I have started a convention of prefixing a commit message with "prompt:" for commits that are entirely AI generated (which I have read through and approved) and then I make my own manual changes and iterations in subsequent commits.

I feel like this is a reasonable attempt at disclosure, but it also seems like it's lacking. I usually include a summary, not the entire prompt (how would I included tool use?) I don't say which models or "agent" was used (cursor sidebar chat vs calude code, etc). Not sure if this matters though?

How have you been managing this in your own workflows?