What do you mean? 👀

Like, many code is written by AI now days So it’s better to commit your prompt and code change at same time for future reference and bug fix and probably, to research how to write better prompt

Actually not only prompt, also some llm meta data

I let the LLM generate docs under a /docs folder and let it update and commit along with the code. It works quite well.

Yes, I think using specific files for this purpose is becoming popular. For example, there is the '.rules' file and specific files for each platform, such as '.cursorrules' and '.roorules'. The problem is that every platform seems to have its own conventions for naming the rules file. It would be great to have one that could be used by different platforms. I've also noticed more people using 'ARCHITECTURE.MD' for agent guidance and, more recently, 'CLAUDE.md'.