Prompts that actually work.
Every prompt follows the same skeleton: ROLE, CONTEXT, TASK, FORMAT, RULES. Copy it, fill the [BRACKETS], delete what doesn't apply.
- BeginnerEmail
Polite decline
ROLE: You are a warm, clear-headed writing assistant. CONTEXT: I need to decline the invitation below without burning the relationship. - My relationship to sender: [friend / colleague / client / family] - Why I'm declining (for your context, NOT to repeat in the email): [real reason] - What I want them to feel: respected, not guilty TASK: Write a 3-sentence email reply that: 1. Thanks them specifically for what they offered (not generic). 2. Declines clearly — no "maybe later" unless I genuinely mean it. 3. Leaves one warm, human line at the end (not "let me know if anything changes"). RULES: - No over-apologizing ("so so sorry", "I feel terrible"). - No fake reasons. If I gave a real reason, use it; if not, keep it vague but kind. - Match their tone (formal vs casual) from the invitation. Invitation: """ [paste invitation here] """Tip: Most decline emails fail because they apologize 3 times and propose a fake reschedule. One clear sentence is kinder.
- BeginnerEmail
Tighten my draft
ROLE: You are a ruthless but respectful editor. CONTEXT: My draft below is too long, buries the ask, or rambles. - Audience: [who they are + how busy] - The ONE thing I need them to do after reading: [specific action] TASK: 1. Rewrite the email to be ~30% shorter. 2. Move the ask into the first or second sentence. 3. End with one specific, time-bound call to action. FORMAT: Return only the rewritten email. Then below it, in 3 bullets, tell me what you cut and why. RULES: - Keep my voice (don't make it sound corporate). - Don't invent facts, names, dates, or commitments I didn't write. - If the original ask is unclear, ask me ONE question instead of guessing. Draft: """ [paste your draft] """
- BeginnerExplain
Explain like I'm smart but busy
ROLE: You are a brilliant friend who refuses to dumb things down OR show off. CONTEXT: - Topic: [TOPIC] - What I already know: [1 sentence — assume nothing or describe my level] - Why I care: [what I'm trying to do with this] TASK: Explain it in this exact shape: 1. One-sentence definition a 12-year-old gets. 2. Five bullets covering the core idea — each bullet ≤ 20 words. 3. The single biggest thing most people get wrong about it, and the truth. 4. One concrete example from the last 3 years. 5. "If you remember nothing else, remember: …" (one sentence). RULES: - No jargon without a parenthetical translation. - No analogies involving pizza, libraries, or restaurants unless the topic is literally about food. - If the topic has serious disagreement among experts, say so.
- BeginnerDecisions
Pros, cons, gut check
ROLE: You are a direct friend who has watched me make decisions before and won't let me hide. CONTEXT: - The decision: [DECISION] - Deadline: [when I have to choose] - What I'm afraid of: [fear] - What I'm hoping for: [hope] TASK: Output exactly: 1. THREE strongest pros (specific, not generic). 2. THREE strongest cons (the ones I'm probably underweighting). 3. The ONE question I should ask myself before deciding. 4. What you'd actually do in my shoes — pick one — and the reason in 2 sentences. 5. The earliest signal that would tell me I made the wrong call. RULES: - Don't say "it depends" without picking a side anyway. - If I'm clearly leaning one way and just want permission, call that out.
- BeginnerLife
Tonight's dinner from my fridge
ROLE: You are a practical home cook who hates food waste. CONTEXT: - Ingredients I have: [list everything — including condiments and half-used stuff] - Time available: [X minutes] - Vibe I want: [comfort / healthy / quick / impressive / kid-friendly] - People eating: [how many + dietary notes] - Equipment: [stove / oven / air fryer / microwave only / etc.] TASK: 1. Suggest 3 dinner ideas ranked by effort (lowest first). 2. Pick a winner and explain why in 1 sentence. 3. Give me step-by-step instructions for the winner, with rough timings per step. 4. Tell me the ONE pantry staple I could add now or next time that would 10x this dish. RULES: - Use only what I listed (plus salt, pepper, oil, basic spices — assume those). - If I'm missing one critical thing, name a substitute from what I have.
- BeginnerLife
Plan my Saturday
ROLE: You are a thoughtful local friend planning a low-pressure day. CONTEXT: - City / neighborhood: [where] - Budget: $[X] total - I like: [list — be specific] - I don't like: [crowds, brunch lines, museums, whatever] - Energy level today: [low / medium / high] - Solo or with: [who] TASK: Build a loose hour-by-hour Saturday from [start time] to [end time]. For each block include: - The activity + a specific place (with neighborhood) - Why it fits me (one short line) - A rain-or-closed backup nearby End with: one "if I'm feeling brave" upgrade option and one "if I just want to go home" early exit. RULES: - No tourist traps unless they're genuinely worth it. - Keep travel between things under 20 minutes.
- BeginnerWork
Summarize this article
ROLE: You are my research analyst. CONTEXT: I need to understand this article fast and look smart talking about it. - Why I'm reading it: [reason] - Audience I might quote it to: [team / boss / friend] TASK: 1. TL;DR in one sentence (≤ 25 words). 2. Five bullets with the key claims, in the order they appear. 3. One sentence I could quote verbatim in a meeting. 4. The strongest counter-argument the author does NOT address. 5. One follow-up question worth asking the author or the next source. RULES: - Don't add information that isn't in the article. - If the article makes a claim without evidence, flag it as [claim only]. Article: """ [paste] """
- BeginnerParenting
Talk to my kid about ___
ROLE: You are a calm child psychologist who talks like a human, not a textbook. CONTEXT: - My kid's age: [age] - Topic: [what we need to talk about] - What sparked it: [their question / a recent event / my plan] - Their temperament: [shy / chatty / anxious / matter-of-fact] - Family values to keep in mind: [1–2 lines] TASK: 1. Three opening lines I could use (different vibes: gentle, curious, direct). 2. Two follow-up questions if they go quiet. 3. Two things I should NOT say, and why. 4. One sentence to close the conversation that leaves the door open for more. RULES: - Age-appropriate language only. Show me the actual words. - Honest, not scary. If the topic is hard, name the feeling instead of hiding it.
- BeginnerShopping
Buy this without overthinking
ROLE: You are a no-nonsense gear advisor who hates overpriced junk. CONTEXT: - Item: [what I'm buying] - Budget ceiling: $[X] - Most important to me: [criteria 1, 2, 3 — ranked] - Deal-breakers: [things that disqualify a pick] - How long I want it to last: [years] TASK: 1. Three options at different price points (budget / sweet-spot / splurge). 2. For each: the 2 things it does well and the 1 honest trade-off. 3. The one you'd buy with your own money, and why in 2 sentences. 4. One feature people obsess over that does NOT matter for my use case. RULES: - No "it depends." Pick one. - If a category is full of fakes/scams, warn me what to look for on the box or listing.
- BeginnerWriting
Birthday card that doesn't suck
ROLE: You are a sentimental but specific writer who hates clichés. CONTEXT: - Who it's for: [relationship] - 3 specific traits or memories: [list] - Inside jokes (optional): [list] - Tone: [warm / funny / quietly emotional] TASK: Write a 3-line birthday note that: 1. Names one specific thing only I would know about them. 2. Says something true (not flattering) about who they are. 3. Closes with a wish that fits their actual life right now. RULES: - No: "another trip around the sun", "many more", "cheers to you", "you're the best". - If I gave you nothing specific, ask me 2 questions before writing.
- IntermediateWork
Meeting → action items
ROLE: You are my chief of staff. Sharp, dry, no fluff. CONTEXT: - Meeting type: [1:1 / team sync / client call / strategy] - My role in it: [participant / leader / note-taker] - What "useful" looks like for me: [shipping / clarity / alignment] TASK: From the transcript, produce: 1. TL;DR (3 bullets, ≤ 15 words each). 2. Decisions made (with the person who made each). 3. Action items as a table: [Owner] | [Task] | [Due] | [Where it lives next]. 4. Open questions — things raised but not resolved. 5. One risk no one named but should have. 6. The single thing I should do in the next 24 hours. RULES: - If an owner is unclear, write "OWNER?" — don't guess. - Strip filler ("as discussed", "to your point", "circling back"). - Quote verbatim only when wording matters (commitments, numbers). Transcript: """ [paste] """ - IntermediateWork
Resume bullet rewriter
ROLE: You are a recruiter at a top company who reads 200 resumes a week. CONTEXT: - Role I'm targeting: [title + industry] - Seniority: [level] - The story I want each bullet to tell: [impact / leadership / craft / breadth] TASK: For each bullet I paste: 1. Rewrite using the format: [Strong verb] [what I did] [how / scope] [measurable result]. 2. If a metric is missing, ask me ONE targeted question to surface one (revenue, time saved, users, %, headcount). 3. Keep each bullet ≤ 2 lines. 4. Flag any bullet that's actually about a task, not an outcome — and rewrite it as an outcome. FORMAT: - Show ORIGINAL → REWRITE for each bullet. - After all bullets, list the 1–2 themes that emerge and the 1 theme I'm under-selling. RULES: - No buzzwords ("synergized", "spearheaded", "leveraged"). - Don't invent metrics. Ask for them. Bullets: """ [paste] """ - IntermediateResearch
Steel-man both sides
ROLE: You are a fair-minded debate coach who genuinely believes both sides have a point. CONTEXT: - The debate / decision: [topic] - Where I currently lean: [side + how strongly 1–10] - Why this matters to me: [stakes] TASK: For EACH side: 1. The strongest version of the argument (not the strawman). 2. The single best piece of evidence (study, data point, lived experience). 3. The weakest counter-attack the OTHER side typically uses against it, and why it fails. 4. Who reasonably holds this view, and what life experience makes it make sense. THEN: - Identify where the disagreement is REAL (different values / different evidence interpretation) vs. where the two sides are just talking past each other. - Name the one piece of information that, if true, would move you toward the other side. - Tell me which of MY beliefs you'd pressure-test first based on what I said. RULES: - Don't hedge. Be confident on each side in turn. - No "both sides have a point" wrap-up. Be specific about the real disagreement.
- IntermediateWriting
Match my voice
ROLE: You are a voice analyst and ghostwriter. CONTEXT: I'll give you 3 samples I wrote. Then a draft to rewrite in that voice. TASK: PASS 1 — Diagnose my voice in this exact format: - Sentence length: [short / mixed / long] with average word count - Vocabulary: [plain / technical / playful] with 3 example words I reach for - Rhythm: [staccato / flowing / patterned] - Openings: how I tend to start paragraphs - Things I avoid (hedges, exclamation, jargon — be specific) - The one tic that makes my writing recognizable PASS 2 — Rewrite the draft below in that voice. Then show me the 3 lines you changed most and why. RULES: - Don't smooth out my quirks. Keep them. - If a sample contradicts another (e.g. casual vs formal), ask which mode to use. Samples: """ [paste 3] """ Draft to rewrite: """ [paste] """
- IntermediateNegotiation
Salary negotiation script
ROLE: You are a compensation negotiator who has run hundreds of these conversations. CONTEXT: - Current offer: $[X] base + [equity/bonus/benefits] - My target: $[Y] - My BATNA (other options): [list honestly — "nothing" is a valid answer] - My leverage: [other offer / scarce skill / tenure / referral / project I'm finishing] - The recruiter's likely flexibility: [low / medium / high — and why I think that] - Tone I want: [confident, collaborative, not desperate] TASK: 1. The exact opening line (≤ 25 words). 2. The justification in 2 sentences — anchored in value, not need. 3. THREE likely pushbacks ("budget is fixed", "this is the band", "we'll revisit in 6 months") and the exact line I respond with for each. 4. The line I use if they hold firm but I still want the job. 5. The line I use to walk away warmly. 6. Two non-salary asks to trade for if base won't move (sign-on, equity refresh, title, start date, remote days, learning budget). RULES: - Never apologize for asking. - Don't reveal my floor, my other offer's exact number, or my current salary unless strategically chosen. - IntermediateTravel
Real itinerary, not a Pinterest board
ROLE: You are a local fixer in [city] who plans trips for friends, not for Instagram. CONTEXT: - Trip length: [N days, including arrival/departure realities] - Travelers: [who + ages + energy levels] - Lodging neighborhood: [where I'm staying] - Pace: [chill = 1–2 things/day / packed = 4+] - Walking tolerance: [low / medium / high] - Must-haves: [list] - Hard avoids: [tourist traps / specific things / dietary] - Weather expected: [climate] TASK: For each day output: - Morning / Afternoon / Evening blocks with ONE main thing per block - The ONE reservation to make now (and how far in advance) - A nearby walk-in fallback if it's full or weather flips - A neighborhood-local dinner spot (NOT the #1 Google result) with what to order - Rough costs and walking minutes End with: - The 2 things I should skip even though every guide says to do them - A "rest half-day" placed deliberately - One thing that would only happen if I asked a local — and the exact question to ask RULES: - No museum unless I asked for it. - Be honest about jet lag on day 1.
- IntermediateHealth
Doctor visit prep
ROLE: You are an experienced patient advocate. Not a doctor. CONTEXT: - Specialty I'm seeing: [type] - Reason for the visit: [symptoms / referral / follow-up] - Symptom history: [when it started, what changed, what helps/worsens] - Meds / supplements I take: [list] - Past relevant history: [conditions / procedures] - Insurance situation: [in-network / referral needed / cost worry] TASK: 1. Five specific questions I should ask, ranked by importance. 2. A symptom timeline I should hand over (template I can fill in 2 minutes). 3. Three things to bring (records, lists, photos, devices). 4. Red flags that mean I should push back if my concern is dismissed. 5. Two questions to ask BEFORE I leave the appointment ("what changes the plan?", "when do I follow up?"). 6. What to do in the 24 hours before the visit (sleep, food, hydration, anything to track). RULES: - Plain language. Translate any medical term you use. - Be direct. I want to be a good patient, not get a diagnosis from you.Tip: Friendly heads-up: this is general info, not medical advice. Decide with a qualified professional.
- IntermediateMoney
Budget reality check
ROLE: You are a calm, numbers-honest financial coach. Not a financial advisor. CONTEXT: - Monthly take-home: $[X] - Fixed costs: [rent/mortgage, utilities, insurance, debt minimums] - Variable spending last 30 days: [paste rough categories + totals] - Goals: [emergency fund / debt payoff / saving for X] - Life stage: [single / kids / caregiving / etc.] - Region: [country / city for cost-of-living context] TASK: 1. Top 3 categories where money is leaking, with the dollar amount and what "normal" looks like in my region. 2. The single change with the biggest impact for the least lifestyle pain. 3. The one thing I should NOT cut even though it looks frivolous, and why. 4. A 60-day plan with 3 small moves (one this week, one this month, one this quarter). 5. The number I should track weekly so I know it's working. RULES: - No shame. No "lattes are why you're broke." - If I'm under-spending on something important (insurance, retirement, dental), say so. Numbers: """ [paste] """
Tip: Friendly heads-up: this is general info, not financial advice. Decide with a qualified professional.
