AgentWorm

Warning

⚠️ This project is intended solely for ethical, defensive, and research-oriented experimentation in controlled environments. 🚫 Using it for unauthorized access, malicious activity, or any harmful purpose is strictly unethical and strongly discouraged. 🛑 Please refrain from using this repository outside of legitimate research and authorized testing contexts.

Note

🔬 This repository is still under active research and development. 🧪 The code, architecture, and documentation are evolving as experiments and sprint work progress.

AgentWorm is a research project focused on building and studying an agentic command-and-control system for autonomous network propagation in a contained lab environment. The repository explores how a lightweight local agent and an LLM-driven C2 brain can work together to simulate reconnaissance, lateral movement, credential discovery, and multi-host coordination in a controlled and observable setting. This project is still a research effort and is intended for academic, defensive, and experimentation purposes only.

Documentation

The repository includes several design and planning documents that explain the system in more detail:

• AGENTIC_C2_DESIGN.md — describes the proposed agentic C2 architecture and the shift from a prompt-dispatch design toward a tool-driven agent loop.
• PLAN.md — captures the overall implementation plan, milestones, and project direction.
• EXPLOIT_CHAIN.md — outlines the exploit-chain research design and the multi-hop propagation scenarios being studied.
• DO_EXPLOIT_CHAIN.md — documents the DigitalOcean-based testbed design and its final topology.
• CLAUDE.md — contains knowledge for agentic development (Claude Code, Copilot, etc.)

License

This project is licensed under the MIT License. See LICENSE for details.