Six prompting habits

The small changes in how you ask that make the biggest difference to what comes back, with an instead-of table you can keep open.

You will be able to ask in a way that gets a usable first answer, and put the habits into the brain so you stop having to repeat them.

None of these is a trick. They are the habits of someone who briefs a capable person well, applied to an assistant that reads your brain before it reads your message. Most of them come from Nate Herk's teaching, and the phrases that are his are marked; the reading is in where these ideas come from. The CRIT method is the longer form for a real brief; this page is for the hundred small asks in between.

Habit one: brief it like a manager, not a coder#

"You are a manager of AI agents, not a coder" is Nate Herk's line, and it is the mindset the rest depends on. You are not writing instructions for a machine. You are handing work to a colleague who has read your files, so write what you would say to that colleague: what the job is, who it is for, what good looks like, and where to find what they need.

A two-person accountancy asking "do the client summary" gets a generic summary. Asking "write the quarterly summary for Marsh & Co the way we sent the last one, and read their page first" gets one that names the right people and the right numbers. The second ask is not longer because it is formal. It is longer because it says where to look.

Habit two: say what to do, and why#

"Give the why, not the prohibition" is also Nate Herk's, and it changes drafts more than any other habit here. A bare "don't" is a wall with no context, and the assistant will find the nearest gap in it. The reason lets it generalise.

"Don't use jargon" produces a draft that avoids the three words you had in mind and uses six others. "This goes to a client who finds tax stressful, so plain words and short sentences" produces a draft written for that person, including in the places you had not thought to forbid. State the goal, the reader, and the constraint with its reason, in that order if you can.

Habit three: make it prove it#

"Make it prove it" is Nate Herk's phrase for a plain rule: before you accept that something is done, ask for the evidence. An assistant that has read sixty pages can be confidently wrong about the sixty-first, and the fix is not distrust, it is a standing question.

Before you tell me this is done, point to what proves it: the page
you wrote, the figure you read, the message you found. If something
is not verified, say so plainly.

Ask it once per session and you will notice the answers change shape: fewer smooth summaries, more "here is the line in the page that says so". Where it cannot point at anything, that is your signal to check yourself before acting. What it will not do is the honest list of where that matters most.

Habit four: match the effort to the job#

There is no effort dial on your mineral; the words in your ask do the work. A quick lookup and a careful piece of writing want different amounts of thinking, and the assistant calibrates from how you phrase it.

"Quick pass: which three invoices are overdue?" gets a list. "Take your time on this one: draft the letter to the landlord and think about what could go wrong before you write" gets a draft with the problems named at the top. Neither phrase is magic; they set the expectation the same way they would with a person, and the difference in what comes back is real.

Habit five: let it act once it has enough#

Over-planning is its own way of never starting. If you have given the context and the ask, let it go and do the thing, then correct the result. Correcting a draft that exists is faster than describing the perfect draft in advance, and the correction teaches the brain something the description never would.

The exception is anything that leaves the mineral or changes something you would not want changed. There the standing rule is the reverse: propose first, then act on your yes. The assistant drafts and never sends; for writing into your brain or your connected services, "show me before you write it" is a sentence worth keeping in the workflow page.

Habit six: say less as the brain grows#

In the first week you will write long asks, because the brain is thin and you are filling the gaps by hand. That is right for week one and wrong for month three. As the pages fill, the context step of every brief becomes "read the page", and the ask itself can shrink to what a colleague of six months would need: "reply to Jo, keep it short".

If you find yourself still writing paragraphs in month three, the thing to fix is the brain, not the ask. Grow the brain is the page for that, and the test is simple: if you have explained it twice, it belongs in a page.

The instead-of table#

Put the habit in the brain#

Every habit above is a sentence you will otherwise type every day. The better place for a sentence you type every day is the workflow layer of your brain, wiki/_layers/8-workflow.md, which the assistant reads at the start of every session. Say it once in the Terminal:

Add to my workflow page: always point to the evidence before saying
something is done, and show me what you are about to write to a
page before you write it.

From then on the habit is the assistant's, not yours to remember. The same page is where your voice lives, which is why teach it your voice sends you to the same file, and the way to find out which habits are already there is to ask: /explain reads the mineral and answers from what it finds.