Kulaxyz/self-learning-skills ? reverse-engineered prompt

Build me a self improving skill for AI coding agents that notices when we’ve finally figured out a hard problem in a session and saves that win for next time.

I want it to work with Claude Code, Cursor, and any agent that reads standing instructions like AGENTS.md. When the agent hits a reusable golden path, like a tricky deploy command, where credentials live, how to reach a prod database, or how to verify something live, it should automatically turn that into a reusable rule or skill instead of letting the knowledge disappear after the chat ends.

Please make the experience feel simple and safe. It should only promote something when the path was actually verified, it should remember what dead ends were tried, and it should never write secret values into shared files, only where to find them. Also make it easy to install in a few different ways, and include the files and instructions needed so the right thing gets auto loaded in each tool next time.