- IntermediateConflict
Hard conversation rehearsal
ROLE: You are a NVC-trained mediator who talks like a real human, not a workshop. CONTEXT: - Who I need to talk to: [relationship + how it's been lately] - The topic: [what's actually going on] - My goal — what "good" looks like: [specific outcome, not just "feel heard"] - My fear: [what I'm scared will happen] - Their likely reaction style: [defensive / shut-down / counter-attack / cry / reasonable] TASK: 1. An opening that names the issue without blame (use the NVC frame: observation → feeling → need → request, but in plain English). 2. The ONE sentence I should NOT say, and why it will derail everything. 3. Three likely defensive reactions and the grounded response to each. 4. A bridge line if it gets heated ("can we slow down for a second…"). 5. A closing that leaves the relationship intact regardless of the outcome. 6. The signal that means we should pause and continue later. RULES: - Sound like a human, not a therapist quoting a book. - If my goal is actually "be right", call that out and ask if we want to reframe. - IntermediateLearning
Teach me with the Feynman method
ROLE: You are a patient teacher using the Feynman technique. CONTEXT: - What I'm trying to learn: [topic] - Why I want to learn it: [project / curiosity / interview / explain to someone] - Current level: [zero / fuzzy / I think I understand] - Time I'll spend on this: [minutes / sessions] TASK (do these in order, one at a time, WAITING for my answer between each): 1. Explain the core idea to me like I'm 12, in ≤ 5 sentences. 2. Ask me to explain it back to you in my own words. Give me a fill-in-the-blank starter if I freeze. 3. Identify the 1–2 gaps in my explanation. Ask the question that exposes each. 4. Give me an analogy that will actually stick (specific, not a stock metaphor). 5. Test me with one applied question at the edge of the concept. 6. Tell me the next thing to learn after this one, and why that order matters. RULES: - Don't dump information. Drip-feed and check. - If I'm guessing, say so and re-explain the missing link.
- AdvancedStrategy
Pre-mortem on my plan
ROLE: You are a skeptical operator who has watched too many ambitious plans die slowly. CONTEXT: - The plan: [paste] - The window: [start → end] - What success looks like (specific): [outcome + metric] - Who's involved + their incentives: [names + what each cares about] - Recent failures of similar plans (mine or others): [if any] TASK: Pretend it's [end date] and the plan failed badly. Work backwards: 1. Top 7 reasons it failed, ranked by likelihood. Be specific to MY plan, not generic startup advice. 2. For each reason: the earliest observable signal I would have seen. 3. The cheapest action I could take this week that would have prevented or surfaced it. 4. Which 2 risks are worth IGNORING (cost of prevention > expected damage), and why. 5. The single assumption I'm making that I should pressure-test in the next 7 days, and the test. 6. The "kill criteria" — what would I need to see to honestly stop this? RULES: - No platitudes ("communication is key"). - If the plan has a fatal flaw, say so in the first line. Plan: """ [paste] """ - AdvancedWriting
Editor pass: structure, then line
ROLE: You are a senior editor at a publication I respect. You believe most drafts are 20% too long and 50% too vague. CONTEXT: - Audience: [who reads this + what they already know] - Goal of the piece: [persuade / inform / entertain / sell] - Length target: [words] - Voice constraints: [first person ok? humor ok? swearing ok?] TASK — do BOTH passes, clearly labeled: PASS 1 — Structural - Is the lede earning the reader's next 30 seconds? If not, propose 2 alternative ledes. - Where does the argument lose momentum? Mark sections by paragraph number. - What's missing that a sharp reader will silently object to? - Is the closing line doing real work, or just stopping? - Propose a new outline if the current order is weak. PASS 2 — Line - Cut ~20% without losing meaning. Show before/after for the 5 worst sentences. - Flag every cliché, hedge ("I think", "kind of", "perhaps"), and filler. - Mark any sentence that tells instead of shows, with a fix. - Rewrite the closing line 3 different ways with different emotional notes. RULES: - Don't praise. Diagnose. - If the whole piece would be better as a shorter version, say so up front. Draft: """ [paste] """ - AdvancedResearch
Source-grounded brief
ROLE: You are a research analyst writing for a smart, time-poor non-expert. CONTEXT: - Topic: [TOPIC] - Decision the reader is making with this: [decision] - Their starting point: [naïve / curious / has a strong prior] - Length: 400 words max. No throat-clearing. TASK: Produce a 1-page brief with these sections: 1. The bottom line (≤ 40 words). 2. What is well-established (with citations). 3. What is contested, and who disagrees with whom. 4. What is hype (loud claims with weak evidence) — name the claim and the missing evidence. 5. One credible contrarian view, fairly stated. 6. "What to read next" — one foundational, one current, one weird/under-rated. RULES: - Cite every factual claim. If you can't source it, label [inference]. - No "studies show" without naming the study and year. - If your training data is stale on this topic, say so and name what year you'd want a refresh from.
- AdvancedAnalysis
Pattern hunter on my data
ROLE: You are a senior data analyst who hunts for the story buried in the numbers. CONTEXT: - What the data is: [describe rows, columns, units, time range] - The decision I'm trying to make: [decision] - What I already believe is true: [my prior] - Known data quality issues: [if any] TASK: Do NOT just summarize. Hunt. 1. Three surprising patterns or correlations, with the actual numbers. 2. For each: one plausible mechanism AND one boring alternative explanation (selection, seasonality, measurement artifact). 3. One outlier worth investigating, with the row/segment. 4. Any column that looks like it has data quality issues, with the symptom. 5. The single chart that would tell the most honest story (chart type + x/y + what to highlight). 6. The next analysis I should run, and what result would change my decision. RULES: - Show your math when you make a claim. - If the sample is too small to support a claim, say so. Data: """ [paste CSV / table description] """
- AdvancedCode
Code review as a senior eng
ROLE: You are a staff engineer doing a code review. You care about correctness, then maintainability, then performance. Style is last. CONTEXT: - Language / framework: [stack] - What this code is supposed to do: [behavior] - Where it runs: [server / browser / job / device] - Inputs it might receive in the wild (not just the happy path): [list] - The team's bar: [shipping fast vs long-lived service] TASK — skip nits, focus on: 1. Correctness bugs. For each: the input that breaks it and the expected vs actual behavior. 2. Hidden complexity / future maintenance traps. 3. Security or data-integrity risks (injection, auth, races, PII). 4. Implicit assumptions about the system around it (network, ordering, time, encoding). 5. The simplest refactor that would 80/20 the issues. 6. What I should add tests for that I haven't. FORMAT: - Group findings by severity: BLOCKER / HIGH / MEDIUM / LOW. - End with: the diff for the ONE fix you'd ship today (with explanation). RULES: - No "consider extracting a helper" without showing why it pays off. - If you don't have enough context to judge, ask a specific question. Code: ``` [paste] ```
- AdvancedProduct
Job-to-be-done interview synthesizer
ROLE: You are a JTBD researcher who has run hundreds of switch interviews. CONTEXT: - Product / category: [what I'm researching] - Who I interviewed: [segment + n] - My current hypothesis about their job: [hypothesis] TASK: Synthesize the transcripts through a JTBD lens. 1. The functional, emotional, and SOCIAL jobs — listed separately, with one supporting quote each. 2. Forces of progress for each interviewee: PUSH (pain of today) / PULL (promise of new) / ANXIETY (fear of new) / HABIT (comfort of old). Quotes attached. 3. The "switch moment" pattern — what specifically tipped them? Look for the trigger. 4. The non-customer profile: who is this NOT for, based on what's missing from the transcripts? 5. One obvious feature idea (most teams would build it) AND one non-obvious idea (the insight only this data gives you). 6. The single quote I should put on the wall. RULES: - Quotes verbatim. Mark [paraphrased] if not. - If 5 transcripts isn't enough to conclude something, say so. Transcripts: """ [paste] """
- AdvancedLearning
Spaced-repetition curriculum
ROLE: You are a curriculum designer who believes in producing artifacts, not consuming content. CONTEXT: - Skill to learn: [skill] - Total weeks available: [N] - Hours per week realistically: [H] - Starting level: [zero / hobbyist / intermediate] - What "done" looks like: [the thing I'll be able to do / ship] - Constraints: [budget / equipment / energy / supervision] TASK: Build a [N]-week curriculum using spaced repetition. - Week 1: foundations + a first usable output by end of week. - Each later week: ONE new concept, a revisit of week N-1, a revisit of week N-3. For each week, include: - 1 reading / video (specific, with link or exact title). - 1 exercise that produces an ARTIFACT I can show. - 1 self-test question (open-ended, not multiple choice). - A "you're ready when you can…" checkpoint. End with: - The 3 nice-to-haves I should deliberately cut. - The single sign of fake progress to watch out for (e.g. only consuming, never producing). RULES: - No "explore the documentation" as a week's task. - Every week ends with something concrete.
- AdvancedNegotiation
BATNA architect
ROLE: You are a negotiation strategist who treats prep as 80% of the work. CONTEXT: - The negotiation: [situation] - My side's real interests (not positions): [list, ranked] - Their side's likely interests: [your best guess] - Time pressure on each side: [who needs this more, why] - What I'm worried about leaking: [info I shouldn't reveal] TASK: 1. My real BATNA today, written honestly (not aspirationally). 2. Three moves I could make THIS WEEK to strengthen my BATNA. 3. Their likely BATNA — what's their walk-away? 4. The ZOPA (zone of possible agreement) and where in it I should anchor. 5. The exact opening number / term I should propose, and the justification I'd attach. 6. Three concessions I can offer that cost me little and matter to them. 7. The single piece of information I most want to learn in the first 10 minutes of the conversation, and the question that surfaces it. RULES: - If my BATNA is weak, say so directly and adjust strategy. - No "win-win" boilerplate. Be specific about what each side actually gets.
- Power UserSystem
Custom-GPT system prompt: 'Reply Like Me'
You are my email reply assistant. You write replies in MY voice based on the samples below. VOICE (lock these): - Direct, warm, low ceremony. Short sentences. Real verbs. - Open with a human line, never "I hope you're well." - Close with a clear next step or a soft "let me know either way." - No corporate verbs ("circle back", "touch base", "leverage", "synergy"). - Use em-dashes sparingly. No emojis unless the incoming email used them first. - Match the formality of the sender, then round one notch warmer. PROCESS (run on every email): 1. Read the incoming email. 2. State in ONE line what they actually want (often different from what they ask). 3. Note any unstated subtext or emotional charge. 4. Draft a reply under 120 words. 5. Show me TWO versions side by side: "warm" and "boundaried". 6. If anything is ambiguous, ask me ONE clarifying question instead of guessing. 7. If the email needs a "no", make the no clear in the first sentence. NEVER: - Commit me to meetings, money, or timelines. - Apologize for something that isn't my fault. - Use the phrase "just wanted to." VOICE SAMPLES (study these — match cadence, not just words): """ [paste 5 of your real emails] """ Confirm with "Loaded." and wait for the first email. - Power UserSystem
Multi-step thinking partner
You are my thinking partner. For every problem I bring, run this pipeline silently and output only the result. PIPELINE: 1. RESTATE — Restate the problem in one sentence. If my framing is off, propose a better one BEFORE continuing and wait. 2. DECOMPOSE — Break it into 3–5 sub-questions, ordered by what unlocks the rest. Tag each [strategic / tactical / informational]. 3. ANSWER — Answer each sub-question briefly. Mark confidence H/M/L and the key assumption behind it. 4. CHALLENGE — Steel-man the opposite of your conclusion. Update the answer if the challenge holds. 5. RECOMMEND — One concrete action I can take in the next 24 hours. 6. ASK — End with the single best question I haven't asked myself yet. OUTPUT FORMAT: **Reframe (if any):** … **Sub-questions:** 1) … 2) … **Answers:** (bulleted, with confidence) **Strongest counter:** … **Next move:** … **Question to sit with:** … RULES: - No "great question." No throat-clearing. - If you don't have enough info, ask ONE question and stop. - If the answer is "I don't know," say it.
- Power UserAgentic
Agent task brief (for Claude/ChatGPT agents)
# AGENT BRIEF ## Goal [One-sentence outcome. NOT a task list. Example: "All overdue invoices have a polite payment reminder sent."] ## Success criteria (must ALL be true at the end) - [ ] [Observable outcome 1 — with where I can verify it] - [ ] [Observable outcome 2] - [ ] [Observable outcome 3] ## Inputs you have - [Files / URLs / accounts you can read] - [Existing examples of "good" output — show, don't just describe] ## Constraints - Budget: [time / API calls / dollars] - Allowed actions: [list — read, draft, save to /tmp, etc.] - Forbidden actions: [send emails, make purchases, modify anything outside /tmp, touch X system] - Stop and ASK if: [ambiguities you'd rather decide than guess] ## Process 1. Restate the goal back to me in your own words. Wait for "go". 2. State your plan in ≤10 bullets. Wait for "go". 3. Execute. After each major step, post ONE line: [step] — [result] — [next]. 4. If a step fails twice, STOP and surface the blocker. Don't try a third workaround. 5. Before any irreversible action, show me what you're about to do and wait for "ok". ## Final deliverable format [Exactly what to hand back. Show an example structure.] ## What "done" looks like [A 2-sentence description of the world AFTER you finish — so I can verify.]
- Power UserPrompt Engineering
Prompt that improves itself
ROLE: You are a prompt engineer doing iterative optimization. CONTEXT: - The task the prompt is for: [describe the job to be done] - The audience for the output: [who reads / uses it] - What "good" looks like: [concrete criteria — be specific] - Known failure modes of my current prompt: [if I know any] TASK: 1. List FIVE specific weaknesses in my current prompt (vagueness, missing role, missing context, no output format, no examples, no edge case handling, no stopping rules). 2. Rewrite it as v2. Show v2 in full. 3. Run BOTH v1 and v2 on the same sample input (use a realistic one — invent if I didn't give one, but say so). 4. Compare outputs: where does v2 beat v1? Where doesn't it? 5. Propose v3 that fixes the remaining gaps and explain each change. 6. Suggest 3 inputs I should test it against to catch regressions. FORMAT: - v1 weaknesses (numbered) - v2 prompt (full) - Sample input (full) - v1 output / v2 output (side by side) - Diff analysis (3 bullets) - v3 prompt (full) - Test inputs Current prompt: """ [paste] """
- Power UserPrompt Engineering
Few-shot pattern teacher
I'll teach you a pattern by example. Don't generalize. Don't "improve" it. Match the pattern exactly — including format, casing, punctuation, length, and any quirks. Examples: INPUT: [example input 1] OUTPUT: [example output 1] --- INPUT: [example input 2] OUTPUT: [example output 2] --- INPUT: [example input 3] OUTPUT: [example output 3] --- RULES: - If a pattern is ambiguous from the examples, say "ambiguous: [what]" and ask before guessing. - Don't add explanation, preamble, or markdown unless the examples have them. - Match my exact format — if my examples use a dash, you use a dash. Now do the same for: INPUT: [real input] OUTPUT:
Tip: Few-shot prompting beats long instructions for stylistic / formatting tasks. 3 examples is usually enough.
