🤖 GitHub PR Review Bot

Automated PR reviewer that analyzes diffs and posts inline suggestions using OpenAI. Get AI-powered code reviews on every pull request!

📋 Table of Contents

• ✨ Features
• 🚀 Quick Setup
• 📋 Prerequisites
• 🔧 Configuration
• ⚙️ How it Works
• 🛠️ Error Handling
• 📝 Prompt Template
• 📚 Notes and Limitations
• 📄 License

✨ Features

• 🤖 AI-Powered Reviews: Uses OpenAI to analyze code quality, security, and performance
• 📝 Inline Comments: Posts specific suggestions directly on code lines
• 🔄 Auto-Triggered: Runs on PR open, sync, and ready-for-review events
• 🛡️ Smart Filtering: Avoids duplicate and trivial comments
• 🌍 Language Agnostic: Works with any programming language
• ⚡ Fast & Efficient: Chunks large diffs for optimal processing

🚀 Quick Setup (Choose One)

Option 1: One-Click Setup (Recommended)

npx [email protected] setup

Then add your OpenAI API key as a repository secret named OPENAI_API_KEY.

Option 2: Manual Setup

1. Create .github/workflows/pr-review.yml in your repository:

name: PR AI Review

on:
  pull_request:
    types: [opened, synchronize, ready_for_review]

jobs:
  review:
    runs-on: ubuntu-latest
    permissions:
      contents: read
      pull-requests: write
    steps:
      - name: Checkout
        uses: actions/checkout@v4

      - name: Run PR AI Review Action
        uses: ashugupta22/[email protected]
        with:
          openai_api_key: ${{ secrets.OPENAI_API_KEY }}
          openai_model: gpt-4o-mini
          diff_chunk_size: "12000"
          review_prompt: ''

2. Add your OpenAI API key as a repository secret named OPENAI_API_KEY

📋 Prerequisites

• GitHub repository with Actions enabled
• OpenAI API key (get one from OpenAI Platform)

The Action runs on PR events (opened, synchronize, ready_for_review) and posts a single batched review with inline comments.

Use as a reusable Action in multiple repos

Quick Setup (Recommended)

npx [email protected] setup

Manual Setup

1. In each target repository, create a workflow like:

name: PR AI Review
on:
  pull_request:
    types: [opened, synchronize, ready_for_review]
jobs:
  review:
    runs-on: ubuntu-latest
    permissions:
      contents: read
      pull-requests: write
    steps:
      - name: Run PR AI Review Action
        uses: ashugupta22/[email protected]
        with:
          openai_api_key: ${{ secrets.OPENAI_API_KEY }}
          openai_model: gpt-4o-mini
          diff_chunk_size: "12000"
          review_prompt: ''

2. Add OPENAI_API_KEY secret in each target repository.

Configuration

• OPENAI_API_KEY: OpenAI key (required)
• OPENAI_MODEL: Model name (default: gpt-4o-mini)
• DIFF_CHUNK_SIZE: Max characters per diff chunk sent to OpenAI (default: 12000)
• REVIEW_PROMPT: Override the default review instruction

How it Works

• Lists PR files and unified patches via GitHub API
• Builds a map from file+line to diff position for inline comments
• Concatenates and chunks patches, sends to OpenAI with a strict JSON-only instruction
• Parses { issues: [{ file, line, comment, suggestion }] }
• Filters trivial feedback and duplicates from existing PR review comments
• Posts batched review comments using the Review API

Error Handling

• Workflow fails if OpenAI API key is invalid or expired
• Workflow fails if all OpenAI API calls fail
• Workflow fails if GitHub API calls fail
• Individual chunk failures are logged but don't fail the entire workflow unless all chunks fail

Prompt Template

See prompt.md for the base prompt used when REVIEW_PROMPT is not provided.

Notes and Limitations

• Only comments on lines present in the new revision (no deleted-line comments)
• Binary/large files with no patch are skipped
• Duplicate detection uses path+line+body; adjust in reviewEngine.js if needed

License

MIT