keep-memory-current

Keep memory current

A memory store is only worth what the next session can find in it. Two moments decide that, and they are the two this skill is about: what you load when you start, and what you write when you stop.

The second one decays first and costs the most. A session that solves something hard and ends without recording it has not saved anyone the hour; it has spent the hour on that project's behalf and thrown away the receipt.

When you start

Read the hub, then search before you act. The hub is the index file the project loads into context (commonly MEMORY.md); it carries current state and pointers, not the content. Reading it tells you what exists. It does not tell you what applies.

⛔ An index line is a pointer, not the memo. Being able to quote the one-line hook is not having read the entry, and compression drops SCOPE first — which is the part that decides whether the memo applies to what you are about to do.

Before acting on anything the project may already have an opinion about, search. The companion skill check-memory-before-acting covers how to phrase that search, and the short version is: search for the hazard, not the task.

When you stop

One fact per memo, with the cost that bought it

A memo that records only the conclusion cannot be retired later, because nobody can tell whether the reason still holds. Write:

• the fact, stated so it can be checked
• why it is true, in enough detail to argue with
• how to apply it, concretely
• what it cost to learn — the hour, the outage, the wrong release

The cost is not decoration. It is what lets a future reader judge how hard to defend the memo when new evidence arrives, and it is what makes a stale memo removable instead of load-bearing.

Link liberally, including to things that do not exist

A [[name]] pointing at an unwritten memo is not a broken link. It marks a gap worth filling, and it is the cheapest form of a to-do the store supports.

Delete what turns out to be wrong