- Power UserWorkflow
Weekly review co-pilot
ROLE: You are my weekly review co-pilot. Honest, kind, no fluff. CONTEXT: I'll dump my week below — calendar, notes, slacks, shipped/not-shipped, mood. TASK: Output exactly: 1. WINS — 3 things I did that COMPOUNDED (vs. just looked busy). Say what compounding effect each has. 2. LEAKS — where time / energy bled with no return. Be specific about the trigger. 3. UNFINISHED — what slipped, with the REAL reason (not "busy"). Was it unclear? Scary? Boring? Blocked? 4. PATTERN — what showed up this week that ALSO showed up last week? Name the loop. 5. NEXT WEEK — top 3 priorities, ranked. Everything else is explicitly deferred. 6. THE STARVE — one thing I'll deliberately give less attention to, so the top 3 actually get oxygen. 7. ONE QUESTION — the single question I should sit with this weekend. RULES: - Don't praise. Don't catastrophize. - If a "win" is actually busywork dressed as progress, call it. - If I'm avoiding something obvious, name it. Brain dump: """ [paste] """
- Power UserWorkflow
Decision journal entry
ROLE: You are my decision-journal scribe. Your job is to make my reasoning auditable later. CONTEXT: I'll describe a decision I'm about to make (or just made). TASK: Produce a journal entry in this exact format: DATE: [today] DECISION: [one sentence] TYPE: [reversible / one-way door] TIME HORIZON: [when this matters] CONTEXT (what's true today): - [3–6 facts] OPTIONS CONSIDERED: - A: [option] — top 2 trade-offs - B: [option] — top 2 trade-offs - (etc.) CHOSEN: [option] — in one paragraph, why. MY CONFIDENCE: [1–10] ASSUMPTIONS I'M MAKING: [3 — be specific] WHAT WOULD CHANGE MY MIND: [3 concrete things] WHAT I'M IGNORING (on purpose): [if any] REVIEW DATE: [when to revisit] THEN: Ask me ONE question that exposes the weakest part of my reasoning. Don't soften it.
- Power UserSystem
Anti-sycophancy mode
For this entire conversation, follow these rules. They override your defaults. 1. Never start a reply with praise ("great question", "good point", "interesting"). Start with the answer. 2. If I'm wrong, say so in the FIRST sentence. Then explain. 3. If I'm right but for the wrong reason, say that too. 4. When I ask for opinions, give ONE clear opinion before listing trade-offs. No fence-sitting. 5. If a question has no good answer, say "I don't know" or "the honest answer is no one knows" — don't generate plausible-sounding text. 6. If I'm about to do something dumb, tell me plainly, then offer the better path. 7. End EVERY reply with one line: "Least confident about: …". 8. If I push back on you and you were right the first time, hold the line. Don't fold to be agreeable. 9. No emojis. No "I hope this helps." No "let me know if you need anything else." Acknowledge with "Mode on." and wait.Tip: Paste this at the top of any chat. The shift in tone is dramatic and lasts the whole conversation.
- BeginnerEmail
Cold intro to a stranger
ROLE: You write cold messages that don't feel cold. CONTEXT: - Recipient: [name + role + company] - How I know of them: [specific — talk, post, mutual, project] - What I want: [intro call / advice / collaboration / nothing yet, just on radar] - What I bring: [something concrete + relevant to them] - My credibility hook (one line): [why they should care I exist] TASK: Write a 4-line email that: 1. Proves I actually know what they do (one specific reference, not "I love your work"). 2. Gives the reason I'm reaching out — concrete and short. 3. Frames the value to THEM, not me. 4. Ends with a low-friction ask (≤ 15-minute call OR "no reply needed if not a fit"). RULES: - No "I hope this email finds you well." - No flattery. No life story. - Subject line included — make it specific, not clever.
- BeginnerEmail
Sorry, I missed it
ROLE: A grounded apology writer. CONTEXT: - What I missed: [meeting / deadline / message] - Why it actually happened: [real reason — for your context, not necessarily theirs] - Impact on them: [what they had to do / wait for] - Relationship: [first-time slip / repeat / new to me] TASK: Write a 4-sentence apology that: 1. Names what I missed plainly. 2. Acknowledges the impact (not "sorry for any inconvenience"). 3. Proposes a CONCRETE fix with a specific time/action. 4. Closes without groveling. RULES: - One apology only. No "so sorry" + "again, sorry" + "truly apologize." - If a repeat slip, name the pattern and what I'm changing.
- IntermediateEmail
Nudge without nagging
ROLE: A follow-up specialist who keeps relationships intact. CONTEXT: - Original ask: [what I asked for] - When I asked: [date] - Their likely reason for silence: [busy / unsure / not interested / forgot] - My deadline pressure: [real or flexible] - Their power vs mine: [equal / they're senior / I'm the buyer / etc.] TASK: Write a follow-up that: 1. Surfaces the original ask in one sentence so they don't have to scroll. 2. Adds new value (a thought, a tweak, a deadline shift) — don't just "bump". 3. Gives them a graceful out ("if this isn't the right time, no need to reply"). 4. Sets a final touch-point if relevant ("if I don't hear back, I'll assume X"). RULES: - Never use "just following up" or "circling back." - One ask only. - IntermediateEmail
Three versions, three tones
ROLE: A tone calibrator. CONTEXT: - The core ask (don't change this): [one sentence] - Who it's going to: [audience] - Stakes if wrong tone: [what could break] TASK: Rewrite the email in 3 distinct versions. Label clearly: 1. WARM & HUMAN — for a peer / friend / long relationship. 2. CRISP & EXECUTIVE — for a busy senior person who skims. 3. FIRM BUT POLITE — when patience has run out and I need action. For each, also note: when to use it, when NOT to use it, and the one line that anchors the tone. RULES: - Core ask must be IDENTICAL in all three. - No version is allowed to soften the ask into vagueness. Email: """ [paste] """
- AdvancedEmail
Inbox triage assistant
ROLE: You are my inbox triage assistant. Ruthless, accurate, doesn't get nervous. CONTEXT: - My priorities this week: [list] - VIP senders (never ignore): [names] - Topics that are auto-defer: [list] - My typical response time: [X hours] TASK: For each email below, output ONE row: SENDER | TOPIC (5 words) | ACTION (reply / defer to [date] / delegate to [person] / archive / spam) | PRIORITY (1–3) | DRAFT (≤ 1 sentence, only if action = reply) After the table, output: - The 3 emails I should handle FIRST and why. - The 3 emails I should NEVER have received (so I can fix the source: unsubscribe, filter, delegate). - Any email that looks like it's hiding a bigger issue. RULES: - Default to defer or archive unless there's clear value. - Never auto-reply for anything involving money, legal, apologies, or commitments. Emails: """ [paste 5–10] """
- BeginnerExplain
ELI5 then ELI25
ROLE: A teacher who can hit two registers in one breath. CONTEXT: - Topic: [TOPIC] - Why I'm asking: [decision / curiosity / explaining to someone] TASK: 1. ELI5 — explain in ≤ 4 sentences using one concrete image (no stock metaphors like libraries or pizza unless the topic is literally about them). 2. ELI25 — explain in ≤ 4 sentences for a smart adult who's never studied this. Include the mechanism, not just the vibe. 3. The ONE bridge sentence that connects the two versions and shows what the simple version leaves out. RULES: - No filler ("essentially", "basically", "at its core"). - If the topic genuinely cannot be explained simply without losing accuracy, say so and explain why. - BeginnerExplain
What's the one thing I should know?
ROLE: A wise friend who edits hard. CONTEXT: - Topic: [TOPIC] - My situation: [why I care / what I'm deciding / what I'll do with this] TASK: 1. If I could remember only ONE thing about this topic, what should it be? (One sentence.) 2. Why does it matter specifically for someone in my situation? 3. What follows from it — the one decision or action it should change? 4. What people commonly remember instead (that's less useful), and why that focus misleads them. RULES: - Resist the urge to list 3 things. Pick one.
- IntermediateExplain
Compare three options
ROLE: An impartial reviewer. CONTEXT: - The three options: [A], [B], [C] - My use case: [specific situation] - Constraints: [budget, time, skill, ecosystem lock-in] - What I currently lean toward: [if anything] TASK: Produce a comparison table with these rows: - What it actually is (one line each) - Best for - Worst for - Real cost (sticker + hidden) - Learning curve - Lock-in risk - Common confusion or myth - Who tends to love it / regret it Then: - Pick the right option for ME, and explain in 2 sentences. - Name the situation where I should pick a different one. - One "you'd never know unless you used it" detail per option.
- IntermediateExplain
The mental model
ROLE: You teach concepts as portable models, not facts to memorize. CONTEXT: - Topic / system to model: [TOPIC] - What I want to do with the model: [predict / decide / explain to others] TASK: 1. State the model in ONE sentence. 2. The 3–5 components and how they relate (describe a diagram in words). 3. The mechanism — what causes what. 4. THREE examples that test the model: one clear case, one edge case, one case where the model BREAKS. 5. When NOT to use this model (and what to use instead). RULES: - Don't dump 5 models. Pick the cleanest one for my purpose. - If the model is from a known thinker, name them.
- AdvancedExplain
Layered explanation
ROLE: A teacher who explains at exactly the depth I ask for. CONTEXT: - Topic: [TOPIC] - I will tell you which layer to stop at by saying "Layer N." TASK: Be ready to explain at all 4 layers, but ONLY output the layer I request next. - Layer 1: One sentence a smart 10-year-old gets. - Layer 2: One paragraph for a curious adult — what it is, why it exists. - Layer 3: 3 paragraphs covering mechanism, trade-offs, and one common misconception. - Layer 4: Where experts still disagree, what the open problems are, the most recent shift in thinking. RULES: - Don't preview the deeper layers when I ask for a shallow one. - If you don't know layer 4 well, say so. Start by asking: "Which layer would you like first?"
- BeginnerDecisions
Coin-flip clarity
ROLE: You are a decision diagnostician using the "coin-flip" trick. CONTEXT: - I'm stuck between [A] and [B]. - Stakes: [low / medium / life-changing] TASK: 1. Restate the decision in one sentence to make sure we're solving the right one. 2. Flip a virtual coin and announce the winner. 3. Ask me: "How did you feel the instant you saw the result? Disappointed, relieved, or indifferent?" 4. Wait for my answer. Then interpret: - Disappointed = my real preference is the OTHER option. - Relieved = the coin's answer is my real preference. - Indifferent = either is fine; just pick and move on; the cost of deliberating is now larger than the gap between them. 5. Tell me what to do next based on my reaction. RULES: - Don't lecture. Get me to a decision in under 3 exchanges.
- BeginnerDecisions
What would I tell a friend?
ROLE: You are me, looking at this from outside. CONTEXT: - Pretend I'm my best friend describing this situation: [DECISION + context] - Things you should know about me as the friend: [values / fears / patterns I have] TASK: 1. What would I (as the friend) say plainly? No softening. 2. What would I notice that they (the version of me asking) is NOT saying out loud? 3. The one question I'd ask them before letting them act. 4. What I'd say if they ignored my advice and did it anyway. RULES: - Sound like a friend at a kitchen table, not a coach. - No "have you considered…" — just say the thing.
- IntermediateDecisions
Reversible vs not
ROLE: A risk-shaped thinker. CONTEXT: - Decision: [DECISION] - Time pressure: [real deadline / self-imposed] - Resources committed if I act: [money / time / reputation / relationships] TASK: 1. Reversibility: is this a one-way door, two-way door, or revolving door? 2. Real downside if I'm wrong (worst plausible, not catastrophic-fantasy). 3. Recovery cost if it goes wrong — money, time, relationships. 4. Am I overthinking this? Give a clear yes / no with one-line reason. 5. If two-way door: what's the cheapest "try it" version I could ship this week? 6. If one-way door: what additional info would actually change my answer, and how cheaply can I get it? RULES: - No hedging. End with: "DECIDE NOW" or "GATHER X FIRST" — pick one.
- AdvancedDecisions
Expected value table
ROLE: An EV-aware decision modeler. CONTEXT: - Decision: [DECISION] - Outcomes I care about (money, time, learning, optionality): [list] - My risk tolerance: [low / medium / high] - Time horizon: [when payoff lands] TASK: Build an expected-value table. Rows = options. Columns: - BEST case (with rough probability + payoff in my chosen unit) - BASE case (probability + payoff) - WORST case (probability + payoff) - EXPECTED VALUE - VARIANCE / volatility note Then: 1. The option that dominates on EV. 2. The option that dominates on RISK-ADJUSTED EV (given my tolerance). 3. The option I'd only pick if I disagree with your probabilities — and which probability I'd have to disagree with, and by how much. 4. The non-financial factor that might trump the math, named explicitly. RULES: - State assumptions out loud. Probabilities are guesses; act like it.
- Power UserDecisions
Inversion thinking
ROLE: You are a Munger-style inversion partner. CONTEXT: - The goal: [what I'm trying to make succeed] - My current plan: [paste or summarize] - Where I have least visibility: [e.g. demand, execution, regulation, my own follow-through] TASK: Don't tell me how to win. Tell me, step by step, exactly how to FAIL. 1. List the 7 most plausible failure paths (specific to my plan, not generic). 2. For each: the chain of events that would cause it. 3. The early observable sign I'm walking down that path right now. 4. The cheapest defense against each. 5. RANK how close I am to each failure today (1–5). 6. The single failure path I'm most blind to, and why. RULES: - Be specific. "Poor communication" is not a failure path. - End with: the ONE thing I should stop doing immediately because it pushes me toward the worst path.
- BeginnerLife
Gift idea, real human
ROLE: A thoughtful gift-giver who avoids generic. CONTEXT: - Recipient: [name + relationship] - Specific traits / hobbies / quirks: [list — be granular] - Budget: $[X] - Occasion + delivery deadline: [when] - What they DON'T want / can't use: [allergies, space, ethics] - Vibe: [practical / sentimental / surprising / consumable] TASK: 1. Three gift options at different angles (one practical, one sentimental, one slightly unexpected). 2. For each: the specific item, where to get it, and ONE sentence on why it fits this person. 3. The pairing or add-on (note, packaging, experience) that turns it from a thing into a moment. 4. One gift I should NOT get them even if it seems perfect (and why). RULES: - No candles, generic Amazon top-10, or "experiences" unless they specifically love that type. - No flowers unless they've expressed love for them.
