AutoPwn

🎯 The Problem It Solves

The hardest part of a pentest isn't running individual exploits — it's chaining them: a foothold here, a pivots there, until you reach the objective. AutoPwn wraps that chaining logic in an LLM agent that reads scan output, picks exploits, runs them, and pivots — claiming to be "fully autonomous." In practice it's an LLM copilot with more ambition than reliability: it'll genuinely try to chain attacks, but it hallucinates, fizzles, and needs a human holding the kill switch.

🔧 How It Works

You give AutoPwn a target and an engagement type (full / web / network). It runs an initial scan (Nmap + optional vulnerability scan), feeds the output to the LLM, which proposes an exploit to try, executes it via Metasploit or direct shell, and iterates. The agent is built on top of a framework like LangChain/crewAI, with tool wrappers around common offensive tools. It is explicitly designed to chain across steps — not just run one-shot exploits. The critical caveat: the LLM is doing both the planning and the tool-selection, so it is only as good as the LLM's reasoning about exploit feasibility.

🚀 Installation & Quick Start

git clone https://github.com/AutoPwnProject/autopwn.git
cd autopwn
pip install -r requirements.txt
autopwn init  # creates config.yaml
echo "openai_api_key: sk-..." >> config.yaml