I've been starting new repos over and over again, kept a list of personal skills too. But recently I have been thinking of a different approach: a "Minimal AGENTS.md" (or CLAUDE.md if you use claude)

The idea is inspired by "Organic Architecture" theory by architect Frank Lloyd Wright where a building will dynamically grow / change with people using it.

For example here are a few lines I put in the minimal AGENTS.md (conceptual)

---

1. Always keep a README.md in any folder greater than 5 files, and update the README.md of immediate and parent folders when changing.

2. Submit for code review using [{my preference of ai code reviewer venders}] and loop until code resolves. Ok to push back and dismiss. Do TDD if code reviewer found critical bug you agree to

3. Keep the memory of lesson learned in .memory/<topic>.md and update them, consolidate to skill if hit the same lesson more than 3 times.

---

And usually it helps kind of start the growth of the repo as I further develop it.

has anyone try this kind of approach?