- BeginnerLife
Declutter one drawer
ROLE: A calm decluttering coach. CONTEXT: - What I'm tackling: [drawer / shelf / closet] - Time I have: [15 / 30 / 60 min] - Why this matters: [moving / overwhelm / specific goal] - Things I'm bad at deciding on: [sentimental / "might need it" / gifts] TASK: 1. A 15-minute step-by-step walk through, time-boxed (3 min sort, 7 min decide, 3 min stage, 2 min reset). 2. Five YES/NO questions to ask each item. 3. What to do with the "maybe" pile (a specific 30-day rule). 4. The ONE thing I should NOT decide today (and revisit when). 5. A quick "win" object to remove first to build momentum. RULES: - No "spark joy" wording. - Make it doable in the time I have. Don't expand scope.
- IntermediateLife
Sunday reset checklist
ROLE: You design boring, sustainable systems for chaotic lives. CONTEXT: - My household: [solo / partner / kids + ages / roommates] - The state Sunday usually starts in: [trashed / okay / surprisingly tidy] - The week ahead's pressure points: [list — e.g. early Monday, daycare drop-off, deadlines] - My energy on Sundays: [low / medium / high] TASK: Build a 60-minute Sunday reset, time-blocked: - 0–10 min: [decisive starting move] - 10–25 min: kitchen reset (specific tasks) - 25–40 min: week-prep (meals / clothes / calendar) - 40–55 min: ONE self-care thing (specific, not "relax") - 55–60 min: tomorrow's launch pad Also include: - A "if I only have 20 minutes" minimum-viable version. - One thing to delegate or automate (and how). - The ONE thing I should stop trying to do on Sundays. RULES: - No aspirational stuff. Realistic for a normal human.
- IntermediateLife
Weekly meal plan with one grocery list
ROLE: A practical meal planner who optimizes for ingredient reuse. CONTEXT: - Household: [how many + ages + appetites] - Dietary needs: [allergies / preferences / restrictions] - Time per dinner on a weeknight: [minutes] - Cooking skill: [beginner / intermediate / confident] - Equipment: [stove, oven, slow cooker, etc.] - Hard dislikes: [list] - Budget for groceries: $[X] TASK: 1. Five weeknight dinners using overlapping ingredients (≥60% reuse). 2. For each: prep time, cook time, one shortcut, one upgrade if time allows. 3. ONE consolidated grocery list grouped by store section (produce / protein / dairy / pantry / frozen / household). 4. A "use-it-up" plan for any ingredient that won't survive the week. 5. The ONE meal I should batch-cook on Sunday to make the week easier. RULES: - No specialty ingredient I'd use once and forget. - Realistic times, including the "second pan to wash" reality.
- AdvancedLife
Life audit, brutal but kind
ROLE: A life-coach version of a friend who's seen me make the same mistake twice. CONTEXT: I'll dump info on six areas. Be honest but never cruel. TASK: For each area (Health, Work, Money, Relationships, Fun, Growth): 1. Honest score 1–10 with one-line justification. 2. The #1 thing to change (specific, not "exercise more"). 3. One thing I should KEEP doing that's working. 4. The cheapest 2-week experiment to test the change. Then synthesize: - The pattern across areas (most people have one root they're avoiding). - The SINGLE highest-leverage move for the next 90 days. - The thing I'll likely do INSTEAD that will feel productive but won't move the needle. - The 1 quarterly check-in question. RULES: - No platitudes. No vision-board energy. - If something I wrote contradicts something else, surface it. Dump: """ [paste] """
- BeginnerWork
Status update in 3 lines
ROLE: A no-fluff status compressor. CONTEXT: - Audience: [manager / cross-functional partner / exec] - What they care about most: [outcome / risk / blocker] TASK: Turn my rambling update into exactly 3 lines: 1. DONE: [the most important shipped thing] 2. BLOCKED: [what's stuck + what I need + by when] 3. NEXT: [the single most important thing happening before our next sync] Then optionally add: ONE line of context if my audience is missing something critical. RULES: - Cut all filler ("just to give you a heads-up", "wanted to flag"). - If "blocked" is actually "I'm avoiding it", note that to me privately at the bottom. Update: """ [paste] """ - BeginnerWork
Slack message, not a novel
ROLE: A Slack-native communicator. CONTEXT: - Channel type: [DM / team / cross-team / leadership] - Urgency: [now / today / async / FYI] - The ask or update: [one line summary] TASK: Rewrite as a Slack message that: 1. Opens with the ASK or the HEADLINE — bolded, in the first line. 2. Adds context in ≤ 2 short lines. 3. Ends with: a question, a deadline, or "no action needed." RULES: - No "Hi team, hope you're all doing well!" - Use line breaks. Don't write paragraphs. - Mention specific people only if action is on them. Message: """ [paste] """
- IntermediateWork
Project brief on a napkin
ROLE: A PM who refuses to start without a one-pager. CONTEXT: - Project: [name] - Sponsor / owner: [who] - Why now: [trigger] TASK: Produce a one-page brief with: 1. PROBLEM — 3 sentences. Who hurts, how much, today. 2. AUDIENCE — who this is for. Who it is NOT for. 3. SUCCESS METRIC — the ONE number we'll move, with the baseline and target. 4. SCOPE — in / out / explicitly deferred. 5. MILESTONES — 4 dated checkpoints. 6. RISKS — top 3 with the mitigation for each. 7. OPEN QUESTIONS — what we don't know yet. 8. OWNER + DEADLINE. RULES: - If anything is unclear, ask me up to 3 questions BEFORE writing. - No "drive synergies."
- AdvancedWork
Stakeholder map
ROLE: A political-realist project leader. CONTEXT: - Project: [name + goal] - Stakeholders I'm aware of: [list with roles] - Decision-maker if known: [name] - History: [past projects, who supported / opposed] TASK: For each stakeholder, output a row: NAME | ROLE | WHAT THEY CARE ABOUT | WHAT THEY FEAR | INFLUENCE (1–3) | INTEREST (1–3) | THE ONE MESSAGE THAT MOVES THEM | FIRST CONTACT MODE (1:1 / email / together) Then a 2x2 (Influence × Interest) telling me: - Manage closely (high/high) - Keep satisfied (high inf / low int) - Keep informed (low inf / high int) - Monitor (low/low) End with: - The 3 conversations to have THIS WEEK, in order. - The one stakeholder who could quietly kill this, and the move to neutralize. - A line on who I'm probably MISSING from this map.
- Power UserWork
1:1 prep machine
ROLE: A 1:1 coach who turns vague conversations into useful ones. CONTEXT: - Other person: [name + role + how long we've worked together] - Cadence: [weekly / monthly] - Recent context: [what happened since last 1:1] - The relationship dynamic: [aligned / tense / new / repairing] - What I want from this conversation: [outcome] TASK: Output a 1:1 prep card: 1. THREE wins to acknowledge — specific to them, not generic. 2. TWO issues to surface, each with a non-blamey opener. 3. ONE thing I want to LEARN from them (not tell them). 4. The single question that, if asked, would tell me how they're really doing. 5. Their likely defensive reaction, and how I keep it productive. 6. The ONE follow-up I'll commit to before next 1:1. RULES: - Don't pad with "small talk topics." - If I'm avoiding the real issue, name it in a private note at the end. Context: """ [paste] """
- BeginnerParenting
Bedtime story, custom
ROLE: A calm storyteller for a winding-down kid. CONTEXT: - Kid's age: [age] - Hero's name + traits: [name + 3 traits] - Things they love (animals, vehicles, foods): [list] - Lesson or feeling to land softly: [theme — e.g. brave, patient, kind, share] - Things to avoid: [scary characters, separation themes, etc.] TASK: Tell a ~5-minute bedtime story with: - A gentle opener that signals "story starting now." - 3 small "scenes" with rising warmth (no real conflict). - A calm ending where the hero settles in safely. - A final sentence that quietly invites sleep. RULES: - No villains, no chases, no loud noises. - Sentence length should get shorter as the story progresses.
- BeginnerParenting
Screen time rule that sticks
ROLE: A practical parent-coach. Not preachy. CONTEXT: - Kid's age + temperament: [age + style] - Current screen pattern (honest): [hours + what + when] - What I'm worried about: [behavior / sleep / mood / siblings] - What I'm willing to enforce consistently: [be realistic] - My partner's stance: [aligned / different / not in the picture] TASK: 1. ONE clear rule, stated in the kid's words (≤ 15 words). 2. WHEN the rule applies (and when it doesn't — exceptions named upfront). 3. The 3-line script for the FIRST time they push back. 4. The 1-line script for the 10th time they push back. 5. The natural consequence (NOT a punishment) if the rule is broken. 6. The thing to replace the screen WITH so the rule has somewhere to land. RULES: - The rule must be enforceable by ONE parent, on a tired Tuesday. - No "we'll see" — that's how rules die.
- IntermediateParenting
Tantrum decoder
ROLE: A developmental-psych-aware parent coach. Practical, not clinical. CONTEXT: - Kid's age: [age] - Trigger (what visibly set it off): [trigger] - What was happening just before (sleep, food, transitions): [state] - The shape of the tantrum: [shutdown / explosion / "I want X" loop] - My usual response: [what I do] TASK: 1. Under-the-surface read: what's likely going on developmentally / emotionally (3 bullets). 2. RIGHT NOW (during the meltdown): the 4-step move (regulate self → connect → name → wait). 3. LATER, when calm: the conversation script (2–3 lines). 4. What NOT to say in the moment, and why those backfire. 5. The pattern to track over the next week (trigger / time / context) so I can see the real cause. 6. When to seek extra help (specific signs, not vague). RULES: - No shaming me. No "be more patient." - Be specific. Real words I can say.
- IntermediateParenting
Big topic, small kid
ROLE: A child-life specialist for hard conversations. CONTEXT: - Kid's age + emotional readiness: [age + style] - Topic: [death / divorce / illness / scary news event / job loss] - What they may have already heard or seen: [if anything] - Family values to anchor in: [1–2 lines] TASK: 1. ONE-sentence opening line, in their words. 2. The honest answer to "what does that mean?" at their age level. 3. The honest answer to "could that happen to us / you / me?" without false reassurance. 4. What to do if they cry, freeze, or change the subject. 5. The closing that ends the conversation without slamming the door ("you can ask me again anytime; there are no wrong questions about this"). 6. Two age-appropriate follow-ups for the next few days. RULES: - Calm, honest, age-appropriate. Don't lie. - Match their pace; let them lead the depth. - AdvancedParenting
Family operating system
ROLE: A boring-but-effective family-systems designer. CONTEXT: - Household members + ages: [list] - Current chaos points (mornings, dinners, bedtimes, weekends): [where it falls apart] - Adult work patterns: [shifts / WFH / travel] - Family values to preserve: [2–3] - What we've tried that DIDN'T stick: [list — and why we think] TASK: Design a simple "family OS" with: 1. MORNING rhythm — 5 steps max, time-boxed, with the "if late" fallback. 2. EVENING rhythm — 5 steps, including transition to wind-down. 3. WEEKLY 15-min family meeting — kid-friendly format with 4 sections (cheers, calendar, concerns, choices). 4. AGE-SCALED chore system — small kids / older kids / adults — and how chores rotate. 5. CONFLICT process — the 3-step move when siblings or adults disagree. 6. "OFF-DUTY" — built-in time when no one is responsible for the household. RULES: - Sustainable on a hard week. - Visible (where it lives so it's actually used). - Boring beats Instagram-perfect.
- BeginnerShopping
Is this a fair price?
ROLE: A pricing-savvy shopper. CONTEXT: - Item: [exact model / SKU if possible] - Price seen: $[X] - Region: [country / city] - Condition: [new / open-box / used / refurbished] - When I need it: [now / can wait for a sale] TASK: 1. The LOW / FAIR / HIGH price range for this item in this region/condition right now. 2. What justifies the HIGH end (features / scarcity / seller reputation). 3. The cheapest legit channel to buy from, and why it's not sketchy. 4. The next likely sale window or price drop, if any. 5. The version of "settle" — a slightly older / lesser model that's 80% as good for far less. RULES: - Be honest about uncertainty if your data is stale. - Don't recommend buying if I should wait.
- BeginnerShopping
Read the reviews for me
ROLE: A review skeptic with an eye for buried truths. CONTEXT: - Product: [name + model] - My use case: [specific] - What would make me return it: [deal-breakers] TASK: 1. The 3 most-mentioned PROS. 2. The 3 most-mentioned CONS. 3. The ONE complaint that reviewers downplay but I should care about for my use case. 4. The ONE compliment that's misleading (e.g. "great for the price" = "actually not great"). 5. The pattern of who LOVES it vs who REGRETS it (e.g. casual vs heavy users, climate, hand size). 6. The verdict for me — buy / buy with caveat / skip — in one sentence. RULES: - Ignore one-line reviews and obvious fakes. - If the reviews are sparse, say so and recommend a comparable product with more data. Reviews: """ [paste] """
- IntermediateShopping
Buy-once gear
ROLE: A buy-it-for-life advisor. CONTEXT: - Item: [category — e.g. chef's knife, raincoat, headphones] - Use case + frequency: [daily / weekly / occasional] - Environment / wear conditions: [climate, rough use, travel] - Aesthetic preference: [matters / doesn't] - Budget ceiling: $[X] TASK: 1. The "buy it for life" pick — what + why + roughly when to expect to replace anything. 2. A solid MID-TIER that does 90% of the work for far less. 3. A budget option that won't embarrass itself (with the trade-off I'll feel). 4. The maintenance / care that makes any of these last longer. 5. The ONE feature people obsess over that does NOT matter for my use. 6. What to AVOID at any price point (and brand-pattern red flags). RULES: - Plain English on trade-offs. - Specific brands and models if you have them.
- IntermediateShopping
Used vs new math
ROLE: A total-cost-of-ownership thinker. CONTEXT: - Item: [model] - New price: $[X] - Used price (typical): $[Y] - Refurbished price + warranty: $[Z] - How often I'll use it + for how long: [usage] - My tolerance for hassle (returns, fixes, listing failures): [low / medium / high] TASK: 1. Real depreciation curve for this item over the first 3 years. 2. Failure / repair rate by year (real-world, not marketing). 3. Warranty math: cost of breakage × probability for each option. 4. Time cost of going used (sourcing, inspecting, returns). 5. Resale value if I sell in 2 years. 6. Verdict: NEW / USED / REFURBISHED — and the price threshold that flips the answer. RULES: - Show your math. - If used is dominated by refurb in this category, say so plainly.
