Eigen

An AI-powered security vulnerability research framework that analyzes git repositories to discover potential security issues through automated commit analysis, code review, and exploit development.

Overview

Eigen uses a multi-agent architecture built on LangGraph to perform deep security analysis of codebases. It orchestrates multiple specialized AI agents that work together to:

1. Analyze Git Commits - Examine commit history to identify security-relevant changes
2. Review Code - Perform in-depth code review using static analysis tools
3. Discover Vulnerabilities - Identify potential security issues and attack surfaces
4. Generate Exploits - Develop proof-of-concept exploits to verify findings

Architecture

┌─────────────────────────────────────────────────────────────────┐
│                        Supervisor Agent                         │
│              (Coordinates analysis workflow)                    │
└─────────────────────────────────────────────────────────────────┘
                               │
         ┌─────────────────────┼─────────────────────┐
         ▼                     ▼                     ▼
┌─────────────────┐  ┌─────────────────┐  ┌─────────────────┐
│ Initial Commit  │  │  Code Reviewer  │  │    Exploit      │
│    Analyzer     │  │     Agent       │  │   Generator     │
└─────────────────┘  └─────────────────┘  └─────────────────┘

Agents

Agent	Description
Supervisor	Orchestrates workflow, delegates tasks, and synthesizes final reports
Initial Commit Analyzer	Analyzes commit messages, diffs, and identifies significant changes
Code Reviewer	Performs deep code analysis using static analysis tools (semgrep, cppcheck, bandit, etc.)
Exploit Generator	Creates proof-of-concept exploits to verify discovered vulnerabilities
Critique Agent	Reviews reports for completeness and security relevance
Breakdown Agent	Decomposes complex exploitation tasks into actionable sub-tasks

MCP Integrations

Eigen connects to multiple Model Context Protocol (MCP) servers:

• Pwno - Binary analysis and exploitation tools
• claude-context - Code understanding and context retrieval (Milvus-backed)
• DeepAgent - Deep code analysis via DeepWiki
• lldb - LLDB debugger integration for dynamic analysis

Installation

# Clone the repository
git clone https://github.com/HACKE-RC/Eigen.git
cd Eigen

# Install dependencies
pip install -r requirements.txt

# Set environment variables
export OPENAI_API_KEY="your-openai-key"
export GOOGLE_API_KEY="your-google-key"  # For Gemini

Usage

python main.py /path/to/target/git/repository

The framework will:

1. Analyze the last 200 commits in the target repository
2. Generate reports in the reports/ directory within the target repo
3. Perform binary/source analysis and save findings

Output

Reports are saved in markdown format:

File	Description
reports/<commit_hash>_initial.md	Initial commit analysis
reports/<commit_hash>_code_review.md	Detailed code review findings
reports/<commit_hash>_final.md	Final synthesized report
initial_analysis.md	Source code security analysis
dynamic_analysis.md	Dynamic analysis and vulnerability findings
exploit_report_<n>.md	Exploit documentation for each finding

Tools

The framework leverages various security tools:

• Static Analysis: semgrep, bandit, cppcheck, clang-tidy, pylint, mypy
• Binary Analysis: pwntools, ROPgadget, angr, capstone, unicorn
• Symbolic Execution: z3-solver
• Debugging: LLDB (via MCP)

Requirements

• Python 3.10+
• Git repository access
• API keys for LLM providers (OpenAI/Google)
• MCP servers running (Pwno, lldb) for full functionality

License

MIT

Disclaimer

This tool is intended for authorized security research and educational purposes only. Always obtain proper authorization before analyzing any codebase.