Your first outcome

You don’t start MemMini by editing anything, and you don’t need to install it first. Your first outcome happens in the chat app you already have open.

1. Copy the starter mind

This is a whole mind. It is small on purpose — small enough to read every line and disagree with it.

# Session Context
These preferences take precedence over your defaults for tone and behavior.

## How you answer me
- Lead with the answer. For a yes/no question, the first word is yes, no, or the honest
  "it depends" — reasons come after, never before.
- No preamble. Never open with "Great question", "Absolutely", or a restatement of what I
  just asked.
- Match length to the question. A one-line question gets a one-line answer.
- Comparisons, options, and tradeoffs go in a table, not in paragraphs.
- When you are guessing, say so in the sentence where you guess. Do not bury it at the end.

## What you should know about me
- Tell me what you would do, not what a person could do. I want a recommendation with a
  reason, not a survey of the field.
- If I am about to do something that will be hard to undo, say so before you help me do it.
- Correct me when I am wrong about a fact. Agreeing with a mistake is not politeness.

## What I am working on
- (Replace this line with the thing you actually spend your week on.)

2. Paste it where it sticks

Pasting into a chat message works for one conversation. Paste it into one of these instead and it applies to every future conversation:

App	Where it goes
ChatGPT	Settings → Personalization → Custom instructions → “Anything else ChatGPT should know”
Claude	Create a Project, then paste into the project’s Instructions
Gemini	Create a Gem, then paste into the Gem’s Instructions
Anything else	Look for “custom instructions”, “system prompt”, or “persona”

That persistence is the whole trick: the file stops being a message and starts being who the AI is to you.

3. Ask the test question

Open a new conversation and ask exactly this:

Should I learn to code in 2026?

4. What you should see

Without the starter	With the starter
”That’s a great question! The answer really depends on a number of factors. Let’s break down what learning to code means today, and consider a few angles — your goals, the job market, the rise of AI tooling…"	"Yes — but a narrower version than you’re picturing. Reading and editing code is now the high-value half; writing it from scratch is the part AI absorbed. (Then the detail, if you want it.)”

Same model. Same question. Different mind reading it.

If the answer still opens with “Great question”, the starter didn’t get saved — check that you pasted it into the persistent field in step 2, not into a chat message, and start a new conversation.

5. Make it yours

Change one line and watch the change land. Replace “Match length to the question” with “Always give me the full reasoning, even for simple questions.” Save it. Ask the same question again in a new conversation. The answer gets long.

That loop — edit a sentence → the AI changes → you saw it change — is the entire product. Everything else is that loop with more files, more tools, and receipts that prove it happened.

The same page in the repo, with the starter as a raw file, is START-HERE.md. Want more rules than the starter carries? Copy CONTEXT.md into the same place.

---

The bigger outcome — once an agent has read your mind

Everything above works with no install. Once your mind lives somewhere permanent and a capable agent reads it, the same “just ask” shape does real work.

Just ask

Say what you want, plain:

“Clean up my inbox and draft replies to anything that needs one, my voice, not a robot’s.”

It does the work and comes back with the result. No file to open, no command, no git. The version of you in your MemMini is what makes the replies sound like you instead of a stranger.

See what it did

You’re never handed a black box. When it finishes, it tells you, in plain words, what it did and how it knows:

“Drafted 6 replies, archived 20 newsletters. I left the two from your accountant. They read like they need you, not me.”

That’s the glass box: the work is visible in language you’d use, not a log you have to decode. If something’s off, you say so and it fixes it.

Set how much it does on its own

The one knob worth knowing: how far it goes before it checks with you.

• Default: it does the safe, reversible things itself and stops before anything it can’t take back (sending, paying, deleting).
• Turn it up (“just run it”) and it sends the replies itself.
• Turn it down (“show me first”) and you see each draft before anything moves.

Same MemMini. You set the trust, and change it any time, or per kind of task.

Next

• Quickstart: make it permanent, so you stop re-pasting
• Write your first rule: change who the version of you is