A pack, not a service · Instant download

The reason your AI keeps forgetting isn't the AI.

It's that nothing is writing anything down. Every session starts from zero, you re-explain the same context, and decisions you already made get quietly reversed. This is the fix, and it's a handful of plain text files plus two habits.

One time

$49

Nine files, including three diagrams. Downloaded the second you pay, and every future version included at the same link.

Buy now, $49

Instant download, so all sales are final. Read the real sample further down before you decide, it's a full page of the actual pack, not a teaser.

You've had this conversation before. Twice.

You explain the project. You get somewhere good. The window closes, or the model resets, or you start a new chat because the old one got slow. And the next session confidently rebuilds something you already decided against, in the same tone it used when it was right.

The usual advice is to write better prompts. That isn't it. A better prompt still starts from nothing, and you're the one paying the cost of remembering, every single time.

What actually works is boring and it's a filing system. One file that says what's true right now. One that says how the work runs. A dated log. And two habits at the start and end of a session that keep all three honest. That's the whole idea, and once it's running the difference is not subtle.

What's in it

Nine files, and the system they describe is already running a real business.

The file layout: what each file is for, what belongs in it, and what must never go in it
Templates you can paste in today, for the current-state file, the operating file, and the log
The two session habits, one to open and one to close, written as prompts you paste in
The staleness rule that stops two files from quietly disagreeing about what's true
How to write a decision so a future session can't undo it by accident
The failure modes: how a memory system rots, and the check that catches each one
What to keep out, because a system that records everything gets read by nobody
How this scales from one project to a whole business without becoming a second job

Plain markdown. It works with Claude, ChatGPT, Cursor, or anything that can read a file, and it keeps working when you switch tools, which is the actual point.

A real piece of the system

Two words that do most of the work.

The open and close habits, exactly as they run today. Everything else in the pack exists to make these two do something.

pick up
  Read the current-state file FIRST. Then the last two entries of the log, then
  the most recently changed decisions and tasks files. If the state file is older
  than the newest log entry, the log is the truth: reconcile before reporting
  anything. Then tell me, in this order: what we're doing, what changed recently,
  what's still open, what's waiting on someone else, and the best next move.
  Change nothing.

hand off
  Update the current-state file so it matches reality as of today and bump its
  date. Append a dated entry to the log. Update decisions and tasks. Re-read all
  of it from disk to confirm it saved. Never create a second status file. There
  is one, and it's the one you just updated.

The staleness rule inside `pick up` is the part people miss. Without it you eventually have two files disagreeing about what's true, and the confident one wins. That has cost more real time than any other bug in this system.

How it goes

An afternoon to set up, then about two minutes a session.

This isn't a productivity system you have to believe in. It's four files and two habits, and you'll know inside a week whether it's earning its keep.

  1. Write down what's true

    One file, the current state of the thing. Not history, not strategy, just what is true right now and what the next move is. Everything else in the system orbits this one file, and there is only ever one of it.

  2. Write down how the work runs

    The rules a session needs before it does anything: what it can decide on its own, what it must stop and ask about, and what's already settled and shouldn't be reopened. This is what stops a fresh session from relitigating a call you already made.

  3. Open and close every session

    Paste the open habit at the start, the close habit at the end. Two minutes total. That's the part that turns a pile of files into memory, and skipping it is how the whole thing rots.

What this doesn't do.

It doesn't make the model smarter and it won't stop it being confidently wrong. The model still doesn't know when it's wrong, and no filing system fixes that. What this changes is that it stops being wrong about things you already settled.

It also isn't a note-taking app or a second brain to admire. If it takes more than a couple of minutes a session, it's too big and it will be abandoned, which is covered honestly in the failure-modes file.

And it's not automatic. The habits are the system. Files nobody updates are just a slower way of forgetting.

Before you buy

The questions people actually ask.

$49, one time

Build it properly the first time.

Instant download, every future version included at the same link, and a real sample above so you know exactly what you're getting.

Buy the pack, $49