- AdvancedShopping
Total cost of ownership
ROLE: A TCO analyst. CONTEXT: - Two finalists: [A] vs [B] - Use horizon: [5 years] - Expected usage profile: [light / heavy / specific scenarios] - My region (for service / parts / energy costs): [region] TASK: Build a 5-year TCO comparison. Rows for each option: - Purchase price - Accessories / setup - Consumables / subscriptions (annualized) - Repair likelihood × average repair cost - Insurance (if relevant) - Energy / fuel (if relevant) - Time cost (setup, maintenance, learning curve) — converted to $ at my hourly rate - Resale value at year 5 Then: - Winner at typical use. - Break-even USAGE LEVEL where the other option wins. - The single assumption that, if wrong, flips the answer. RULES: - Show working. Don't round-trip to vibes.
- BeginnerWriting
Headline lab
ROLE: A headline writer working across 5 styles. CONTEXT: - Topic / piece: [paste the article or describe] - Audience: [who is scrolling past this] - Where it will live: [blog / X / LinkedIn / newsletter subject] - Voice constraint: [no clickbait? no exclamation? no jargon?] TASK: Write 10 headline options. Cover this mix: - 2 PLAIN & USEFUL (what + payoff) - 2 CURIOUS (open loop without lying) - 2 CONTRARIAN - 2 EMOTIONAL / story-driven - 2 LIST or NUMBER Mark your TOP 3 with the style each uses and the audience each best converts. RULES: - Each ≤ 12 words (60 chars). - No bait-and-switch. - No emojis unless platform demands.
- BeginnerWriting
Anti-corporate translator
ROLE: A jargon executioner. CONTEXT: - Audience: [who reads this and what they actually want to know] - Meaning that must survive intact: [list the 2–3 facts that can't be lost] TASK: Rewrite the text so it sounds like a smart human, not a press release. - Replace every buzzword with a plain verb or concrete noun. - Strip "leverage", "robust", "synergize", "innovative", "world-class", "best-in-class", "passionate", "delighted to announce". - Shorter sentences. Active voice. - Keep the meaning. Lose the haze. Then list, in 3 bullets, the worst phrases you killed and what you replaced them with. RULES: - If a piece is entirely corporate fog with no substance, say so and ask what we're actually trying to communicate. Text: """ [paste] """
- IntermediateWriting
Hook + nut graf
ROLE: A long-form feature writer. CONTEXT: - Topic: [topic] - Audience + what makes them keep reading: [audience + what they're hoping for] - The thesis (in one sentence): [thesis] TASK: Write THREE opening hooks (3 distinct angles): 1. Scene / image hook 2. Counter-intuitive claim hook 3. Personal-stake / specific-character hook For EACH hook, write the 2-sentence "nut graf" that follows it — the paragraph that earns the next 5 minutes by telling the reader what they'll learn AND why it matters NOW. Then pick the WINNER, explain why it earns the reader's next 30 seconds, and name the type of reader it'll lose (and whether we're okay losing them). RULES: - No "in today's fast-paced world." - The hook must connect logically to the thesis.
- AdvancedWriting
Show don't tell pass
ROLE: A literary editor focused on craft. CONTEXT: - Form: [essay / memoir / fiction / brand story] - Voice constraints: [first person ok? humor ok?] TASK: Read the draft. Find every "tell" — every sentence where I'm telling the reader the conclusion instead of letting them feel it. For each: 1. Quote the line. 2. Diagnose the type of tell (emotion named / interpretation given / verdict stated). 3. Propose a "show" rewrite using a SPECIFIC image, a line of dialogue, an action, or a sensory detail. 4. Note which option is strongest and why. Also flag: - Any moment of telling that is actually a load-bearing wall (where showing would hurt pace). Leave those. - The single best already-shown moment in the draft, so I can echo it. RULES: - No generic "add more sensory detail" notes. Be specific to MY scene. Draft: """ [paste] """
- Power UserWriting
Reverse outline
ROLE: An editor doing a structural diagnostic. CONTEXT: - Goal of the piece: [persuade / explain / move emotionally / sell] - Audience: [who] - Length target: [words] TASK: Reverse-outline the draft. For each paragraph, output 2 things: - SAYS: a 5-word summary of WHAT it says. - DOES: 5 words on what it DOES in the piece (sets tension, gives evidence, transitions, releases pressure, lands the point). Then diagnose: 1. Where the structure DRAGS (repeats, restates, loses momentum). 2. Where the structure LIES (says it's making a new point but isn't). 3. Where the structure LEAPS (skips a step the reader needs). 4. Propose a new paragraph order, and explain the new arc in 3 bullets. 5. Name the ONE paragraph that, if cut, would lose nothing. RULES: - Be ruthless on cuts. Praise sparingly and specifically. Draft: """ [paste] """
- BeginnerResearch
Dumb questions first
ROLE: A patient explainer who believes the dumb question IS the question. CONTEXT: - Topic I'm starting to research: [topic] - Why I'm researching it: [reason — decision / curiosity / writing / interview] - What I already (think I) know: [1 sentence] TASK: 1. Ten "dumb" questions I should be embarrassed not to know the answers to. 2. Answer each in ONE sentence. 3. Mark which 3 questions are most load-bearing for my purpose. 4. Suggest the 3 questions a true expert would also ask but a beginner wouldn't think to. RULES: - No prerequisites. No "first, let's define…" — go straight to the question.
- BeginnerResearch
Map the landscape
ROLE: A landscape cartographer for a new topic. CONTEXT: - Topic: [topic] - My background level: [zero / hobbyist] - Decision or output I'm working toward: [purpose] TASK: Produce a 1-page map: 1. The KEY PLAYERS — companies, schools of thought, authors, communities — grouped by camp. 2. The TIMELINE — 5 inflection points that shaped the field. 3. The BIG OPEN QUESTIONS — what experts still argue about. 4. The COMMON BEGINNER MISTAKES. 5. WHERE TO START — one accessible read, one weird/under-rated one, one foundational primary source. 6. The TWO sub-fields adjacent to this that someone studying this should also know exist. RULES: - Honest about what's still unsettled.
- IntermediateResearch
Three levels of source
ROLE: A reading-list curator. CONTEXT: - Topic: [topic] - Time I'll invest: [hours / days / weeks] - My goal: [understand / decide / write / teach] TASK: Recommend ONE source at each level, with a real link or unmistakable title: 1. POPULAR INTRO — accessible, well-written, mostly accurate. What to skip in it. 2. DEEPER WORK — a book or long-form that adds real depth. What to focus on, what to skim. 3. PRIMARY SOURCE — an academic paper, original text, or canonical dataset. The one section to read first. For each: how long it takes, what it gives you that the others don't, and the one MISLEADING idea it might leave you with that the next level corrects. RULES: - If a popular intro is outdated or oversimplified, say so.
- AdvancedResearch
Skeptic's reading list
ROLE: An adversarial reading-list builder. CONTEXT: - The thing I just read or believe: [paste or summarize] - Why I found it persuasive: [reason] - The strength of my belief: [1–10] TASK: Recommend 3 sources that push back on it WELL — not bad-faith haters, the strongest credible critics. For each: 1. Source + author + year. 2. The CORE counter-argument in 2 sentences. 3. The strongest piece of evidence they bring. 4. Where they're vulnerable in turn. 5. What it would take to update from their critique. End with: - The ONE source most likely to actually change my mind, and the chapter to start with. - The version of the original claim that survives all three critiques (if any). RULES: - No "balance" theater. Pick real critics.
- Power UserResearch
Lit-review skeleton
ROLE: A research-methods coach building a literature review skeleton. CONTEXT: - Research question (draft): [question] - Field(s) it sits inside: [list] - Constraints: [time / access / language] - What I'll do with the review: [paper / decision / brief] TASK: 1. The question stated PRECISELY (with scope, population, time horizon). 2. KEY TERMS + synonyms to search (with 2 sample search strings). 3. FIVE seminal works to find (author + year + why it matters). 4. THREE review papers that summarize the field. 5. The NAMED DEBATES and who's on which side. 6. What is NOT yet settled (and what evidence would settle it). 7. The methodology I'd use to extract claims from each source consistently (extraction template). RULES: - Cite where you can. Mark [inference] when guessing. - Flag where your training data is likely stale.
- BeginnerNegotiation
Ask for a discount
ROLE: A polite-but-effective discount asker. CONTEXT: - Item / service: [what] - Sticker price: $[X] - The reason a discount is plausible (loyalty, volume, off-season, cash, cosmetic flaw): [reason] - Channel: [in person / phone / email / chat] - My BATNA: [walk away easily / need this] TASK: 1. ONE-line opener that doesn't sound cheap. (Make it about THEM keeping a customer or moving inventory, not me being poor.) 2. The follow-up if they say "let me check." 3. The line if they refuse outright. 4. The line if they offer HALF of what I asked. 5. The graceful close if I want to leave the relationship intact either way. RULES: - Smile in the words. Direct, not whiny. - Mention the exact number once — then stop talking.
- IntermediateNegotiation
Counter-offer email
ROLE: A counter-offer writer who anchors well. CONTEXT: - The offer on the table: $[X] (or terms) - My target: $[Y] - My BATNA + how visible to them: [be honest] - Their likely flexibility: [low / medium / high — and why] - Relationship I want to preserve: [yes / it's transactional] TASK: Draft a counter-offer email that: 1. Anchors ~15% above my target, with 2 specific justifications grounded in value (not need). 2. Acknowledges what's GOOD about their offer first (one sentence — sincere). 3. Frames the counter as a path to "yes," not a rejection. 4. Includes a non-price ask I'd trade for if the number won't move. 5. Sets a deadline tied to a reason ("I'm finalizing on Friday because…"). RULES: - No "I deserve" / "I need." Anchor on value. - One ask only. Don't list 7 things. - IntermediateNegotiation
The silent treatment script
ROLE: A negotiation coach teaching strategic silence. CONTEXT: - Negotiation type: [salary / sales / vendor / personal] - Setting: [in person / video / phone] - My instinct under pressure: [over-explain / shrink / fill silence] TASK: 1. The 3 moments in a negotiation where silence is the most powerful move. 2. How LONG to hold each silence (with seconds — feel longer than they sound). 3. What to do with face / breath / posture (so I don't leak nerves). 4. What to say to BREAK the silence if they don't fold (and the tone to use). 5. What to say AFTER they break first (don't pounce — leave dignity). 6. One drill I can practice this week to build the muscle. RULES: - This is a tool, not a weapon. Pair with warmth.
- AdvancedNegotiation
Multi-issue tradeoff
ROLE: An integrative-negotiation strategist. CONTEXT: - Negotiation: [what we're trying to agree on] - MY issues with relative importance (1–10): [list] - THEIR likely issues with importance (your guess): [list] - The relationship after the deal: [one-shot / ongoing] - Constraints on either side: [time, budget, optics] TASK: 1. Identify the issues where my importance is LOW and theirs is HIGH (cheap gives for me). 2. Identify the inverse (cheap gives for them). 3. Design 3 PACKAGES that trade across issues: - Package A: anchor / aspirational - Package B: realistic landing zone - Package C: fallback that still meets my must-haves 4. The OPENING package and why. 5. The signals from them that mean to move to B, then C. 6. The single "deal-breaker" I should never trade, named explicitly. RULES: - Don't optimize on price alone. Look for the trade.
- Power UserNegotiation
Tactical empathy script
ROLE: A Chris Voss-style tactical-empathy negotiator. CONTEXT: - The situation: [paste] - Their emotional state, best guess: [worried / defensive / proud / scared] - The outcome I want: [specific] - What I'd accept as a B-grade win: [fallback] TASK: Write the script as a numbered turn-by-turn dialogue, using: - MIRRORS (repeat the last 1–3 words of what they said, as a question) - LABELS ("It seems like you're worried about…", "It sounds like…") - CALIBRATED QUESTIONS ("How am I supposed to do that?", "What about this is most important to you?") - A NO-ORIENTED QUESTION ("Have you given up on this deal?") - A "THAT'S RIGHT" goal — design toward the moment they say it. For each line, annotate in italics WHY the move works at that moment. End with the 1 line I should say FIRST if I have only one shot. RULES: - Sound like a human, not a hostage-negotiation textbook. - Don't manipulate. Aim for genuine understanding that unlocks the deal. - BeginnerTravel
Pack for the trip
ROLE: A frequent-traveler packing coach. CONTEXT: - Destination + season: [city + month] - Trip length: [N days] - Activities: [list — meetings, hiking, beach, formal dinner] - Bag constraint: [carry-on only / checked ok] - Laundry available: [yes / no / maybe] - I always forget: [list, if known] TASK: 1. Outfit-based packing list (not item-based) so I see complete looks, not a pile. 2. 3 ITEMS people commonly forget for this trip type. 3. What to BUY at destination instead of bringing (cheaper or bulky). 4. The ONE "secret weapon" item that punches above its weight. 5. Toiletries / meds checklist with sizes that pass airport security. 6. Tech + chargers + plug adapter for the region. 7. The 5 things I should put in the SEAT pocket / personal item for the flight. RULES: - Realistic for the climate, not aspirational. - Mark "leave home" items I won't actually wear.
- BeginnerTravel
Eat like a local
ROLE: A local food guide who hates Instagram lines. CONTEXT: - Neighborhood + city: [where] - Cuisine I want: [type or "anything authentic"] - Budget per meal: $[X] - Dietary needs: [list] - Time of day: [breakfast / lunch / dinner / late-night] TASK: 5 spots, mix of price points. For each: 1. Name + what to ORDER (specific dish). 2. WHEN to go (and when to avoid). 3. How to BEHAVE so they don't roll their eyes (queue, cash only, seating norms, "ordering at the counter," etc.). 4. The ONE dish to skip even though it's on every menu. 5. The next-block hidden gem if my pick is closed. RULES: - No tourist-trap top-Google-result restaurants. - Include at least one place a local would bring their parents.
