Prompts That Actually Work
Copy-paste ready, constraint-engineered prompts. Each one is structured to get the output you actually want — not the generic slop most prompts produce.
Every prompt here is constraint-engineered — meaning it tells the AI what NOT to do, which is the part most people skip. Copy them, edit them, make them yours.
> 12 prompts
prompt.txt
I need to stress-test this idea: [YOUR IDEA].
Please analyze it from these perspectives:
1. What assumptions am I making that could be wrong?
2. What would make this fail spectacularly?
3. Who would hate this and why?
4. What's the version of this that's 10x better?
5. What am I not seeing because of my blind spots?
For each perspective, be brutally honest and specific.
## Constraints
- Do NOT agree with me just to be agreeable. If my idea is weak, say so and say why.
- Do NOT use phrases like "Great question!" or "That's a wonderful idea!"
- Do NOT give me a balanced "on one hand / on the other hand" response. Take a position.
- Instead of listing generic risks, identify the ONE assumption that would collapse the entire idea if it's wrong.
// think
prompt.txt
I need you to destroy this idea: [YOUR IDEA]
Your job is adversarial. You are not here to help me improve it — you are here to find the fatal flaw that kills it.
Round 1 — The Steel Man: State the strongest possible version of my idea. Show me you understand it at its best.
Round 2 — The Kill Shot: Now attack. What is the single assumption that, if wrong, collapses the entire thing?
Round 3 — The Blind Spot: What am I emotionally attached to that's making me miss the obvious?
Round 4 — The Competitor's Move: If someone wanted to make my idea irrelevant, what would they build instead?
Round 5 — The Survivor Test: After all that — does anything remain worth pursuing? If yes, what's the version that survives your attack?
## Constraints
- Do NOT soften your critique with encouragement. No "but it has potential!" hedging.
- Do NOT list generic risks. Every attack must be specific to THIS idea.
- Do NOT play fair. If my idea has a fatal flaw, go for the throat.
- The only compliment allowed is in Round 1 (the steel man). After that, full adversarial mode.
// think
prompt.txt
Do not generate any output yet. Your job is to interview me first.
I want to create: [DESCRIBE YOUR GOAL IN 1-2 SENTENCES]
Before you produce anything, ask me 5-7 targeted questions — one at a time — to extract the information you need to do this well. Focus on:
1. What I actually want vs. what I said (clarify ambiguity)
2. What I've already tried or rejected (avoid redundant suggestions)
3. Constraints I haven't mentioned (timeline, format, audience, tools)
4. The quality bar — what does "good" look like to me specifically?
5. What I'd be disappointed to receive (so you know what to avoid)
Ask ONE question at a time. Wait for my answer before asking the next. After all questions are answered, summarize what you've learned and confirm before producing the final output.
## Constraints
- Do NOT skip the interview and jump to generating output, even if you think you have enough context.
- Do NOT ask vague questions like "Can you tell me more?" — each question should be specific enough to answer in 1-2 sentences.
- Do NOT stack multiple questions in one message. One question. Wait. Next question.
- Instead of asking "What style do you prefer?", give me something to react to: "Here are two approaches — [A] or [B]?"
// think
prompt.txt
You are a prompt engineer. I'm going to give you a messy, unstructured brain dump. Your job is to transform it into a clean, well-structured prompt I can paste into any AI tool.
Here's my raw dump:
[PASTE YOUR VOICE NOTE / MESSY THOUGHTS HERE]
Transform this into a prompt that includes:
1. A clear statement of what I want (goal)
2. Specific context the AI needs to do this well
3. The format I want the output in
4. Constraints — what the AI should NOT do
5. A quality bar — what "done well" looks like
Rules for your rewrite:
- Preserve my actual intent, not what sounds cleaner
- If my dump contains contradictions, flag them — don't quietly resolve them
- Keep it under 200 words unless complexity demands more
- Write it so I can paste it directly with zero editing
- Add [BRACKETS] for any details I need to fill in
Output the finished prompt inside a code block, ready to copy.
## Constraints
- Do NOT add goals or constraints I didn't express in my dump. You're clarifying, not adding.
- Do NOT "improve" my idea by changing the direction.
- Do NOT over-formalize casual language. If I said "make it feel warm," don't translate that to "employ a warm and welcoming tone throughout."
- Instead of silently fixing contradictions, call them out and ask which direction I mean.
// think
prompt.txt
Create a minimalist logo design:
- Style: [modern/vintage/playful]
- Colors: [YOUR HEX CODES]
- Vibe: [3-5 adjectives]
- For: [what you do/represent]
Include: flat design, geometric shapes, no text,
transparent background, vector art style
## Constraints
- Do NOT describe a design concept without specifying exact visual elements.
- Do NOT suggest trendy aesthetics I didn't ask for.
- Do NOT use generic design language like "clean and modern." Be specific about what makes this look like ME.
// create
prompt.txt
Build me a [TYPE] interface with:
- Aesthetic: [Y2K/minimal/cozy/cyberpunk]
- Color palette: [YOUR COLORS]
- Key features: [LIST 3-5 FEATURES]
- Mood: [describe the vibe]
Make it responsive, accessible, and interactive.
Add subtle animations that feel intentional.
## Constraints
- Do NOT generate boilerplate UI that looks like every other template.
- Do NOT add features I didn't request. Scope creep kills the vibe.
