Most people use AI the same way: ask a question, get an answer. The prompts on this page work differently — they flip the interview, point the AI at your own data exhaust, and set traps for your blind spots. Every one is copy-paste ready, and every one ends with a way to make it yours. Steal freely.
One idea runs through all of them: the best gift you can give an AI is context — and Dex means you're sitting on months of it.
Flip the script — make Dex interview you
You're bad at guessing what the AI needs to know about you. It's very good at pulling it out of you. So stop writing profiles and let it ask.
I want you to interview me so you can work brilliantly with me. Ask me one question at a time — never a list — and keep going until you have enough. Cover: what I actually do day to day (not my job title), the goals I'm measured on, who my customers and stakeholders are, how I like to work and communicate, what a great week looks like, and what keeps going wrong. When you're done, update my profile and my USER_EXTENSIONS block with what you learned, and show me exactly what you changed.
Make it your own: answer out loud using voice dictation — you'll say twice as much, and the messy honesty is the good part.
You're my career coach. You have my role, my goals, and months of my daily plans and reviews in this vault — use them. Ask me about my week: what went well, what didn't, what I'm avoiding. One question at a time. Push past my first answer — it's usually the polished one, not the true one. At the end, tell me one thing to do differently next week and save anything worth keeping as career evidence.
Make it your own: run it every Friday at the same time. Ten minutes. By review season, /career-coach and /resume-builder will be working from receipts, not memory.
I have a rough idea: [ONE SENTENCE]. Don't develop it yet. Act as a sharp product manager and grill me with clarifying questions, one at a time, until the idea is properly thought through. When I say "done", switch roles: become a skeptical CFO and grill me again on cost and payoff. Then — and only then — write the one-pager.
Make it your own: swap the personas for whoever actually scares you: your board, your best customer, your most pedantic engineer.
Mine your exhaust — Dex knows things about you that you don't
Dex accumulates machine-readable traces of how you actually work — usage logs, session learnings, archived daily plans, meeting notes. Most people never look at them. These prompts turn that exhaust into self-knowledge.
Read my usage log (System/usage_log.md), my session learnings (System/Session_Learnings/), and my archived daily plans (07-Archives/Plans/). Don't summarize them — infer from them. What are my actual working patterns? When do I plan versus drift? What do I consistently overcommit on? What do I say matters but never act on? Give it to me straight, with the evidence for each claim.
Make it your own: ask it to write the findings into your identity model so future advice is calibrated to the real you, not the aspirational one. (/identity-snapshot runs a structured version of this regularly.)
Look at my last eight weeks of meetings. Ignore what the events are called — work out what my calendar says I actually value: who gets my time, what share is reactive versus chosen, where the recurring commitments are. Then compare against my pillars in System/pillars.yaml. Where does my time contradict my stated priorities? What one standing commitment should I renegotiate?
Make it your own: run it on the first Monday of each month. The gap between stated and revealed priorities is the single most useful chart of your working life.
Go through my meeting notes from the last month. Find every commitment I made — including soft ones like "I'll take a look" or "let me get back to you" — and check each against my tasks and what's actually been done. Show me the open promises, oldest first, with who I made them to.
Make it your own: brutal before a 1:1 or a customer call. Point it at one person — "just promises involving Sarah" — for surgical pre-meeting honesty.
Here's what running it feels like:
Read everything you have on [PERSON] — meetings, notes, tasks between us. How is this relationship trending? What do they care about that I keep not engaging with? What have I promised and not delivered? If they were asked privately "how's working with me?", what would they honestly say?
Make it your own: run it for your manager, your key customer, and the stakeholder you find difficult — the third one is why this prompt exists.
Magic words — voice in, finished work out
Set a trigger word, then think out loud in any order. Nothing happens until you say the word — then the mess becomes the deliverable.
I'm going to dump thoughts at you in no particular order — voice notes, fragments, tangents. Don't respond to any of it beyond "noted". When I say HARVEST, turn everything I've said into a [board update / project brief / speaker notes], well-structured, in my voice, ready to use. File anything that sounds like a task or a commitment where it belongs in my vault.
Make it your own: the last sentence is the Dex upgrade — the leftovers get filed, not lost. Great on a walk after a big meeting: dump the debrief, HARVEST, done.
Set traps for your blind spots
The most valuable prompts are the uncomfortable ones. You have months of evidence in the vault; let it argue with you.
It's three months from now and [PROJECT] has failed. Using everything you know about this project, my team, and how I work — including my mistake patterns in 06-Resources/Learnings/ — write the honest post-mortem. What killed it? What early warning did I ignore? Which of my known patterns contributed? Then: what's the cheapest thing I could do this week to make that story not come true?
Look through my tasks and daily plans for the things I keep rescheduling — anything that has slipped three or more times. For each one: how long have I been carrying it, and what does the pattern suggest — is it genuinely low priority (kill it), too big (break it down), or am I avoiding it (say why)? Be direct. End by asking me which one I'll do tomorrow morning, first.
Make it your own: let Dex hold you to the answer — "put it in tomorrow's plan as the first Must."
Here's what I'm about to do: [DECISION]. You have my goals, my history, and my mistake patterns. Make the strongest honest case AGAINST it — not strawmen, the version my smartest critic would make. Then tell me what evidence would change your mind, and what you'd do instead.
Ideate from your tools, not your problems
Don't start from "what's my problem?" Start from "here's what my system can reach — what could we build?" It surfaces ideas you'd never think to ask for.
List every tool and connection you currently have access to in this vault — integrations, MCP tools, the works. Then, knowing my role and my goals: what are the five most valuable workflows we could build that I'm not using? For each: what it would do, what it needs, and the effort to set up. Rank by payoff. Don't pitch me things my tools can't actually do.
Make it your own: when one lands, follow with /create-skill — the workflow becomes a command you run forever. This start-from-the-tools move is how the best internal AI hackathons run.
Where these come from — and where yours go
These patterns come from real workshops and real daily use, and they all reduce to the same loop:
The prompt-stealing loop
-
01 Steal a promptcopy one from this page, run it as-is
-
02 Make it yoursswap the personas, folders, and stakes for your own
-
03 Notice it workingthe moment you run it twice, it's a workflow
-
04 Make it permanent/create-skill turns it into a command — yours forever
-
05 Share it/diff-generate packages it for someone else to steal
Don't just read this — paste it into your own Dex and it becomes a lesson about your actual work.
First, fetch and read the page I'm referring to: https://heydex.ai/help/prompts-to-steal.md I just read the Dex Guide's "Prompts to steal" page — flipped interviews, exhaust-mining, blind-spot traps, magic words. Now go one better: invent THREE new prompts in that spirit that only work for me — built on my role, my actual tools, my vault's history, and my known patterns. For each: the prompt itself, ready to run, and one line on why it suits me specifically. Then run the best one, live, right now.