DXv-3/self-improving-system-builder — reverse-engineered prompt

Reverse engineered prompt

Build me a small Python tool that acts like a self improving task router for local work. I want to be able to give it a plain task string, have it choose the best matching skill or path with a confidence score, run the plan safely, and keep a record of what happened so it can get better over time. If something looks risky, it should stop unless I explicitly allow it, and if execution fails it should roll back to a safe state.

Please include the IDKWIDK 7 gate audit before anything is marked done, with the final next steps turned into tracked actions. I also want a simple learning loop that saves outcomes, updates reliability for each skill, and can notice successful direct actions that should become reusable skills later.

Keep it lightweight and mostly standard library if possible. Add a solid test setup that covers the main flow, safety behavior, and skill conflicts. If anything is unclear, look up current docs online and make sensible choices.

Want more depth? Deep Reverse