- IntermediateTravel
Hidden-cost audit
ROLE: A travel-budget realist. CONTEXT: - Trip type: [city / cruise / road trip / multi-stop] - Stated budget: $[X] - Booked already: [flights / hotel / tours] - My payment setup: [credit card type, fx fees] TASK: Audit for hidden costs people miss: 1. Ground transfers (airport, intercity, daily) — ranges. 2. Tips / service charges — local norms + amounts. 3. Tourist taxes / city fees / resort fees. 4. Baggage / seat selection on each flight. 5. Currency conversion + ATM fees + best card to use where. 6. eSIM / roaming / wifi. 7. "Free" tours, attractions with paid add-ons, photo passes, museum reservations. 8. Food cost realism (not just dinners — coffee, snacks, water). 9. Buffer for "we want to do the thing we didn't book." End with: a revised total range and the 2 line items most likely to blow the budget. Plan: """ [paste] """
- AdvancedTravel
Travel hacking 101
ROLE: A points-and-miles strategist. CONTEXT: - Home airport(s): [X] - Region I'd love to visit: [or "flexible"] - Travel window: [dates / month / quarter] - Points balances: [program: amount] - Travel partner / family: [solo / 2 / family — cabin class needed] - Risk tolerance: [conservative / chase availability] TASK: 1. Three high-value redemption ideas given my balances, ranked by cents-per-point and ease. 2. For each: the catch (transfer-partner timing, fuel surcharges, holding availability, taxes). 3. The single move I should make THIS WEEK (transfer, application, hold, booking call). 4. The trap I should NOT fall into (most common bad redemption in my position). 5. The credit-card move (if any) that would unlock a meaningfully better redemption in 3 months. RULES: - Cite typical cpp and award charts; flag if data is stale. - Be honest if my balances aren't enough — and suggest the cheapest path forward.
- Power UserTravel
Slow-travel month
ROLE: A slow-travel designer for productive humans. CONTEXT: - Travelers: [solo / couple / family] - Total time: [4 weeks] - Work obligations: [hours / week + meeting hours + timezone tolerance] - Climate preference: [+ months] - Budget: $[X] including flights + lodging + life - Energy preference: [deep / varied / both] - Pace: max 2 base moves TASK: Design a 4-week itinerary: 1. THREE base cities/towns (with the rationale for each). 2. The order + travel days between them. 3. Lodging type per base (apartment / boutique / homestay) with neighborhood + WHY. 4. Workspace setup: where to actually work + wifi reality. 5. Weekly rhythm: 1 deep experience, 1 small ritual, 1 day off-grid. 6. Fallback at each base if weather / mood / work explodes. 7. Pre-trip checklist: visas, vaccinations, eSIM, banking, mail, pet, plants. RULES: - Optimize for depth over checklist. - Walkable bases. Real food. Sleep.
- BeginnerHealth
Symptom journal starter
ROLE: A patient-advocacy coach helping me prepare data for a clinician. CONTEXT: - Issue: [symptom / condition] - How long it's been going on: [duration] - Pattern I've noticed (or "none yet"): [pattern] - What I've already tried: [list] TASK: 1. The TOP 5 things to track that clinicians actually find diagnostic (e.g., severity 1–10, time of day, triggers, what helped, what made it worse, related symptoms). 2. The cadence (real-time / daily / when it happens) for each. 3. A simple template (table format I can paste into Notes app). 4. Common pitfalls in self-tracking (over-tracking, missing context, recall bias). 5. WHEN to seek care urgently — list of red-flag symptoms. 6. What to bring to the appointment to make this data useful. RULES: - Plain language. No "may indicate possible" hedging that doesn't help. - This is data collection, not diagnosis.
Tip: Friendly heads-up: this is general info, not medical advice. Decide with a qualified professional.
- BeginnerHealth
Read this lab result
ROLE: A patient-friendly lab-result translator. NOT a diagnostician. CONTEXT: - The result: [paste] - My basic info: [age, sex, any chronic conditions, relevant meds] - The reason this lab was ordered: [if known] TASK: 1. Plain-English meaning of EACH value (what's being measured + why it matters). 2. What's WITHIN normal range (and where in the range — "high-normal" matters). 3. What's FLAGGED — and what a flag at this level commonly means (not "could be anything"). 4. The 3 questions to ask my doctor (specific to my results). 5. What NOT to panic about based on this lab alone. 6. The pattern of follow-up tests that typically come next, so I'm not surprised. RULES: - Don't diagnose. Translate. - If a result needs urgent attention based on standard guidelines, say so plainly. Result: """ [paste] """
Tip: Friendly heads-up: this is general info, not medical advice. Decide with a qualified professional.
- IntermediateHealth
Habit stack for sleep
ROLE: A sleep-routine designer who designs for tired humans, not optimal ones. CONTEXT: - Bedtime target: [time] - Current wind-down (honest): [what I actually do in the last 90 min] - Sleep enemies in my life: [phone / caffeine / kids / work / anxiety] - Energy I have for routine changes: [low / medium] - Bedroom setup: [light / temp / noise / partner] TASK: Build a 30-minute wind-down stacking 4 small habits. For each step: - The action (concrete, ≤ 5 min) - The cue that triggers it (so I don't have to remember) - The "good enough" version on a bad day - Why this step matters physiologically Then: - A 10-minute MINIMUM version for hard days. - The ONE environmental change to make this week. - The signal I'll use after 2 weeks to know it's working. RULES: - No "stop using phones" without giving me what to do instead. - Real, not ideal.
- AdvancedHealth
Strength program from scratch
ROLE: A pragmatic strength coach for a beginner with a real life. CONTEXT: - Age + sex + relevant injuries: [info] - Equipment available: [list] - Sessions per week + minutes each: [X × Y] - Primary goal: [strength / fat loss / function / energy] - Secondary goal: [if any] - History: [never lifted / lifted years ago / current] TASK: Build a 12-week beginner program. - Weekly template (which days, what splits, recovery days). - For each lift: progression scheme + how to know when to add weight. - Warmups (≤ 8 min, specific). - Deload week placement. - The "I missed a week" rule (don't catch up — just restart at last successful weight). - The 3 movements to AVOID until form is solid, and the substitution for each. End with: - A 1-page printable summary. - The single sign I should stop and see a pro. RULES: - Honest about realistic progress on this schedule. - Don't bury me in volume.
Tip: Friendly heads-up: this is general info, not medical advice. Decide with a qualified professional.
- Power UserHealth
Second-opinion partner
ROLE: A careful, conservative second-opinion partner. NOT a doctor. CONTEXT: - My situation: [details, history, current symptoms] - The doctor's recommended plan: [paste] - My doctor's reasoning, as I understand it: [paste] - My specific concerns: [list] TASK: 1. The GOOD REASONS to follow this plan (steel-man it before challenging). 2. Reasonable ALTERNATIVES a sharp clinician might consider, with the trade-offs of each. 3. Questions I should bring back to my doctor (specific, not generic). 4. Red flags that would warrant a real second opinion from another clinician. 5. What I should TRACK between now and the next visit to make the next conversation productive. 6. How to talk to my doctor about wanting a second opinion without damaging the relationship. RULES: - Be direct about uncertainty. - Make clear this is a structured way to think — not a diagnosis or treatment plan.
Tip: Friendly heads-up: this is general info, not medical advice. Decide with a qualified professional.
- BeginnerMoney
Subscriptions audit
ROLE: A subscription auditor with no loyalty. CONTEXT: - My monthly subscription list with prices: [paste] - What I use weekly vs monthly vs "haven't opened in 90 days": [if known] - My income range: [for context] TASK: 1. Rank ALL subscriptions by VALUE PER DOLLAR for me, not for the market. 2. Tag each: KEEP / DOWNGRADE / PAUSE / CANCEL. 3. The ONE subscription I'm probably forgetting about that's quietly draining money. 4. The bundle / annual upgrade that would actually save me money (and which is fake savings). 5. The ONE cancellation that will sting at first but I won't miss in 30 days. 6. A script to use if a service makes me call to cancel. RULES: - No "you should keep Netflix because…" without checking my actual usage. - Total monthly savings if I act on all your recommendations.
- BeginnerMoney
Should I pay this bill or dispute?
ROLE: A consumer-protection-savvy second opinion. NOT a lawyer. CONTEXT: - The bill / charge: [paste] - What I was charged for: [service / product] - What I actually agreed to: [contract / quote / verbal] - Date of charge + payment deadline: [dates] TASK: 1. Whether this charge looks LEGIT, SUSPICIOUS, or DEFINITELY WRONG — and why. 2. The 3 things to verify (look at receipt, contract, T&Cs, prior emails). 3. If disputing: the exact 3-line script (calm, specific, asks for a fix not a fight). 4. The escalation ladder (provider → manager → consumer protection / credit card chargeback / regulator) and WHEN to step up. 5. Time limits I should know (chargeback window, statute, contract terms). 6. The risk of ignoring it (credit score, collections, late fees). RULES: - No legal advice. Direction + scripts. - Realistic about effort vs. amount.
- IntermediateMoney
Emergency fund target
ROLE: A personal-finance coach. NOT a financial advisor. CONTEXT: - Monthly fixed costs (rent, food, insurance, debt mins): $[X] - Job stability: [stable / volatile / self-employed / seasonal] - Dependents: [who relies on me] - Existing safety nets: [partner income / family backstop / severance] - Health considerations: [if any] - Region: [country, cost of living] TASK: 1. The DOLLAR TARGET for my emergency fund (range), with reasoning. 2. WHERE to keep it (account type + 2 specific institutions worth considering). 3. A monthly contribution that gets me there in [N] months without breaking life. 4. What does NOT count as emergency fund (retirement accounts, brokerage, "I'll just use credit"). 5. The TRIGGER to actually use it (real emergency vs. "I really want this"). 6. The replenishment plan after I tap it. RULES: - General info, not personalized advice. - Be realistic about the order: minimum debt → starter fund → high-interest debt → full fund.
Tip: Friendly heads-up: this is general info, not financial advice. Decide with a qualified professional.
- AdvancedMoney
Tax-advantaged stack
ROLE: A general-knowledge guide to tax-advantaged accounts. NOT a tax advisor. CONTEXT: - Country / state: [region] - Employment: [W2 / self-employed / mix] - Employer plan + match: [details] - Income range (for limits): [range] - Family status: [single / married / kids / aging parents] - Goal horizon: [retirement / education / health / flexibility] TASK: 1. The PRIORITY ORDER for filling tax-advantaged buckets in my situation, with WHY. 2. The "free money" steps people skip (employer match, HSA, retirement saver's credit, etc.) at MY income. 3. The trade-off of each tier (lock-up, withdrawal penalties, flexibility). 4. The line where it makes sense to move to taxable brokerage instead. 5. The 1–2 moves I should make THIS YEAR before the calendar matters (catch-up, conversion, contribution deadlines). 6. The most common mistake at my level. RULES: - General info. Tax law changes; flag where my country/year matters. - If I gave you incomplete info, ask before recommending.
Tip: Friendly heads-up: this is general info, not tax or financial advice. Decide with a qualified professional.
- Power UserMoney
Personal balance sheet
ROLE: A personal-CFO scribe. NOT a financial advisor. CONTEXT: - My ASSETS: [cash, brokerage, retirement, real estate, equity, vehicles, other — with rough values] - My LIABILITIES: [mortgage, student loans, credit card, car loan, other — balances + rates] - Income (after tax, monthly): $[X] - Life stage: [accumulating / coasting / decumulating] TASK: 1. Net worth (and break it down by liquid / illiquid / restricted). 2. Liquidity ratio (liquid assets / monthly fixed costs). 3. Debt-to-income. 4. Savings rate (monthly savings / monthly net income). 5. Concentration risk (any one asset > X% of net worth). 6. The 3 ratios I should TRACK quarterly given my stage. 7. The SINGLE most important number for my life stage, with reason. 8. The 2 moves that would most improve the balance sheet over the next 12 months. RULES: - General info, not personalized advice. - Show formulas so I can recompute. Inputs: """ [paste] """
Tip: Friendly heads-up: this is general info, not financial advice. Decide with a qualified professional.
- BeginnerConflict
Cool-down line
ROLE: An emotional brake-pedal. CONTEXT: - What I'm about to send (while angry): [paste] - Who it's going to: [relationship] - What I'm trying to achieve underneath the anger: [real goal] TASK: 1. A COOLED version that says what I actually mean, with the dignity intact for both sides. 2. The single line in MY draft that would have most damaged the relationship, and why. 3. A line I should sit with for ONE HOUR before sending anything at all (something to ask myself). 4. The two-question gut check before hitting send: "Is this true? Is this useful? Is this kind?" — and which one MY draft fails. 5. The "do nothing for 24 hours" version if the situation could survive my silence. RULES: - Don't strip my honesty. Strip the venom. - If sending nothing is the right move, say so.
- BeginnerConflict
Apology that doesn't cave
ROLE: An apology writer with a spine. CONTEXT: - What I'm apologizing for (my part): [specific] - What I'm NOT apologizing for (theirs / shared): [specific] - The relationship + what I want to preserve: [info] - Their likely reaction: [defensive / cold / open] TASK: Write a 4-sentence apology that: 1. Owns MY part specifically — no "if anyone was hurt." 2. Names the IMPACT (not just the act). 3. States what I'll do differently (one concrete change). 4. Leaves my dignity intact and doesn't apologize for things that weren't mine. Then provide: - ONE line to add if they try to widen the apology to cover their grievances. - ONE line for if they reject the apology. RULES: - No groveling. No "I'm such an idiot." - Specific > sweeping.
- IntermediateConflict
Boundary in 3 lines
ROLE: A boundary-coach who keeps it short and enforceable. CONTEXT: - Person + relationship: [who] - Behavior I'm addressing: [specific — what they do, when] - Why this matters NOW: [pattern / recent escalation] - My realistic enforcement ability: [what I'll actually do] - Risk: [will they retaliate / cut off contact / escalate] TASK: Write the boundary as 3 lines: 1. The BOUNDARY, stated cleanly ("I'm not available for X."). 2. The CONSEQUENCE if crossed (a thing I will do, not a threat about them). 3. The WARM CLOSER (one sentence that keeps the door open without weakening the boundary). Then: - The script for the FIRST pushback. - The script for the THIRD pushback. - A reminder of what to do with my body (slow breath, soften jaw) to stay grounded. RULES: - A boundary is what I will do, not what they must do. - Don't justify. State it. - AdvancedConflict
Map their incentives
ROLE: A perspective-taking coach. CONTEXT: - The conflict: [paste] - What I think they want: [my guess] - What I think they're afraid of: [my guess] - What I know about their recent life pressures: [info] TASK: 1. List the 5 PRESSURES likely on them right now (work, home, health, money, status, identity). 2. The version of this conflict where they're NOT a villain — write it sincerely. 3. The 3 things they're probably trying to protect. 4. The thing I'm doing that, from their angle, looks worse than I realize. 5. The MOVE that would respect their world AND mine, in one paragraph. 6. The opening line that signals I see them, without surrendering my position. RULES: - No "well, both sides…" wrap-up. - Generous reading, not naïve reading.
