The exact shape of the answer you want back.
What the model must not do, and where it must stop.
The situation, in the detail somebody new would need.
What it should ask you before it starts.
Who it is being, and what that person would know.
The rules the answer has to obey.
How much of its thinking you want shown.
One or two of what good looks like.
How you will know the answer is good.
The thing to do, in one sentence.
3 of 10 filled in
3 of the 6 load bearing parts are empty
Your prompt
OUTPUT A welcome email, subject line and body, ready to paste. CONTEXT A skincare brand doing $1.4M a year, selling direct to consumers. TASK Write the first email new subscribers get after they sign up.
Still empty
- WarningSTOPMost disappointing answers are a model continuing past the point you wanted.
- WarningCONSTRAINTSConstraints are what separate a usable answer from an impressive one.
- WarningSUCCESSIf you cannot say this, you cannot tell whether the answer worked.
- PassASK FIRST (optional)It turns a guess into a question, which is cheaper than a confident wrong answer.
- PassROLE (optional)A role narrows the vocabulary and the assumptions. It is not a costume.
- PassREASONING (optional)Shown reasoning is checkable. Hidden reasoning has to be taken on trust.
- PassEXAMPLES (optional)One example moves an answer further than three paragraphs describing the same thing.
Everything here runs in your browser. Your numbers are never sent anywhere, not to me and not to anyone else.
How this is calculated, and what to put in
Why the order matters
OUTPUT and STOP come before anything about the work. That is deliberate, and it is the whole framework.
A model told the shape first produces that shape. Told the task first, the shape becomes a correction you make afterwards.
STOP is second for a reason. Most disappointing answers are a model carrying on past where you wanted.
TASK comes last on purpose
By the time it arrives, the model knows the terms and the rules. It knows what finished looks like.
Put the task first, and every line after it revises something already begun.
You do not need all ten
Six are load bearing. The rest earn their place on harder work. On simple work they get in the way.
The tool says which are empty and what each one costs you. It does not nag you to fill them all in.
There is no score
A prompt cannot be scored. The only measure is whether the answer was any good.
That is what SUCCESS is for. If you cannot write that line, no amount of structure above it will help.