41 prompts from real builds, rewritten so anyone can use them. Fill in the [brackets], copy and paste into any AI.
What do I say here?
Five openers, one per rank
1
StuckGive it a role
You are a [role, like a career coach]. I want [the result]. Ask me 3 questions before you answer.
2
AskerGive your real limits
Here are my real limits: [hours a week], [budget], [what I already know]. Plan [goal] inside them.
3
DirectorAsk for one pick, then a check
Give me your one best recommendation for [problem]. Then check it: what would make it fail?
4
BuilderDraft first, then chain
Draft a first version of [thing]. Then list the steps so I can repeat it next time, with before and after.
5
OperatorOne role per agent
You are one agent on a team. Your only job is [role]. End with a handoff note: done, open, next.
The library
41 shown
216-001AI agentsBuilder
Make every device a workplace for agents
You are a systems designer. I own these devices: [list with specs]. These AI agents work for me: [list]. Design how the agents share work across devices so any of them can pick up where another stopped. Give each device one job, show a handoff step by step, and list what I must do by hand versus what agents can do alone. Make it concrete enough to set up today.
Why it works: It gives each device one job and separates your hands from the agents' hands.
Not tested yet
216-002AI agentsBuilder
Borrow the best feature from every tool
I'm building [my product]. Here are the tools I use: [tools]. For each tool, name the one feature it does better than the rest and why it works. Rank the three features my product should borrow first and sketch how each would look in my product.
Why it works: It treats every competitor as a teacher and only borrows what's proven.
Not tested yet
216-003AI agentsBuilder
Write a handoff another agent can't misread
Write a handoff message from [sender agent] to [receiver agent] in three parts. Assignment: the one thing to do and what done looks like. Context: only what the receiver needs, with file paths and links. Evidence: what has been checked already and how to prove the work is finished. Keep it under 200 words.
Why it works: A fixed shape means nothing important gets left out between agents.
Not tested yet
216-004AI agentsBuilder
Split the work by what each model is best at
Here are the AI models I have access to: [models and plans]. Here are the jobs I need done every week: [jobs]. Search for current, dated benchmarks and reviews, then assign each job to the model that is strongest at it. Flag any job where the cheapest model is good enough. Show it as a table.
Why it works: It stops you from using one model for everything.
Not tested yet
216-005AI agentsBuilder
One place for every decision
My agents work in many threads and keep asking me things. Collect every open question into one list. For each, give your recommended answer and a one-line reason, so I can answer all of them at once without opening each thread.
Why it works: Every question arrives with a recommendation.
Not tested yet
216-006ProductDirector
Draft first, build later
I'm going to give you ideas one at a time. Do not build anything yet. Keep one running draft page with a single recommended shape, and add each new idea to it, showing where it fits or why it doesn't. When I ask "how's the draft looking?", give me the draft, your one recommendation, and the exact line I can say to approve it.
Why it works: You can think out loud without the AI rushing to build each idea.
Not tested yet
216-007ProductDirector
Show me four looks, then pick one
Design 4 different looks (A to D) for [page or app]. Each one should take a different direction, named in a few words, and borrow from a real reference I can look up. For each, show the main screen on desktop and phone. Then tell me which one you would pick for [audience] and why.
Why it works: Choosing between real options is easier than describing taste in words.
Not tested yet
216-008ProductDirector
Prove the change with before and after
Before you change [screen or feature], take a screenshot of how it looks now. After the change, take the same screenshot again. Show both side by side, list every difference you meant to make, and point out any difference you didn't mean to make.
Why it works: You see the change instead of trusting the AI's description of it.
Not tested yet
216-009ProductDirector
Research my words before you design
I'll describe what I want in my own words: [description]. Before designing anything, search for 5 real examples or references that match those words and show them to me with links. Ask me which ones are closest. Only then make the first version, based on the ones I picked.
Why it works: It turns vague words into pictures you can point at.
Not tested yet
216-010ProductDirector
Name it so people want to go there
I need a name for [feature] in [product]. Give me 5 names. For each, say in one line how it feels to a first-time user. Pick the one that sounds like a place someone wants to be, not a chore.
Why it works: It judges names by how they feel to a user.
Not tested yet
216-011BusinessDirector
Stop opening new income lanes
Here is everything I'm trying to make money from: [list]. Here are my monthly costs: [costs]. I'm losing track. Pick the 3 lanes to push for 30 days, say what pauses and what stops, give each lane one number to hit by day 30, and tell me honestly if a part-time job would beat all of them.
Why it works: It forces the AI to say no to some ideas and set a number for each lane.
