GreyDGL/PentestGPT ? reverse-engineered prompt

Build me an AI agent that can run automated penetration tests for me. I want a command-line tool where I can just give it a target IP address and it will autonomously start probing for vulnerabilities and trying to exploit them. It should use a large language model to reason about its next steps, kind of like an expert hacker thinking out loud.

I'd like to be able to watch its progress in real time, with a live feed of the actions it's taking and its thought process. The whole thing should be packaged in Docker so the setup is really simple—maybe just a make command to install and run. For the AI part, let's have it connect to an external API like Claude, but also include an option to use local models running through something like Ollama so I have flexibility. It should be good enough to solve Capture the Flag (CTF) style challenges across different categories without me having to guide it.