Skip to content

MITHILESHK11/Code_Surgery_Bot

BranchesTags

Repository files navigation

🔬 Code Surgery Bot

Autonomous multi-agent AI system for GitHub Pull Request code reviews
Powered by AMD Developer Cloud + ROCm + CrewAI

License: MIT Python 3.10+ Built for AMD Hackathon

🎯 What It Does

Paste a GitHub PR URL → Get a full expert code review in under 60 seconds.

Code Surgery Bot dispatches 6 specialized AI agents that work in parallel:

  • 🔍 Logic Checker — Finds bugs, null pointers, off-by-one errors, missing error handling
  • 🔒 Security Agent — Detects SQL injection, hardcoded credentials, XSS, insecure dependencies
  • ✨ Style Agent — Enforces naming conventions, checks complexity, finds code duplication
  • 📝 Documentation Agent — Audits docstrings and comments
  • 🤖 Coordinator Agent — Orchestrates all agents and resolves conflicts
  • 📊 Report Generator — Synthesizes findings into a human-readable review with code fixes

Each agent is powered by specialized open-source LLMs running on AMD MI300X GPU via ROCm.

🚀 Live Demo

Try it on Hugging Face Spaces →
(Demo available after deployment)

📊 Architecture

User submits GitHub PR URL
        ↓
PR Analyzer Agent (fetches files, diffs, metadata)
        ↓
Coordinator Agent (decides which agents to run)
        ↓
        ┌──────────────────────────────────┐
        │  Runs in parallel:               │
        ├──────────────────────────────────┤
        │ • Logic Checker Agent            │
        │ • Security Agent                 │
        │ • Style Agent                    │
        │ • Documentation Agent            │
        └──────────────────────────────────┘
        ↓
Report Generator Agent (writes final review)
        ↓
Output: Scored review report with fixes

⚡ Quick Start

Prerequisites

  • Python 3.10+
  • GitHub Personal Access Token
  • AMD Developer Cloud account (for GPU)
  • Or use HuggingFace API as fallback

Installation

# Clone the repository
git clone https://github.com/your-org/code-surgery-bot.git
cd code-surgery-bot

# Create virtual environment
python -m venv venv

# Activate it
# On Windows:
venv\Scripts\activate
# On macOS/Linux:
source venv/bin/activate

# Install dependencies
pip install -r requirements.txt

# Copy environment template
cp .env.example .env

# Edit .env with your credentials
nano .env  # or use your editor

Configuration

Add to .env:

GITHUB_TOKEN=your_github_personal_access_token
AMD_API_KEY=your_amd_developer_cloud_key
MODEL_ENDPOINT=http://your-amd-instance:8000/v1
HF_TOKEN=your_huggingface_token

Run the Demo UI

python app/gradio_app.py

Then open http://localhost:7860 in your browser.

Programmatic Usage

from src.github import PRFetcher
from src.agents import CoordinatorAgent

# Initialize
pr_fetcher = PRFetcher(github_token="your_token")
coordinator = CoordinatorAgent()

# Fetch and review
pr_url = "https://github.com/facebook/react/pull/28000"
pr_data = pr_fetcher.fetch(pr_url)
review_report = coordinator.review(pr_data)

print(review_report)

🏗️ Project Structure

code-surgery-bot/
│
├── src/
│   ├── agents/
│   │   ├── pr_analyzer.py        ← Fetch PR data from GitHub
│   │   ├── coordinator.py        ← Orchestrate agents
│   │   ├── logic_checker.py      ← Find bugs
│   │   ├── security_agent.py     ← Find vulnerabilities
│   │   ├── style_agent.py        ← Check code quality
│   │   ├── report_generator.py   ← Write final review
│   │   └── __init__.py
│   │
│   ├── github/
│   │   ├── pr_fetcher.py         ← GitHub API integration
│   │   └── __init__.py
│   │
│   ├── models/
│   │   ├── llm_client.py         ← LLM inference client
│   │   └── __init__.py
│   │
│   ├── utils/
│   │   ├── ast_parser.py         ← Code parsing utilities
│   │   └── __init__.py
│   │
│   └── __init__.py
│
├── app/
│   ├── gradio_app.py             ← Web UI demo
│   └── fastapi_backend.py        ← REST API backend
│
├── tests/
│   ├── test_agents.py            ← Agent unit tests
│   └── __init__.py
│
├── docs/
│   ├── architecture.md
│   ├── setup-amd-gpu.md
│   └── agent-prompts.md
│
├── examples/
│   ├── sample_review_output.md   ← Example report
│   └── test_pr_urls.txt
│
├── requirements.txt
├── setup.py
├── LICENSE (MIT)
├── .gitignore
├── .env.example
└── README.md (this file)

🔧 Tech Stack

Component Technology Why
GPU Compute AMD Instinct MI300X + ROCm Fast code inference on open-source
Agent Framework CrewAI Best for multi-agent orchestration
LLMs DeepSeek-Coder-33B, Llama-2-70B, Mistral-7B Code-specialized, open-source
Inference vLLM Batched, fast inference on ROCm
GitHub Integration PyGithub Fetch PR data, post reviews
Code Parsing tree-sitter, Python AST Language-aware analysis
Backend API FastAPI REST endpoint for orchestration
Frontend Demo Gradio Clean UI for live demo
Deployment Hugging Face Spaces Free, public, scalable