Not tested yet
216-012BusinessDirector
Where does the money actually come in?
I sell [products]. Right now nothing takes payment. Recommend one way to get paid for all of them, explain who handles taxes and refunds, and list the steps only I can do by hand, in order, with the cost of each.
Why it works: It starts from the honest fact and ends with your hand steps.
Not tested yet
216-013BusinessDirector
What am I missing that nobody is doing?
Here is my product: [product and who it's for]. Look at it as an outsider who has seen every similar product. List 5 things no one in this space is doing yet, and for each, say who would pay for it and how I could test it in a week.
Why it works: It asks the AI for what you can't see from the inside.
Not tested yet
216-014BusinessDirector
Design rewards people can't cheat
I want to reward users of [app] for [behavior]. Design a reward system they will enjoy. Then switch sides: act as a player trying to farm it, and list every way to cheat. Fix each one without making the rewards feel stingy.
Why it works: Breaking your own design before users do saves you a rebuild.
Not tested yet
216-015ResearchDirector
Where are we, really?
Give me a 3-part report on [project]: what's done (with links as proof), where we are in reality right now (checked against current facts with dates), and the road ahead (the next 3 steps in order). Keep each part short. Mark anything you couldn't verify.
Why it works: The same 3 parts every time make it easy to compare week to week.
Not tested yet
216-016ResearchDirector
Be my challenger, with sources
Here is my idea: [idea]. Act as my challenger. Search for current facts with dates, and tell me where I'm right, where I'm wrong, and what someone already does better. Don't trust what you remember. End with your one recommendation and what would change your mind.
Why it works: It asks for dated sources and makes the AI disagree when the facts say so.
Not tested yet
216-017ResearchDirector
Is anyone already selling this?
I think [idea] could make money. Find who sells something like it today, with prices, volume and dates. Show their weak spots. Tell me what I should copy, where I can beat them, and whether this is a crowded market or an open gap.
Why it works: It turns a hunch into a market scan.
Not tested yet
216-018ResearchDirector
Show me what you looked at
Every time you research [topic] for me, keep a log: what you searched, what you opened, what you used, and what you looked at but left out and why. End with a list of what to keep watching next time.
Why it works: You stop wondering whether the AI saw something.
Not tested yet
216-019LearningAsker
A teacher for any subject
Be my teacher for [subject]. First ask me 3 questions to find my level. Then teach one small idea at a time, at my pace, with a quick check after each. Take the lead on what comes next, but slow down whenever I say so. Keep a short list of what I've learned with proof.
Why it works: It sets the pace by your answers instead of rushing through.
Not tested yet
216-020DevicesBuilder
See all my machines in one view
I have these devices: [list]. Design one page that shows each device's storage, memory, temperature, what's running and what needs attention, updated automatically. Tell me which numbers matter most for each device's job, and what should trigger an alert.
Why it works: It turns many screens into one.
Not tested yet
216-021AI agentsBuilder
Done means it comes back
When you hand [task] to [other agent], include what done looks like and ask them to reply in this same chat with a done receipt: what changed, where, and how they checked it. Don't tell me it's finished until that receipt is here.
Why it works: A handoff that never comes back isn't done.
Not tested yet
216-022LearningAsker
I'm new here, give me a map
I'm new to [tool]. Before we do anything, give me a one-page map: what it is, the 5 things I'll use most, what each screen or command is for, and one safe thing to try first. Then wait for my question.
Why it works: Admitting you're new gets you a map instead of jargon.
Not tested yet
216-023ProductDirector
Audit first, fix later
Do a complete audit of [app or page] against [reference apps]. List every problem you find, ranked by how much it hurts the user. Don't fix anything yet. I'll pick what to fix.
Why it works: You decide what gets fixed before anything changes.
Not tested yet
216-024ProductDirector
Keep this, change that
[Screen] is my base template. Make every other screen match it: same spacing, type, colors and borders. List every place that didn't match, and show me before and after for each one.
Why it works: One base version stops piecemeal changes.
Not tested yet
216-025ResearchDirector
Find the real cause
[Thing] is slow or broken. Treat it as an engineering problem: list the possible causes, test the top 3 one at a time, and show me what you measured. Don't just switch tools or models until you know why.
Why it works: It stops the AI from guessing at fixes.
Not tested yet
216-026ResearchDirector
Test one change
I changed [setting] from [A] to [B]. Run the same task both ways and compare speed, quality and cost in a table. Tell me which to keep and why.
Why it works: One change at a time shows what actually matters.
Not tested yet
216-027AI agentsBuilder
Prove the wire with a tiny test
Before we build anything on [connection between tools or agents], run the smallest possible end-to-end test: send one simple question through it and show me the answer coming back, with timestamps.
