PROMPTS THAT MAKE AI DO THE WORK.

Free copy-paste prompts for Claude Code and ChatGPT. Built from what we teach in our workshops — the same frameworks, no signup, no email. Copy one, paste it into your agent, read the plan before you approve it.

STEP 01

COPY

Hit the copy button on any prompt below.

STEP 02

PASTE

Drop it into Claude Code or ChatGPT and hit enter.

STEP 03

READ THE PLAN

It shows you what it's about to do before it touches anything. Read it. That's the whole skill.

SIX PROMPTS. YOUR FIRST SYSTEM.

Run them in order and you'll have a workspace that knows your business, respects your rules, and drafts your weekly work. Or grab the one you need.

[01]

The Agent Interview

Works in
Claude Code
Time
~10 min

Every new chat starts with amnesia — the agent knows nothing about you or your business. This prompt turns that around: it interviews YOU, then writes what it learns into CLAUDE.md, the file it reads every time it wakes up in that workspace.

How to use it
  1. 1.Open Claude Code and open the folder you want it to work in.
  2. 2.Paste the prompt. Answer its questions — out loud is fine, turn dictation on.
  3. 3.When it shows you CLAUDE.md, correct anything it got wrong. Those corrections are the most valuable part.
$ The prompt
You're setting up as my assistant in this workspace. Before you write anything, interview me.

Ask me 8–10 questions, a few at a time, about: what my business does, who my customers are, the documents I produce every week, how I like things written, and the things I never want done.

When the interview is done, write my answers into a file called CLAUDE.md at the top level of this workspace. Plain headings, no fluff, keep it under one page.

Then read it back to me so I can correct anything.
[02]

The ChatGPT Bridge

Works in
ChatGPT
Time
~5 min

Been using ChatGPT for a while? It already knows your tone, your rules, your formats. This prompt exports that knowledge as a markdown file you can drop straight into a new Claude Code workspace — so it starts smart, not blank.

How to use it
  1. 1.Run this in the ChatGPT chat where you've done most of your work.
  2. 2.Answer its two or three scoping questions.
  3. 3.Save the file it produces into your new workspace folder.
$ The prompt
I'm setting up a dedicated workspace on my computer for one area of my work: [THE ONE THING — e.g. social content for my clients / my monthly client reporting / my proposals].

Look back through what you know about me and what we've done in this chat, and pull out only what applies to that one area. I want the rules, not the history.

Give me a single markdown file containing:
- how I want the work done — tone, format, structure, length
- the things I never want done
- any templates or formats I reuse
- specifics about the business it's for

Leave out: anything about other areas of my work, examples of past outputs, and general advice about the topic.

If something only came up once and you're not sure it's a standing preference, don't put it in the main file — list it at the bottom under "check these".

Before you write anything, ask me two or three questions to make sure you've scoped it right.

Pick ONE area of your work. A single tight file beats three thin ones — you can repeat the prompt for other areas later.

[03]

The Never-List

Works in
Claude Code
Time
~5 min

Telling an agent what NOT to do is as important as telling it what to do. That's your fence. This prompt builds the guard rails into your CLAUDE.md — and writes your most important rule twice, so it can't be forgotten.

How to use it
  1. 1.Run this in a workspace that already has a CLAUDE.md.
  2. 2.Tell it your hard rules — the things it must never do, ever.
  3. 3.Read the result. If a rule matters enough to say, it matters enough to check.
$ The prompt
Read the CLAUDE.md in this workspace. Add a section called NEVER.

Ask me for the things I never want done in this workspace — the hard rules, not preferences. Write them as short, direct commands.

Then take the single most important rule and write it twice: once at the very top of the file and once at the very bottom, each marked clearly. Rules written twice don't get forgotten.

Show me the finished file before you save it.
[04]

The Desktop Cleanup

Works in
Claude Code
Time
~5 min

The fastest confidence builder there is. Point it at a messy desktop and watch it organise years of chaos into folders in minutes — without deleting a single thing. Perfect first job for a new workspace.

How to use it
  1. 1.Open Claude Code in the folder you want cleaned (yes, your Desktop can be a workspace).
  2. 2.Paste the prompt. Answer its questions.
  3. 3.Read the plan before you approve it. Nothing gets deleted — everything doubtful goes to a review folder.
$ The prompt
I've had this computer for years and there's a bunch of files and folders on my desktop that may be redundant. Clean it up and organise it into folders.

Don't delete anything. If something may be redundant, put it in a folder called "to-be-reviewed".

Ask me a couple of questions first, then show me your plan before you move anything.
[05]

The Weekly Document

Works in
Claude Code
Time
~15 min

Everyone has one — the document you produce every week that you wish drafted itself. This prompt doesn't write it for you once. It builds the machine: a repeatable process you run forever after.

How to use it
  1. 1.Put a couple of past examples of the document in your workspace.
  2. 2.Paste the prompt and name your document.
  3. 3.It gives you a plan first — read it, fix its assumptions, THEN let it build.
$ The prompt
Every week I produce [NAME THE DOCUMENT — e.g. my client status report / my social media calendar / my quoting summary].

Look at the examples in [FOLDER]. Draft a plan for how you would produce this document for me from now on:

- where you get the inputs
- what format and structure you follow
- what you'd need from me each week
- what you'd never guess or invent

Don't produce the document yet. Show me the plan first and wait for my approval.
[06]

The Delegation Brief

Works in
Claude Code or ChatGPT
Time
~5 min

"Make this better" is how you get generic back. A real brief answers three questions: what does done look like, what does it need to know, and what's off limits. This prompt rebuilds a lazy instruction into a proper brief — and teaches you the habit.

How to use it
  1. 1.Paste the prompt along with whatever vague instruction you were about to give.
  2. 2.Answer its three questions plainly — one long response beats many short ones.
  3. 3.Compare the rebuilt brief to your original. That gap is the skill.
$ The prompt
I'm about to give an AI this instruction:

"[PASTE YOUR VAGUE INSTRUCTION — e.g. make this email better]"

Don't run it. Instead, turn it into a proper delegation brief by asking me three questions:

1. What does DONE look like? (the specific result I want)
2. What does it need to KNOW? (context, examples, constraints)
3. What's OFF LIMITS? (the things it must not do or change)

Once I've answered, write the rebuilt brief in my words, ready to paste. Tell me what the original instruction would have made it guess about.

TEACH YOUR AGENT TASTE.

A taste file is a set of design rules — fonts, colour, spacing, motion, and a never-list — that an agent reads before it designs anything. Drop one into your project, tell your agent to follow it, and every page it builds comes out with a point of view instead of a default. These are the same files we use on client work.

[T1]

Swiss Brutalist

Bauhaus meets tech startup. Zero radius, hard shadows, one red accent. The taste file behind this website.

TYPECOLOURSPACINGMOTIONNEVER
.MD
[T2]

Bauhaus

Form follows function. Jost + Inter, primary colours as structure, geometry as layout.

TYPECOLOURSPACINGMOTIONNEVER
.MD
[T3]

Modern Retro

Sun-faded '70s warmth with modern spacing. Fraunces + Karla, cream and espresso, arches and badges.

TYPECOLOURSPACINGMOTIONNEVER
.MD

How to use: copy or download a file, save it as DESIGN-TASTE.md at the root of your project, then tell your agent — "read DESIGN-TASTE.md and follow it for every visual decision." The file includes the exact install prompt.

THESE PROMPTS ARE THE FIRST HOUR OF THE WORKSHOP.

In three hours, in person, you'll set up your own workspace, build your first real automation, and leave with the confidence to keep going. Small group. Your laptop. Real work.