🚀 PR Review Agent

AI-powered pull request reviewer built with Gemini
that analyzes code against issue context and company standards.

🔥 Features

• AI-powered PR analysis
• Bug detection and suggestions
• Code quality scoring
• Web UI plus CLI support
• Company guideline enforcement
• GitHub PR fetch and auto-review
• Optional PR comment posting back to GitHub
• Saved review history per user

💡 Why this project?

Code reviews are time-consuming and inconsistent.
This project uses AI to standardize and accelerate PR reviews
while ensuring alignment with company guidelines.

📸 Demo

🧠 What This Project Does

The agent evaluates PRs by:

• ⭐ Rating the PR out of 10
• 📝 Summarizing changes
• 🐞 Detecting bugs
• ⚠️ Checking edge cases
• 🔧 Suggesting fixes
• 🔒 Flagging security issues
• 🧭 Comparing the PR against issue scope
• 🏢 Enforcing company review standards

🧩 Core Capabilities

• Review using issue URLs, issue text, PR URLs, PR descriptions, and code blocks
• Compare previous code and current code side by side
• Fetch live PR title, body, and file patches from GitHub
• Post generated feedback back to GitHub as a PR comment
• Support email login, Google login, and per-user review history

🗂️ Project Structure

pr-review-agent/
├── agent.yaml
├── SOUL.md
├── RULES.md
├── README.md
├── assets/
│   └── ui.png
├── skills/
│   └── pr-review/
│       └── SKILL.md
├── public/
├── data/
│   └── store.json
├── gemini.js
├── server.js
├── run-agent.js
├── .env
└── package.json

⚙️ How It Works

1. Input

You can provide:

• issue URL
• issue text
• PR URL
• PR title or PR description
• previous code and current code
• optional company rules

2. GitHub Enrichment

If a GitHub PR URL is provided, the app can:

• fetch PR title and body
• fetch changed files and patches
• use real GitHub data for review generation

3. AI Review

The backend sends the combined context to Gemini, which returns a structured review.

4. Output

The UI displays the review in a scannable format and can optionally post it back to GitHub.

🧪 Example Output

## Rating
8/10

## Summary
Fixes multiplication bug

## Issues
- Incorrect operator used earlier

## Corrections
- Replace + with *

## Verdict
APPROVE

🛠️ Requirements

• Node.js 18 or newer
• npm
• git
• a valid Gemini API key

📦 Installation

npm install

🔐 Environment Setup

Create or update .env like this:

GEMINI_API_KEY=your_gemini_api_key
GOOGLE_API_KEY=your_gemini_api_key
GEMINI_MODEL=gemini-2.5-flash
SESSION_SECRET=change_me
AUTH0_DOMAIN=your_auth0_domain
AUTH0_CLIENT_ID=your_auth0_client_id
AUTH0_CLIENT_SECRET=your_auth0_client_secret
AUTH0_CALLBACK_URL=http://localhost:3000/auth/auth0/callback
GITHUB_TOKEN=your_github_token

What each value is used for

• GEMINI_API_KEY: Gemini review generation
• GOOGLE_API_KEY: gitclaw compatibility
• SESSION_SECRET: signed login sessions
• AUTH0_*: Google login via Auth0
• GITHUB_TOKEN: GitHub PR fetch and comment posting

▶️ Run The Project

Web App

npm start

Open:

http://localhost:3000

CLI Mode

npm run cli

Sample CLI Review

node run-agent.js --sample

🌐 GitHub Integration

The app can:

• fetch PRs using the GitHub API
• auto-review fetched PRs
• post generated comments back to GitHub

To use it:

1. Paste a GitHub PR URL
2. Click Fetch PR
3. Keep Auto-fetch PR details from GitHub before review enabled
4. Optionally enable posting review comments back to GitHub
5. Run the review

👤 Authentication

The app supports:

• email sign-up and login
• Google login through Auth0
• per-user saved review history

📁 Main Files

• agent.yaml defines the agent metadata
• SOUL.md defines the agent identity
• RULES.md defines review rules and constraints
• gemini.js builds and sends the review prompt
• server.js serves the app and APIs
• public/ contains the UI
• data/store.json stores users and reviews
• run-agent.js provides CLI support

🚀 Future Improvements

• inline GitHub review comments on changed lines
• richer scoring breakdown
• downloadable markdown and JSON reports
• team dashboards for shared review history
• better diff visualizations inside the UI