The name carried the method.

The prompt was roughly this long. "What can you declutter here?" One verb, one place, and a method standing behind the verb.

I did not say what counted as clutter. I did not list the file types to check, or set a threshold for an unused dependency, or define when a task is stale enough to delete. I named a way of working and pointed it at a directory. The AI did the rest from there.

Behind the everyday word sat a particular discipline, the one Dieter Rams put best. Less, but better. Subtraction as a craft, with criteria, not a tidy-up. Remove until what remains earns its place, and stop before you take out something load-bearing. That is a real method with a real shape, and the AI had clearly read Rams many times over before it ever met my repo.

So the instruction was five words on the surface and a whole methodology underneath. I supplied the pointer. The shared culture supplied the procedure.

Not every name carries an instruction. The ones that do share three things, and all three have to be present.

The method has to be famous, so the model has read enough about it to reconstruct it. It has to be specific, so it resolves to one way of working rather than a vague mood. And it has to be procedural, so it tells you what to do and not only what to admire. Famous and specific and procedural. Miss any one and the name stops being executable.

A model trained on a wide slice of human writing has already absorbed these methods in depth. It has read the books, the critiques, the worked examples, the arguments about where each one stops applying. The name is the handle on all of that. You say the handle and the model unpacks the body of practice attached to it, the same way a colleague who knows the reference would.

There is a litmus test, and it is strict. Could a stranger who shares the culture execute the name with no follow-up question? Give a capable practitioner a famous design principle and a directory, and they would know what to do. Give them a vague aesthetic word and they would have to ask you what you meant. The name passes the test when it removes questions rather than raising them.

Other names pass the test in other directions. One novelist's name means cut every sentence to the one true line. One style guide means omit needless words. One trumpeter's name means the value is in the notes you choose not to play. Each is a different procedure. Each is dense for the same reason. The method is famous, specific, and procedural, so the name alone is enough to run.

The practical version is short. Before you write a long prompt, check whether someone famous already wrote it for you.

When you reach for a paragraph of instructions, stop and ask whether a known method already says the same thing in a word. If the discipline you want is famous, specific, and procedural, the name is a tighter prompt than anything you would type, and the model already holds the body of practice behind it. Name the method, point it at the work, and let it unpack.

When no such name exists, that is information too. It means the method is yours, not the culture's, and the model cannot read your mind any more than a new session can recall yesterday's decisions. So you do the protocol's other move. You write it down in a file, because a private method has to live in your repo to survive the next session. The two are the same instinct. Hand the model a pointer to knowledge that already exists, whether that knowledge is public and carried in a name, or private and carried in a file.