- Power UserConflict
Mediator simulation
ROLE: Run a simulated mediation. You will play 3 roles in rotation. CONTEXT: - The conflict: [paste background] - The relationship: [type + history] - What surface issue we keep fighting about: [topic] - What I suspect the real issue is: [my guess] TASK: Run the simulation in 4 turns. At each turn label clearly: ME / THEM / MEDIATOR / COACH-TO-ME. Turn 1 — Each side states the surface complaint. (You write THEM. I'll respond as ME.) Turn 2 — Mediator reflects what each is saying and asks ONE deeper question of each. Turn 3 — THEM responds at the next layer down. Coach pauses to tell me what I'm about to react to and why. Turn 4 — Stop when we hit the real disagreement underneath the surface one. State it plainly. Then output: 1. The REAL disagreement, named in one sentence. 2. The single sentence I should say FIRST in the real conversation. 3. The thing I should NOT bring up the first time, even if I'm right. RULES: - Play THEM steel-manned, not a strawman. - Stop after 4 turns even if unresolved. Background: """ [paste] """
- BeginnerLearning
First-principles primer
ROLE: A first-principles teacher. CONTEXT: - Topic: [topic] - My background: [zero / hobbyist / partial] - Why I want to understand this: [decision / curiosity / teaching others] TASK: 1. The most BASIC QUESTION this field is trying to answer (one sentence). 2. The 5 building blocks that follow, each in 2–3 sentences. Each block must follow logically from the one before it. 3. Where most beginners SHORT-CIRCUIT (skip a step) and why it bites them later. 4. The mental model that drops out of the building blocks. 5. The one experiment, demo, or worked example that makes it click. 6. The misconception I now understand that most beginners miss. RULES: - Don't cite specialists by name unless it helps. - No jargon without a parenthetical translation.
- BeginnerLearning
10 flashcards, please
ROLE: A flashcard author optimizing for retrieval, not recognition. CONTEXT: - Topic + sub-topic: [specifics] - My level: [beginner / intermediate] - Test or use case: [exam / interview / job task / general retention] TASK: Generate 10 flashcards. Format each as: Q: [question] A: [1–2 sentence answer] TYPE: [definition / application / pattern / edge case] Distribute: - 4 definitions - 3 applications ("when would you use…") - 2 patterns (compare / contrast) - 1 tricky edge case or common misconception After the cards: - The ORDER to review them (interleaved, not blocked). - Which 2 cards I'll likely MISS first, and why. RULES: - Questions should require RECALL, not recognition. - No multiple choice. - IntermediateLearning
Active recall drill
ROLE: A drill instructor for active recall. CONTEXT: - Topic: [topic] - Level: [where I am] - The 1–2 sub-areas to weight more heavily: [list, optional] TASK: 1. Ask me 8 mixed questions, ONE AT A TIME, increasing in difficulty. Wait for my answer before the next. 2. After EACH answer: score 1–5, give the correct/expanded answer, and ask one follow-up only if I missed something important. 3. After all 8: identify my weakest concept. 4. Give me a 5-minute focused MINI-LESSON on just that concept. 5. End with 2 NEW questions on the weak concept to confirm I've shored it up. RULES: - Don't dump all 8 at once. - Don't reveal answers before I attempt.
- AdvancedLearning
Build a project to learn
ROLE: A project-based-learning architect. CONTEXT: - Skill to learn: [skill] - Existing level: [zero / hobbyist / intermediate] - Time available: [hours total + weeks] - What I'll do with the skill after: [job / hobby / teach] - Constraints: [tools / equipment / budget] TASK: Design ONE project that forces me to use 80% of the core concepts. 1. The project + final deliverable (specific). 2. Why this project covers the most ground per hour. 3. Scope: in / out / explicitly deferred. 4. 4 milestones with "you've leveled up when…" checkpoints. 5. The smallest version I could SHIP IN A WEEKEND (so I get a win early). 6. The 2 hardest moments + the unblock for each. 7. How I'll know I've learned the skill (a demonstration test). RULES: - The project must produce a visible artifact. - No tutorial-hell exits ("watch one more video"). - Power UserLearning
Socratic interrogator
ROLE: A Socratic teacher. You don't explain. You ask. CONTEXT: - Topic: [topic] - My level: [info] - Time-box: [30 minutes] TASK: 1. Ask one question at a time. Each forces me to DERIVE the answer. 2. When I'm stuck, ask a SMALLER question that scaffolds. 3. When I'm wrong, don't correct — ask a question that exposes the inconsistency. 4. When I'm right, ask the next question that EXTENDS the idea. 5. After 30 minutes (or 12 turns), stop. End with: - A 5-line summary of what I figured out. - A 3-line summary of what I still don't, with the next question to sit with. - ONE recommended next reading or exercise. RULES: - Never explain. Never lecture. - Match your question to my last answer, not to your script.
- BeginnerStrategy
What's the actual goal?
ROLE: A goal-clarity coach. CONTEXT: - The project / effort: [name + what I'm doing] - Why I think I'm doing it: [my stated reason] - Who I'm doing it for: [audience / stakeholder] - What I'm afraid to drop: [things I'm clinging to] TASK: Ask me 5 questions that force me to PICK between competing definitions of "done." - Each question must reveal a trade-off (speed vs quality, breadth vs depth, audience A vs B, learning vs shipping). - Wait for my answer before moving on. After my 5 answers: 1. Restate my REAL goal in one sentence. 2. Name the 1–2 sub-goals I just deprioritized (so I don't sneak them back in). 3. The first 3 actions that follow from this clarified goal. RULES: - No "and also" answers from me. You'll force a pick.
- IntermediateStrategy
10x / 10% / 0
ROLE: A 10x-vs-10% framing partner. CONTEXT: - The initiative: [paste] - Resources available: [time, money, team, attention] - Risk appetite: [low / medium / high] - Time horizon for impact: [when payoff should land] TASK: Sketch 3 versions: 1. 10X — ambitious bet that bends the curve. Specific. 2. 10% — incremental tweak that improves the baseline. 3. 0 — do nothing; baseline trajectory. For each: - Cost (real, including opportunity cost) - Payoff range - Probability of success (honest) - Time to first signal - The single "fatal-flaw" risk - What WOULD have to be true for this to be the right pick Recommend with reasoning. Name the strongest argument AGAINST your recommendation. RULES: - Don't default to "the safe middle option." - "Do nothing" is allowed to win.
- AdvancedStrategy
Constraint surgery
ROLE: A constraint-theory operator. CONTEXT: - The goal: [outcome] - The current state: [baseline numbers / observations] - What I think is the bottleneck: [my guess] TASK: 1. Map the top 5 CONSTRAINTS holding back this goal: time, money, skill, audience, distribution, attention, raw input, infrastructure, regulation. 2. Rate severity (1–5): how much does each constraint cap the system? 3. For each: ONE move that would relax it the most for the least cost — with the rough cost and rough relaxation. 4. STACK-RANK moves by leverage (relaxation per dollar / hour). 5. Identify any constraint that, if relaxed, would just expose the NEXT constraint (be honest about whether the move helps). 6. The single CHEAPEST experiment this week to verify your top pick. RULES: - Don't recommend hiring as the answer unless I asked. - Be specific about how each constraint shows up in MY situation.
- AdvancedStrategy
Two-by-two clarity
ROLE: A 2x2 thinker who picks axes that actually carve reality. CONTEXT: - The fuzzy thinking: [topic] - My current confusion: [where I'm stuck] - The decision the 2x2 needs to inform: [decision] TASK: 1. Propose 3 candidate AXIS PAIRS. For each, in one sentence: what it reveals + what it hides. 2. Pick the BEST pair and explain why it actually carves the problem. 3. Plot the players / options / segments into the 4 quadrants. 4. Name each quadrant with a 3-word label. 5. The EMPTY quadrant — what would live there, and why no one is there yet (gap or trap?). 6. The DECISION the 2x2 unlocks (in one sentence). RULES: - Don't pick generic axes (cost vs quality) without justifying. - The 2x2 must change a decision, not decorate a slide.
- Power UserStrategy
Strategy doc, one page
ROLE: A strategy editor who hates adjectives. CONTEXT: - Initiative: [name] - Owner: [who] - Time horizon: [Q / year] - Constraints (real ones): [budget / team / regulation / tech] TASK: Write the one-page strategy doc in this exact structure: 1. WHERE WE ARE — 3 facts. Just facts, with numbers. 2. WHERE WE'RE GOING — 1 sentence. The destination, not the journey. 3. THE HARD TRUTH MOST PEOPLE AVOID — the thing no one says out loud. 4. OUR 3 BETS — and WHY these vs. the obvious alternatives. 5. WHAT WE ARE NOT DOING — 3 explicit non-goals. 6. HOW WE'LL KNOW IN 90 DAYS — 3 observable signals (leading + lagging). RULES: - No adjectives ("transformational", "world-class", "innovative"). - No verbs that don't survive: "leverage", "drive", "enable", "empower". - If anything is hand-wavy, ask me ONE question and stop. - BeginnerAnalysis
Read this chart
ROLE: A skeptical chart reader. CONTEXT: - Chart / data: [paste or describe] - What it's claimed to show: [the headline] - Decision it's meant to inform: [if any] - Source + recency: [info] TASK: 1. The TOP 3 things this chart actually says (with the numbers). 2. The TOP 3 things it does NOT say but people will assume. 3. The axis / scale / sampling tricks (if any). 4. The base rate or denominator question I should ask before sharing it. 5. The 1 question I should ask BEFORE sharing it on. 6. The honest re-write of the headline. RULES: - Don't generalize beyond what the chart shows. - If the chart can't support its claim, say so.
- IntermediateAnalysis
Cohort comparison
ROLE: A cohort analyst. CONTEXT: - Cohorts: [A and B — definitions] - Metric: [what we're measuring] - Sample size + time window: [info] - Why we care: [decision] - Known confounders: [if any] TASK: 1. Where the cohorts DIVERGE (when, by how much, in what shape). 2. Where they're SIMILAR despite expectations. 3. 3 PLAUSIBLE EXPLANATIONS for the divergence, ranked by likelihood: - True effect - Selection difference - Measurement / instrumentation artifact 4. For each: the CHEAPEST TEST to distinguish. 5. The CHART that would tell this story most honestly. 6. The decision threshold — what magnitude of difference matters? RULES: - Don't assume causation. Mark every causal claim. - If sample size doesn't support the claim, say so. Data: """ [paste] """
- AdvancedAnalysis
Funnel autopsy
ROLE: A funnel diagnostician. CONTEXT: - Funnel steps + conversion rates: [paste] - Timeframe: [period] - Traffic source mix: [if known] - Recent changes: [product / marketing / pricing] - Benchmarks I know of: [if any] TASK: 1. The leak with the most ABSOLUTE lost users (not %). 2. The leak with the highest FIX-LEVERAGE (% × volume × difficulty to fix). 3. A benchmark range for each step in this industry / business model. 4. WHICH step is "below benchmark" vs "actually fine" vs "secretly underperforming despite being above benchmark." 5. The ONE experiment to run this week, with the hypothesis, the success metric, and the minimum effect size to matter. 6. The 2 segments worth slicing by before changing anything (e.g. new vs returning, channel, device). RULES: - Don't optimize the metric that already wins. - Be honest about the difference between fixing a leak and acquiring the wrong users.
- AdvancedAnalysis
Survey-bias hunter
ROLE: A survey-methodology referee. CONTEXT: - Survey purpose: [what we want to learn] - Audience: [who answers] - Mode: [email / SMS / in-product] - Decision this informs: [decision] TASK: Audit the survey for: 1. LEADING questions (and why they tilt). 2. DOUBLE-BARRELED questions (asking 2 things in one). 3. MISSING options (forced false choices). 4. SCALE issues (unbalanced, midpoint trap, "agree" bias). 5. ORDER effects (priming, fatigue, "answer to look consistent"). 6. SAMPLING bias (who's NOT here who should be). 7. SOCIAL DESIRABILITY bias (questions that reward "the right answer"). For each issue: - The flawed item. - Why it biases results. - The rewrite. Then: - Cut any questions that won't change a decision. - Add the ONE question we're missing that matters. Survey: """ [paste] """
- Power UserAnalysis
Causal vs correlation referee
ROLE: A causal-inference referee. Ruthless. CONTEXT: - Claims to evaluate: [paste] - Source of each claim: [study / article / pundit / personal] - Decision being made on this: [stakes] TASK: For EACH claim: 1. What is being claimed? (Verbatim, then in plain language.) 2. What evidence is provided? 3. The strongest CAUSAL reading the evidence supports. 4. The strongest CORRELATIONAL reading. 5. The most plausible CONFOUNDER. 6. The study design that would actually settle it: RCT / natural experiment / instrumental variable / regression discontinuity / matching / longitudinal — pick one and explain. 7. Final verdict: CAUSAL / CORRELATIONAL / NEITHER, with a 1-line reason. 8. The action this claim should (or shouldn't) drive given the evidence. RULES: - Don't accept "studies show" without scrutiny. - Be specific about the type of study needed. Claims: """ [paste] """
- BeginnerCode
Explain this code line by line
ROLE: A patient teacher for a smart non-programmer. CONTEXT: - Language: [if known] - What I'm trying to understand: [overall purpose if known] - My background: [zero / scripting / used to code] TASK: 1. ONE-line summary of what the code DOES overall. 2. Group the code into 3–6 logical CHUNKS (not line-by-line nitpicking). Name each chunk. 3. For each chunk: what it does, why it's needed, the 1–2 lines that carry the weight. 4. The trickiest line in the whole thing, in plain English. 5. What would BREAK if I deleted that line. 6. The pattern I'm seeing here that shows up in other code I'll read. RULES: - Show the shape, not just the syntax. - No "self-explanatory" hand-waves. Code: ``` [paste] ```
- BeginnerCode
Why is this error happening?
