haddock-development/claude-reflect-system ? reverse-engineered prompt

Build me a Python based Claude Code skill system that can learn from my corrections and remember them across sessions.

I want it to watch for things like when I say use X instead of Y, always do this, or that worked well, then turn those into permanent skill updates that Claude Code can reuse next time. It should feel safe, with backups, review before applying changes, and a way to undo anything bad. I also want a simple command flow like checking status, running reflection manually, and optionally turning on automatic reflection at the end of a session.

Please include a small demo skill too, something like a Python project creator that can learn preferred tools and style choices. Keep it local, easy to copy into the Claude skills folder, and make the learning transparent so I can see what changed. If you need current Claude Code docs or hook details, look them up online.