🎓 How Each Agent Works

1️⃣ PR Analyzer Agent

Input: GitHub PR URL
Process: Fetches changed files, unified diffs, commit messages using PyGithub API
Output: Structured JSON with { files, diffs, metadata, ast_nodes }

2️⃣ Coordinator Agent

Input: Structured PR data from analyzer
Process: Decides which agents are relevant, dispatches tasks in parallel
Output: Task assignments and conflict resolution

3️⃣ Logic Checker Agent

Model: DeepSeek-Coder-33B
Detects:

  • Null pointer / undefined variable risks
  • Off-by-one errors in loops
  • Missing error handling (try/catch gaps)
  • Wrong return types
  • Unreachable code blocks
  • Infinite loop risks
  • Race conditions in async code
  • Edge case violations

4️⃣ Security Agent

Model: Llama-2-70B-Chat
Detects:

  • SQL injection vulnerabilities
  • Hardcoded credentials & API keys
  • XSS vulnerabilities
  • Insecure dependencies (CVE cross-check)
  • SSRF & path traversal risks
  • Authentication/authorization gaps
  • Sensitive data in logs
  • Deprecated crypto functions

5️⃣ Style & Quality Agent

Model: Mistral-7B-Instruct
Checks:

  • Naming convention consistency
  • Function complexity (>50 lines → split)
  • Code duplication (DRY violations)
  • Magic numbers
  • Complex conditionals
  • Import organization
  • Commented-out code
  • Type hints coverage

6️⃣ Report Generator Agent

Model: Llama-2-70B-Chat
Process:

  • Combines all agent outputs
  • Calculates quality score (0-100)
  • Groups issues by severity
  • Generates corrected code snippets
  • Renders GitHub-compatible Markdown
  • Produces verdict: ✅ Approve / ⚠️ Request Changes / ❌ Reject

📈 Output Example

## 🔬 Code Surgery Bot Review — PR #47

**Overall Score: 72/100** ⚠️ Request Changes

### Issue Summary
| Severity | Count |
|----------|-------|
| Critical | 0     |
| High     | 2     |
| Medium   | 5     |
| Low      | 3     |

### 🔒 Security Issues (2 High)
**auth.py line 34**: API key hardcoded in source

❌ BEFORE: API_KEY = "sk_live_abc123xyz" users_db = connect(API_KEY)

✅ AFTER: API_KEY = os.getenv("API_KEY") if not API_KEY: raise EnvironmentError("API_KEY not set") users_db = connect(API_KEY)


**db.py line 89**: Raw SQL with f-string interpolation

❌ BEFORE: query = f"SELECT * FROM users WHERE id={user_id}" cursor.execute(query)

✅ AFTER: query = "SELECT * FROM users WHERE id = %s" cursor.execute(query, (user_id,))


### 🔍 Logic Issues (3 High)
[...]

### ✅ Verdict: REQUEST CHANGES
Fix 2 critical security issues before merge.

🚀 Deployment

Deploy to Hugging Face Spaces

# Create Space on huggingface.co
# Choose "Gradio" SDK

# Clone Space
git clone https://huggingface.co/spaces/your-username/code-surgery-bot
cd code-surgery-bot

# Copy app
cp app/gradio_app.py ./app.py
cp requirements.txt .

# Push
git add . && git commit -m "Deploy Code Surgery Bot" && git push

Visit: https://huggingface.co/spaces/your-username/code-surgery-bot

Deploy Backend to AMD Developer Cloud

# SSH into MI300X instance
ssh user@amd-instance-ip

# Install dependencies
pip install -r requirements.txt

# Start inference server
python -m vllm.entrypoints.openai.api_server \
  --model deepseek-ai/deepseek-coder-33b-instruct \
  --port 8000 \
  --tensor-parallel-size 4

# In another terminal, start FastAPI backend
python app/fastapi_backend.py

# Gradio UI connects to http://amd-instance-ip:8000

🧪 Testing

# Run all tests
pytest tests/

# Run specific test
pytest tests/test_agents.py::test_logic_checker

# With coverage
pytest --cov=src tests/

🤝 Contributing

We welcome contributions!

How to Contribute

  1. Fork this repo
  2. Create a feature branch: git checkout -b feature/my-feature
  3. Commit changes: git commit -m "Add my feature"
  4. Push: git push origin feature/my-feature
  5. Open a Pull Request

Areas We Need Help

  • Add support for more languages (Go, Rust, Java)
  • Improve security detection rules (OWASP Top 10)
  • Add comprehensive test suite
  • Performance optimization for large PRs
  • Better diff visualization in UI

📚 Documentation

📜 License

MIT License — Free to use, modify, and distribute.
See LICENSE for details.

🏆 Built For

AMD Developer Hackathon 2025 — AI Agents & Agentic Workflows Track
lablab.ai/ai-hackathons/amd-developer

Prize Goal: $2,500 (Track) + $5,000 (Grand Prize)
Deadline: May 11, 2025 — 12:30 AM IST

🙋 Questions?

🌟 Star History

Help us grow! Give this repo a ⭐ if you find it useful.

Built with ❤️ for the AI community

About

No description, website, or topics provided.

Resources

Readme

License

MIT license

Contributing

Contributing
Activity

Stars

0 stars

Watchers

0 watching

Forks

0 forks
Report repository

Releases

No releases published

Packages

 
 
 

Contributors

Languages