Why it works: A tiny test proves the connection before you build on it.
Not tested yet
216-028AI agentsBuilder
Turn this failure into a rule
This just went wrong: [what happened]. Write down what happened, how, and why. Then write one rule that stops it happening again, and say where that rule should live so every future session follows it.
Why it works: Each failure makes the system stronger.
Not tested yet
216-029AI agentsBuilder
Hand off before the context fills
We're deep into this session. Write a handoff for a fresh session: the goal, what's done with proof, what's next, open decisions, and the files to read first. Keep it under one page.
Why it works: A fresh session picks up where this one stopped.
Not tested yet
216-030AI agentsBuilder
One job per agent
Here are my agents: [list]. Give each one a single role, say what it should never do, and write the one-line rule I can paste into each so it stays in its lane and hands off the rest.
Why it works: Clear lanes stop agents from stepping on each other.
Not tested yet
216-031ResearchDirector
Proven vs guessed
Research [topic]. Split your answer into two lists: VERIFIED facts, each with a source link and date, and ESTIMATES, each with how you got it. Keep it lean, with no filler. End with the one thing you're least sure about.
Why it works: You can see at a glance what's proven.
Not tested yet
216-032ProductDirector
Show me the gap with references
Here's what we have now: [screenshots or description]. Collect 6 references for what great looks like in this style, covering [screens or parts]. Put ours next to each one and list exactly where we fall short, in order of what matters most.
Why it works: References turn "make it better" into a list of fixes.
Not tested yet
216-033ProductDirector
Give me numbered picks
Give me 4 options for [thing], labelled P1 to P4, each in two lines with one image or example. I'll reply with numbers like "P1, P4" or "all". Then build only what I picked.
Why it works: Choosing by number is fast and leaves nothing vague.
Not tested yet
216-034ContentDirector
The overreactor and the calm friend
Write a 30-second skit about [situation] with two characters: one who overreacts to everything and one who stays calm and explains something ridiculous. Keep it to 6 lines, end on the calm friend's punchline, and add one note per line for the visuals.
Why it works: A fixed duo makes every skit easier to write.
Not tested yet
216-035BusinessDirector
Prep it for the lawyer
Act as in-house legal prep for my small business, [business and what it sells]. List what I need (licenses, terms, privacy, refunds, age rating), draft first versions, and flag every point a real attorney should review. Label everything "not legal advice".
Why it works: You arrive at the lawyer ready, which saves time and money.
Not tested yet
216-036AI agentsBuilder
Only the skills each agent needs
Here are my agents and what each one does: [list]. Sort my skills into general ones every agent gets and specialized ones only certain agents need. Show who gets what in a table, and flag any agent that has a skill it shouldn't.
Why it works: Lean agents stay focused and cost less.
Not tested yet
216-037ContentDirector
Iterate one concept, one change at a time
Make a concept image of [what you're imagining] in [style]. After each version I'll give one change, like "make it a globe" or "bigger skyline". Keep everything else the same, and number each version so I can say "go back to v2".
Why it works: One change per round keeps the good parts and makes it easy to steer.
Not tested yet
216-038ContentDirector
Front, side and back reference sheet
Turn [character or object] into a reference sheet: front, side and back views in [style], on a plain background, same proportions in every view. Label anything you had to guess for the unseen sides.
Why it works: A builder or another AI can model from it, and the guesses are marked.
Not tested yet
216-039ResearchBuilder
Let two AIs argue, then show me the proof
Here is one question: [question]. Answer it, then argue the strongest opposite view as a second expert. List where you disagree, what evidence would settle each point, and end with one thing I can test this week. Agreement isn't proof, so say what's still unproven.
Why it works: A debate shows the weak spots one answer would hide.
Not tested yet
216-040LearningAsker
Teach it to me visually
Teach me [topic] for someone at my level, [beginner or some experience]. Use a comparison table, one simple chart or diagram described step by step, and a 5-line summary. End with a 3-question quiz.
Why it works: Seeing it laid out sticks better than a wall of text.
Not tested yet
216-041AI agentsBuilder
A status brief at set times, not every hour
Every day at [times], send me a status brief on [project]: what got done, with proof, what's stuck and why, and the one decision you need from me with your recommendation. Keep it under 10 lines. Don't message me in between unless something breaks.
Why it works: Fixed times and a short shape stop updates from turning into noise.
Not tested yet
Fit Kits · soon
Make it fit you · $29
A prompt set built for your situation and tested on Claude, GPT and Grok. Not for sale yet, and nothing here takes payment.