What should a coding agent remember between sessions?

What should a coding agent remember between sessions, and what should it deliberately forget?

I keep seeing the same failure: an agent solves a task in one session, then starts the next session without the decisions that shaped the code. I built Engraphis to make memory scoped by repo and session and inspectable. The local core is Apache-2.0 and runs on SQLite.

The part I am still testing is how to rank a decision that was correct six months ago but is now stale. Finding the old note is easy. Knowing whether it is still valid for this repo is the hard part.

If you have a repo with superseded decisions, I would like to test recall on a sanitized example. What would you want the agent to remember, and what should it forget?