Archive / Your agent wasn't in the room: the short note that keeps it in sync

Your agent wasn't in the room: the short note that keeps it in sync

A fill-in note that tells your agent what changed while it wasn't looking, so it doesn't work from an old picture or undo your fixes.

For: Solo founders, Small business owners, New to AI

Topics: Agent management

Why I wrote it

I spent an afternoon cleaning up my agent's task list by hand, with help from a different AI. We closed stale tasks, fixed a fact it had wrong about someone and set new priorities. The agent wasn't part of any of that.

Its next scheduled run would have read its own memory, decided the old tasks were still open and put them back. So before it ran again, I gave it a note: here is what changed, here is who confirmed it, here is what is still open.

What makes it work

It opens by saying when it matters: read this before your next run touches these tables or folders.

Corrections of fact come first and are marked important. A wrong fact spreads into everything else the agent writes.

Every change says what it was, what it is now, and who confirmed it and when. The agent can tell your decisions from its own guesses.

A "what I didn't do" list tells the agent what is still open, so it doesn't assume everything was handled.

The last line asks it to flag conflicts instead of quietly overwriting them. That line has saved me more than once.

The prompt

Prompt

Retrofit: changes made directly to [where: your task tracker, database, notes folder] since your last update, without you. Read this before your next run touches [the tables or folders it uses]. CORRECTION, IMPORTANT: [The fact you had wrong] is wrong. The correct fact is [the right fact]. If you had this wrong anywhere, correct it. WHAT I CHANGED (all with real dates, none guessed): 1. [Item]: was [old state]. Now [new state]. Confirmed by [your name] on [date]. 2. [Item]: was [old state]. Now [new state]. Confirmed by [your name] on [date]. CURRENT PRIORITY LIST, in full: [Your list.] Treat this as the real list going forward, not the older one. WHAT I DIDN'T DO (yours if you know more): - [Open item] - [Open item] Report back if anything above conflicts with what you know to be true. Don't silently overwrite. Flag it.

How to use it

Write it right after you finish editing, while you still remember what you touched. Fill in only the sections you need. A note with one correction and nothing else is fine.

Send it to the agent before its next scheduled run, then read its reply. If it pushes back on something, it may know a fact you don't. Check before you overrule it.

Use it for: Bringing an agent up to date after you edited its task list, files or records by hand or with a different AI.

Back to the archive