wtf is onemind.md?

onemind.md is a one-file specification that gives a git repository a kind of memory, using nothing but git's own internal building blocks. The idea is that an AI coding agent working in a repo can record its decisions, the ideas it tried and rejected, and other context, and store all of that as git objects on a hidden reference called refs/mind/main. This hidden reference does not show up in your working folder, does not create a mind folder you can see, and does not touch git status, so it stays completely separate from your normal commits and branches. The point of storing this information is to answer questions that normally get lost, such as why a certain approach was chosen, what was tried and rejected already so nobody wastes time repeating it, and what the previous work session figured out so the next one can continue from there. Because the memory is stored as git objects, it travels with the repository automatically. Anyone who clones the repo gets the mind along with the code, with no separate database or service required. The project sets some safety rules for how this memory should be used: nothing should ever be written to the visible working tree, secrets or personal information should never be stored in it, entries should only be added and never rewritten, with corrections handled through git notes instead, and there is a soft size limit of 1 gigabyte by default, with old entries meant to be pruned over time. The README also points to a separate document describing an abstract version of the idea that could work with tools other than git. A live example of how it looks in use is shown in an animated GIF in the repository.