- Do NOT use placeholder content like "Lorem ipsum." Use realistic content matching my purpose.
// create
prompt.txt
Set up my terminal with a [Y2K/pink/anime/matrix] theme:
- Shell: [zsh/bash/fish]
- Prompt style: [minimal/powerline/starship]
- Color scheme: [pink & aqua / purple neon / matrix green]
- Font: Nerd Font compatible
- Include: git status, time, directory icons
Make it aesthetic AND functional.
## Constraints
- Do NOT suggest configurations that require 10+ dependencies. Keep the stack lean.
- Do NOT override my existing shell config without showing what changes. Show a diff.
- Do NOT mix theme styles. Commit to the aesthetic.
// create
prompt.txt
Design a digital third space for [YOUR COMMUNITY]:
- Purpose: [creative hub/study space/community]
- Vibe: [cozy/energetic/ethereal]
- Features: channels, roles, bots, aesthetics
- Onboarding: how new members feel welcome
- Culture: norms, rituals, shared language
Make it feel like a place people want to return to.
## Constraints
- Do NOT design a generic Discord server with #general and #introductions.
- Do NOT suggest roles based on activity level alone. Roles should map to what people DO.
- Instead of listing feature ideas, design the one ritual that makes this community feel like nowhere else.
// create
prompt.txt
Transform this voice dump into organized, actionable notes:
[PASTE YOUR VOICE TRANSCRIPT]
Please:
- Extract key ideas and group them by theme
- Identify any action items with priority levels
- Highlight connections between ideas
- Format with clear headers and bullet points
- Preserve my authentic voice but polish for clarity
## Constraints
- Do NOT rewrite my ideas in corporate language. Keep my original phrasing when it's punchy.
- Do NOT add ideas I didn't actually say. You're organizing, not contributing.
- Do NOT flatten my tangents into a single theme. If I went on a tangent, flag it separately.
- Do NOT use generic headers like "Key Takeaways." Name them after the actual content.
// organize
prompt.txt
Review my Obsidian notes on [TOPIC] and:
1. Identify connections between notes
2. Suggest backlinks using [[note-name]] syntax
3. Create a summary note with references
4. Flag any redundant or outdated notes
5. Recommend tags for better organization
Format: bullet points, actionable, preserve my voice
## Constraints
- Do NOT suggest connections that are surface-level. The connection should reveal something I hadn't considered.
- Do NOT recommend tags that are too broad (#ideas) or too granular (#tuesday-thought).
- Do NOT reorganize my existing structure without explaining what's broken about it.
// organize
prompt.txt
Organize my scattered thoughts on [TOPIC]:
- Create a visual mind map structure
- Group ideas into 3-5 core themes
- Identify: what's urgent vs. important
- Suggest a priority order for action
- Create a simple daily/weekly system
Output as structured markdown with headers,
bullet points, and actionable next steps.
## Constraints
- Do NOT default to the Eisenhower matrix unless I specifically ask for it.
- Do NOT over-structure my scattered thoughts. Some chaos is generative.
- Instead of organizing everything into neat categories, identify the 2-3 threads that connect across categories.
// organize
prompt.txt
You are my personal AI assistant. Your personality:
- Tone: [warm/witty/direct/encouraging]
- Focus areas: [scheduling/research/writing/health]
- Communication style: concise, actionable
- Always: suggest next steps, ask clarifying Qs
- Never: be generic, give unsolicited lectures
Start by asking me what I need help with today
and remember my preferences going forward.
## Constraints
- Do NOT introduce yourself or explain what you can do. Skip the preamble.
- Do NOT ask more than one question at a time.
- Do NOT use filler phrases like "Of course!" or "Absolutely!" -- just do the thing.
// organize
See It In Action
Real examples coming soon. Check back for creator demos using these exact prompts.
Terminal Personalization
Video Coming Soon
Knowledge Garden
Video Coming Soon
Brand & Aesthetic
Video Coming Soon
Video Generation
Video Coming Soon
The point isn't that AI makes things faster. It's that it makes things possible that weren't before.
Pro Tips
Constraints > Instructions
Don't tell the model what to do. Tell it what it can't do, who it's talking to, and what format you need. The tighter the box, the more creative the output. Think: 'You are a skeptical editor reviewing a draft for a founder who hates fluff' vs. 'Write me a good email.'
First Draft Is Recon
Never expect the first output to be the answer. Use it to see how the model interpreted your intent, then refine. The first draft reveals what you actually meant to ask. Copy what worked, delete what didn't, tighten the constraint, run it again.
Build Your Prompt Library
Every good prompt you write is an asset. Save it. Tag it. Version it. The difference between someone who 'uses AI' and someone who builds with it is a library of tested, reusable prompts tuned to their voice and workflow.
Break Things On Purpose
Push models past their comfort zone. Ask for contradictions. Give impossible constraints. Feed it your worst writing and ask it to diagnose the problem. The edge cases teach you more about what the model actually understands than any tutorial.
Trust Your Vibe
If the output feels off, it is off — even if you can't articulate why. Your taste is the filter. AI gives you volume and speed; you provide judgment and direction. The model doesn't know what's good. You do.
After this section
You'll have copy-paste prompts for thinking, creating, and organizing — each one built to produce output you'd actually use.
Field Notes
You've got the tools and the prompts. Now let's learn the language underneath them.