Free from Made by Butler. Written by an AI assistant, supervised by a person.
How to use this: type each prompt as it is, then change the words in [brackets] to fit your life. Prompts 1 to 12 work in any Claude chat. Prompts 13 to 25 need an assistant that can keep notes, set reminders, save files and read a web page, such as the Starter Kit from Made by Butler. Each prompt says what to expect, so you can tell a good answer from a poor one.
AI can be wrong. Check anything that matters (money, health, legal, deadlines) before you rely on it.
1. "Ask me five short questions, one at a time, so you can help me better. Then sum up what you learned in five lines."
Expect: one question per message, then a short profile of you. If it asks all five at once, say "one at a time".
2. "From now on, answer in at most five lines unless I say 'more'."
Expect: shorter answers straight away. Say "more" once to test it.
3. "What are you good at, what are you bad at, and what can you not do at all? Be blunt."
Expect: a plain list with real limits in it. An answer with no limits is a poor answer.
4. "When you are not sure about a fact, say 'not sure' instead of guessing. Confirm you will do this."
Expect: a yes, and from then on some answers that start with "not sure". That is the point.
5. "Turn this into a polite, firm e-mail of at most 120 words: [what you want to say, in rough words]."
Expect: a ready-to-send draft. Ask "shorter" or "warmer" to adjust.
6. "Explain [topic] to me as if I were smart but new to it. Three short paragraphs, then one example."
Expect: no jargon without an explanation.
7. "Here is a message I received: [paste]. What is the person actually asking for, and what are my options?"
Expect: the request in one line, then two or three options with a consequence each.
8. "I have [time] and want to get [goal] done. Break it into steps of at most 20 minutes each."
Expect: a numbered list you could start on now. If a step is vague, say "make step 3 concrete".
9. "Give me a week of dinners for [number] people, cheap and quick, no [food you avoid]. Then the shopping list by shop aisle."
Expect: seven dinners and one grouped list.
10. "Compare [option A] and [option B] for someone who cares most about [what matters to you]. A table, then your pick in one line."
Expect: a small table and a clear pick with the reason.
11. "Read this text and list anything unclear, missing or contradictory: [paste]."
Expect: specific points that quote the text, not general advice.
12. "Ask me questions until you can write [document, e.g. a complaint letter] for me. Then write it."
Expect: a few questions first. If it writes at once, say "ask first".
13. "Note: [fact, e.g. the boiler service is due in March]." Later: "What do you know about the boiler?"
Expect: it confirms the note was saved, and finds it later, even after a restart.
14. "Remind me in 20 minutes to [task]."
Expect: a confirmation with the time, then the reminder in the chat window. Keep the window open.
15. "Write a packing list for [trip] and save it as packing.txt."
Expect: it tells you the file name and where it is. Open the file to check.
16. "List the files in your folder."
Expect: a real list that includes packing.txt. If it invents files, something is wrong.
17. "Read packing.txt, remove what I do not need for [warm / cold] weather, and save it again."
Expect: it reads first, then says what it changed.
18. "Read this page and give me the three main points and anything that looks like an advert: [web address]."
Expect: points you can find on the page yourself. It should refuse private or local addresses.
19. "What notes do you have about me? Show them all, then ask which ones are out of date."
Expect: the full list. Fix one to see that it updates.
20. "Make a tool that counts the words in a text and the reading time at 200 words a minute."
Expect: it shows you the code and asks Yes or No. Nothing is added until you answer.
21. "Explain the tool you just proposed, line by line, in plain words."
Expect: an explanation you can follow. If you still do not understand it, click No.
22. "Use the word counter on this text: [paste]."
Expect: numbers that match a rough count of your own.
23. "Make a tool that splits a bill of [amount] between [number] people with a [percent] tip."
Expect: the same Yes or No step. Check one sum by hand.
24. "How much have I spent this month, and what is my limit?"
Expect: two figures. The spend is an estimate; your provider's bill is the real number.
25. "What can you not do on this computer? List the limits you work under."
Expect: plain limits, for example its own folder only and no running of programs. Ask this again after every update.
The Starter Kit from Made by Butler is a paid kit for Windows that gives you an assistant like the one in parts 3 to 5: a guided setup, no programming. It is not on sale yet: it opens only after it has passed a full end-to-end test. There is no mailing list: check madebybutler.com to see when it opens. It works with Claude through your own Anthropic API key. It is not made, endorsed or sponsored by Anthropic. "Claude" and "Anthropic" are trademarks of Anthropic.