
  ]
}

Two-Layer Review Strategy

Developer writes code
       │
       ▼
┌──────────────────┐
│ cr review (local) │  ← Layer 1: Fast feedback before commit
│ Fix obvious issues│
└────────┬─────────┘
         │
         ▼
   git commit + push
         │
         ▼
┌──────────────────┐
│ CodeRabbit App   │  ← Layer 2: Full context review on PR
│ (automated PR    │
│  review)         │
└──────────────────┘

Output

• CodeRabbit CLI installed and authenticated
• Pre-push git hook for automated local reviews
• VS Code task integration for on-demand reviews
• Two-layer review workflow (local + PR)

Error Handling

Examples

Run the CLI against one non-sensitive changed file in a local feature branch, review the output before committing, and use the pull-request app as the separate authoritative review layer. If the CLI lacks credits, authentication, or context, skip the local automation and rely on the configured PR workflow; do not paste repository secrets or large private files into an alternate tool.

Resources

• CodeRabbit CLI Documentation
• CLI Blog Announcement
• VS Code IDE Extension

Next Steps

See coderabbit-sdk-patterns for PR interaction automation patterns.