ASCIT31/Dark-Moon ? reverse-engineered prompt

Build me a self hosted AI penetration testing tool that I can point at an authorized target and have it run the assessment on its own.

I want it to discover the target, figure out what technologies are in use, pick the right specialist agents, run through the testing flow, and then give me a clear report with proof for each finding, including the command used, the raw output, and what made it vulnerable. It should work for web apps, APIs, Active Directory, Kubernetes, cloud, CI CD, databases, and common local infrastructure.

Please keep it private by default so the model never sees real IPs, hostnames, or credentials, and make sure that mapping happens locally before any tool runs. I also want a simple install and run experience with Docker, a setup script, and a shell command to start an assessment and watch live logs.

If you need to check current docs or tool behavior, look them up online and keep everything self hosted.