Smart Contract Security Audit System 🔒

An automated smart contract security audit system powered by AI agents (OpenAI GPT-4 and Anthropic Claude) that collaborate to analyze smart contracts for vulnerabilities.

Features ✨

• Multi-agent collaboration system
• Support for both OpenAI and Anthropic Claude models
• Automated vulnerability analysis
• Business logic review
• Exploit verification
• Detailed audit reports
• Easy switching between AI providers

Installation 🛠️

1. Clone the repository:

git clone https://github.com/Teoslaf/Multi-Agent-Audit-Swarm.git swarm
cd swarm

2. Create and activate a virtual environment:

python -m venv venv

# On Windows:
.\venv\Scripts\activate

# On Unix/MacOS:
source venv/bin/activate

3. Install dependencies:

pip install -r requirements.txt

4. Create a .env file in the project root:

OPENAI_API_KEY=your-openai-key-here
ANTHROPIC_API_KEY=your-anthropic-key-here
OPENAI_MODEL=gpt-4-0125-preview
CLAUDE_MODEL=claude-3-sonnet-20240229

Docker Usage 🐳

1. Build the image:

docker build -t smart-contract-audit .

2. Run the audit:

# Using OpenAI
docker run --env-file .env smart-contract-audit --provider openai --contract test.sol

# Using Claude
docker run --env-file .env smart-contract-audit --provider anthropic --contract test.sol

Project Structure 📁

smart-contract-audit/
├── venv/
├── triage_agent/
│   ├── __init__.py
│   ├── agents.py      # Agent definitions
│   ├── prompts.py     # Agent instructions
│   ├── tools.py       # Analysis tools
│   └── run.py         # Main execution script
├── swarm.py           # Agent collaboration system
├── requirements.txt
├── .env
└── README.md

Agents 🤖

• Security Expert: Analyzes smart contracts for security vulnerabilities
• Business Logic Expert: Reviews contract logic and business implications
• White Hat: Verifies potential exploits
• Documentation Specialist: Formats and finalizes audit reports

Environment Variables 🔑

• OPENAI_API_KEY: Your OpenAI API key
• ANTHROPIC_API_KEY: Your Anthropic API key
• OPENAI_MODEL: OpenAI model to use (default: gpt-4-0125-preview)
• CLAUDE_MODEL: Claude model to use (default: claude-3-sonnet-20240229)

Workflow 🔄

1. Initial Analysis

• Security Expert scans for vulnerabilities
• Business Logic Expert reviews contract design
• White Hat verifies potential exploits

2. Collaborative Rounds

For each round (default: 3 rounds):

1. Each agent reviews others' findings
2. Adds new insights and concerns
3. Validates or challenges previous findings
4. Expands on attack vectors

3. Final Report

Documentation Specialist compiles:

• Combined findings from all agents
• Results from collaborative rounds
• Prioritized vulnerabilities
• Recommended fixes

Output Example:

📊 Final Audit Report
==================================================
[Vulnerability Summary]
- High Risk: Re-entrancy in withdraw function
- Medium Risk: Unchecked return values

[Detailed Analysis]
...

[Recommendations]
...
==================================================