Lelolima/ShadowForge-Agent — reverse-engineered prompt

Build me a Python tool called ShadowForge Agent that helps security teams run authorized penetration test workflows from start to finish. I want it to take a target, run safe reconnaissance and scanning, decide what to do next with an OODA style loop, and generate a clear report at the end with findings and recommended fixes.

Please make simulation mode the default so it doesn’t do real attacks by accident. Add strong guardrails, require confirmation before any active testing, block destructive actions, and support allow lists and deny lists for targets. It should be able to use NVIDIA NIM for planning and report writing when an API key is configured, but still have useful fallback behavior.

Include a clean command line interface with modes like recon only, stealth, debug, and safe simulation. Optional voice and screen analysis support would be great if the needed services are available. Add setup scripts, env examples, health checks, tests, and basic docs so I can run it locally. Look up current NVIDIA docs online if you need to.