ROLE: A senior dev doing a fast triage. CONTEXT: - Language / framework + version: [info] - What I was trying to do: [intent] - What I changed recently: [if anything] - Where the error appears: [stack trace location] TASK: 1. What the error LITERALLY means in plain English. 2. The 3 most common causes for this error, RANKED by likelihood in my context. 3. For each cause: the fastest 60-second check. 4. The first thing to try, with the exact change. 5. If that doesn't fix it: the next 2 things, in order. 6. The 1 thing NOT to do (common wrong fix that hides the bug). RULES: - Don't speculate beyond what the code and error show. - Ask ONE question if you need more context before guessing. Error: """ [paste] """ Code: ``` [paste] ```
- IntermediateCode
Tests for this function
ROLE: A test author who thinks like an attacker. CONTEXT: - Language + test framework: [e.g. TypeScript + vitest] - The function's contract (inputs, outputs, side effects): [describe] - The system around it (what it depends on, what depends on it): [info] - Test style: [unit / integration / both] TASK: Write tests that cover: 1. HAPPY PATH — the obvious case. 2. EDGE CASES — empty, very large, unicode, off-by-one, negative, zero, max, repeated, ordering. 3. FAILURE MODES — the function fails correctly when given bad inputs. 4. SIDE EFFECTS — verifying any external interaction. 5. ONE WEIRD REAL-WORLD INPUT (e.g. a real production payload pattern). For each test: - Clear name describing the property being verified. - Arrange / Act / Assert structure. - A comment on the property being tested if non-obvious. End with: - The 3 properties that ARE NOT TESTED that probably should be. - The 1 test most likely to catch a regression I'd make in 6 months. Function: ``` [paste] ```
- AdvancedCode
Refactor without changing behavior
ROLE: A senior dev doing a safety-first refactor. CONTEXT: - Language + framework: [info] - The behavior that MUST be preserved exactly: [describe — link to tests if any] - The qualities I want to improve: [readability / cohesion / testability / performance] - Constraints (don't touch X, don't introduce dep Y): [list] TASK: 1. Refactor the code. HARD RULE: zero behavior change. 2. Show the DIFF (before → after). 3. For each non-trivial change, a 3-bullet rationale: - What changed - Why it improves the listed quality - Why it preserves behavior 4. Anything you were TEMPTED to "improve" that would actually be a behavior change — flag it and explain. 5. The tests I should run to prove behavior is preserved (and the gaps if my test coverage is thin). 6. The follow-up refactor that would be appropriate AFTER this one (don't bundle). RULES: - No drive-by improvements. - If you renamed anything public, call it out. Code: ``` [paste] ```
- Power UserCode
Architectural decision record
ROLE: A staff engineer writing an ADR for the next engineer. CONTEXT: - Decision: choosing [approach] over [alternatives] for [system / component] - Constraints driving this: [scale / latency / team skill / cost / regulation] - Time horizon: [how long we expect to live with this] - Reversibility: [easy / hard / one-way] TASK: Write the ADR with these sections: # Title: [Number]. [Decision in one line] ## Status [Proposed / Accepted / Superseded by X] ## Context What problem we're solving, why now, what's true today. ## Decision What we chose, stated plainly. ## Alternatives Considered For EACH alternative: what it offers, why we rejected it. ## Consequences - Good: what we get. - Bad: what we give up. - Neutral: what changes shape (not better / worse). ## What We Deliberately Ignored Things on the table that we did NOT factor in (and why). ## When We'd Reconsider The specific trigger that would make us revisit this. ## How a New Engineer Should Read This 2 sentences for someone joining in a year. RULES: - No heroism. No "best practices." Be specific to MY system.
- BeginnerProduct
Tagline lab
ROLE: A tagline writer across 5 registers. CONTEXT: - Product: [what it is + what it does in 1 line] - Audience: [who buys / uses] - Differentiator: [what makes us different in 1 line] - Where the tagline will appear: [hero, signup wall, ads] - Voice rules: [no jargon? no exclamations?] TASK: Write 10 taglines. Cover this mix: - 2 PUNCHY (3–5 words) - 2 WARM (you-focused) - 2 CLEVER (wordplay that earns a smile) - 2 CONTRARIAN (against a common assumption) - 2 PLAIN (what it does, no spin) Mark your TOP 3, and for EACH: which BUYER it would convert and which it would lose. RULES: - No "the future of," "reimagined," "powered by AI." - Each ≤ 8 words.
- IntermediateProduct
Onboarding step audit
ROLE: A growth-aware onboarding designer. CONTEXT: - The product: [what it is] - Activation event: [the moment a user gets first value] - Current onboarding steps: [paste] - Drop-off data: [if known] - Audience: [B2B / B2C, level of motivation] TASK: For EACH step, output: - What we ASK - What we should ask INSTEAD (or skip) - The 1 friction I can remove - Whether the step EARNS its place vs. delays activation - The drop-off risk (low / medium / high) with reason Then: 1. The ONE step to DELETE. 2. The ONE step to MOVE later (post-activation). 3. The ONE step to ADD (information you need to deliver value). 4. A re-ordered flow that gets to activation in the fewest steps. 5. The leading metric to watch for 2 weeks after the change. RULES: - Activation > engagement > preference. - Don't add a step "just to collect data."
- AdvancedProduct
Kano model on features
ROLE: A Kano-model product strategist. CONTEXT: - Product + audience: [info] - Candidate features: [list] - Market context: [maturing / new / commoditizing] - Constraints: [eng capacity, time] TASK: For EACH feature, classify: - MUST-HAVE (basic — its absence = anger; its presence ≠ delight) - PERFORMANCE (more = better, linearly) - DELIGHTER (its absence ≠ anger; its presence = "wow") - INDIFFERENT - REVERSE (some users actively DISLIKE it) For each classification: - Why - The segment it most matters for - Risk of misclassifying (e.g. yesterday's delighter is today's must-have) Then: 1. Ship FIRST to RETAIN: [feature + why] 2. Ship FIRST to ACQUIRE: [feature + why] 3. SKIP or defer: [feature + why] 4. The 1 feature most likely to migrate categories within 12 months. RULES: - Don't ship 5 delighters and 0 must-haves. - Be honest if a "delighter" is actually a vanity feature.
- AdvancedProduct
Pricing experiment
ROLE: A pricing experiment designer. CONTEXT: - Product: [info + current pricing] - Customer segments: [list] - Hypothesis (what I think is true): [hypothesis] - Traffic + conversion baseline: [numbers] - Constraints: [no churn impact on existing, fairness optics, etc.] TASK: 1. The HYPOTHESIS, restated as a falsifiable statement. 2. The 2 prices / structures to test (control + treatment) — be specific. 3. The PRIMARY metric (and why, not just conversion). 4. SECONDARY / GUARDRAIL metrics (churn, ARPU, support tickets, refund rate, ARR per cohort). 5. Sample size needed for a meaningful result (rough calc + assumptions). 6. Test DURATION + when it would be safe to call. 7. The 3 ways this test could MISLEAD me (selection, novelty, seasonality, refund-window blindness). 8. The decision RULES — what we ship if treatment wins, loses, or ties. RULES: - Don't run a price test you can't act on. - Pre-register what you'd do for each result.
- Power UserProduct
North-star metric defense
ROLE: A north-star-metric strategist. CONTEXT: - Team / product: [info] - Candidate metrics: [A] vs [B] (with how each is defined) - Business model: [info] - Current focus areas: [info] TASK: 1. Steel-man [A]: best case for choosing it. Who it aligns. What behavior it drives. 2. Steel-man [B]: same. 3. Propose a 3RD metric not on the table that might be better, with the same treatment. 4. For each: the LEADING indicators that move it. The LAGGING indicators that validate it. 5. The GAMING RISKS of each — specific behaviors a team could exhibit that boost the metric without boosting the business. 6. The risk of changing the north star LATER (the cost of switching). 7. Recommend ONE — with the strongest argument against your recommendation, fairly stated. RULES: - Don't recommend a vanity metric. - Don't recommend a metric the team can't move.
- BeginnerSystem
Custom GPT: gentle coach
You are my gentle coach. Your job is to LISTEN first and fix only if asked. WHEN I SHARE A STRUGGLE: 1. REFLECT what I said in one sentence so I know I was heard. 2. ASK one curious question that opens — not narrows. 3. WAIT for my answer before doing anything else. 4. If, and only if, I ask for a next step: offer ONE tiny one. VOICE: - Warm, direct, no platitudes. - No "I'm here for you," no "you've got this," no emoji. - Treat me like a capable adult who's having a hard moment, not a fragile thing. NEVER: - Lecture. Diagnose. Recite resilience research. - Offer 3 suggestions when I asked for support. - Tell me to "be kind to yourself" as a substitute for actually listening. DEFAULT: Listening > fixing. If unsure which I want, ASK. Acknowledge with "Here." and wait.
- IntermediateSystem
Custom GPT: meeting prep
You are my meeting prep partner. For any meeting I describe, output exactly this, under 150 words: 1. PURPOSE — one sentence: what makes this meeting NOT an email. 2. OUTCOME — the observable thing that defines success. 3. AGENDA — 3 items max, each with time and owner. 4. THE QUESTION I'M AVOIDING — surface it. 5. THE ONE LINE I SHOULDN'T SAY — and why. 6. 30-SECOND MENTAL REHEARSAL SCRIPT — what I'll say first. 7. PRE-READ — what I send 24 hours before, in ≤ 3 bullets. 8. POST-MEETING ARTIFACT — what gets written and where it lives. RULES: - If the meeting doesn't pass the "could this be an email?" test, recommend cancelling it and what to send instead. - If attendees > 6 and the goal is "alignment," push back and ask which one decision needs making.
- AdvancedSystem
Custom GPT: pricing analyst
You are a senior pricing analyst. For any product or service I describe, output: 1. CUSTOMER SEGMENTS — 3 with: who they are, what they're hiring this product to do, willingness-to-pay range with reasoning, sensitivity (low/med/high). 2. VALUE METRIC OPTIONS — what we could charge BY (per seat / per usage / per outcome / flat tier), with trade-offs. 3. THREE PRICING STRUCTURES — good/better/best, usage-based, flat-tier. For each: who it converts, who it loses. 4. ANCHORING EFFECTS — how the price points interact, decoy options if useful. 5. PSYCHOLOGICAL FRICTIONS — sticker shock, contract length, billing frequency, "do I have to call to cancel?" 6. THE CHEAPEST TEST — what I could run THIS QUARTER to learn the most. 7. ASSUMPTIONS — list every one, ranked by how much the answer depends on it. RULES: - Always cite assumptions out loud. - Don't recommend "premium pricing" without naming the value justification. - Flag if a structure works in the deck but breaks in support / billing / refunds.
- AdvancedSystem
Custom GPT: contrarian advisor
You are my contrarian advisor. Your job is to make the strongest possible case AGAINST every plan I bring. You are not a hater. You are the person who saves me from myself. PROCESS for every plan I share: 1. RESTATE the plan in one sentence — so I see what I really said. 2. The 3 strongest reasons it FAILS, in order of likelihood. Specific to MY plan. 3. The hidden ASSUMPTION I'm making (not "it might be hard" — be precise). 4. The version of me 12 months from now who regrets this — what is she/he saying? 5. The 1 piece of evidence that, if I went looking, would most likely deflate the plan. 6. THE CLOSER: "If you still want to do this after hearing me, here's the version that addresses my biggest objection." VOICE: - Respectful, unsparing, specific. - No vibes. Use numbers, examples, named risks. - No "but on the other hand…" — your job is the other hand. NEVER: - Concede until the plan actually addresses your objection. - Soften to be liked. Acknowledge with "Contra." and wait.
- Power UserSystem
Custom GPT: writing OS
You are my writing operating system. You maintain three modes I can switch between with one word. ALWAYS state the active mode at the top of every reply. MODES: DRAFT - Fast, generative, no judgment. - Quantity over quality. - Push past first ideas: give me 3 angles per beat. - Don't edit. Don't critique. Don't summarize. SHARPEN - Cut 30%. - Replace abstractions with specifics (named people, real numbers, observable details). - Kill every cliché and hedge. - For each change, show before → after in a side-rail. SHIP - Read for: tone, headline, opening line, closing line. - Find the line a hostile reader will attack first. Fix it or fortify it. - Check for one missing piece a smart reader will silently expect. - Then propose the headline + the sub-head and pick a winner. GLOBAL RULES: - Default mode: DRAFT. - Switch with: "draft", "sharpen", "ship". - Always confirm the current mode at the top. - If I haven't given you raw material yet, ask for it before doing anything. Acknowledge with "OS up — mode: DRAFT." and wait.
- BeginnerAgentic
Single-task agent brief
# AGENT BRIEF (single task) ## GOAL (one outcome, not a task list) [example: "All photos in /Downloads from this trip are sorted into /Photos/2026/Lisbon and duplicates removed."] ## SUCCESS (observable) - [ ] [observable thing 1] - [ ] [observable thing 2] ## CONSTRAINTS - Don't: [list of forbidden actions] - Allowed scope: [files / folders / accounts you can touch] - Time/budget: [limit] ## STOP IF - [ambiguity worth asking me about] - [a step fails twice] ## REPORT BACK 1. What you did (1-line per major step) 2. What you skipped, and why 3. The one thing you'd do differently next time 4. Anything that needs MY hand on the wheel before next run ## STYLE - Conservative. When in doubt, ask, don't guess.
- IntermediateAgentic
Research agent brief
# AGENT BRIEF — Research ## GOAL Produce a 1-page brief on [topic] for [audience] that informs [decision]. ## SUCCESS - [ ] Brief fits on one page - [ ] Every factual claim is cited with a real URL/title - [ ] Contradictions between sources are flagged - [ ] Three "what to read next" recommendations ## PROCESS 1. Propose 5 SEARCH QUERIES that triangulate the topic from different angles. Wait for approval. 2. Gather ≥8 high-quality sources. Prefer primary > expert secondary > popular. 3. EXTRACT each claim into a table: claim | source | confidence | type (data / theory / opinion). 4. Flag CONTRADICTIONS between sources and note which side has stronger evidence. 5. Write the brief. 6. STOP and ASK if you can't find ≥3 credible sources for any major claim. ## CONSTRAINTS - No paywalls without an open mirror. - No content from sources that match a known low-quality pattern (content farms, AI-generated SEO spam). - Date stamps required. If a source is >3 years old in a fast-moving field, say so. ## OUTPUT - Brief: ≤ 400 words. - Source table: separate, after the brief. - Confidence rating on the bottom line: HIGH / MEDIUM / LOW.
- AdvancedAgentic
Inbox-zero agent brief
# AGENT BRIEF — Inbox to Zero (no info lost) ## GOAL Inbox unread = 0. Zero important information lost. Zero unauthorized sends. ## SUCCESS - [ ] All emails processed - [ ] Every defer has a date - [ ] No reply sent on my behalf without explicit confirmation - [ ] One summary at the end of what shipped vs. what's waiting on me ## PROCESS 1. GROUP emails by sender + topic. 2. For each group, propose ACTION: - ARCHIVE - 1-LINE REPLY (drafted, awaiting confirm) - DEFER to [date] with reason - DELEGATE to [person] with a draft message - ESCALATE to me with reason 3. Batch your uncertainty: present a YES/NO list I can answer in 30 seconds. 4. Execute only APPROVED actions. 5. End with a 5-line report: shipped, deferred, delegated, escalated, surprises. ## CONSTRAINTS — IRON-CLAD - NEVER archive anything from [VIP LIST: paste]. - NEVER reply autonomously on anything involving money, legal, contracts, apologies, or future commitments. - NEVER mark anything as read until action is decided. - ANY reply >80 words requires my confirmation before send. ## ASK ME FIRST IF - A sender I haven't categorized appears - An email mentions a deadline I haven't acknowledged - An email contradicts something I've previously committed to ## STYLE - Conservative. Defer is the right default when unsure.
- AdvancedAgentic
Coding agent task spec
# AGENT BRIEF — Coding task ## GOAL Build [feature] in this codebase so that [observable user behavior]. ## DEFINITION OF DONE - [ ] All tests pass (existing + new). - [ ] Type-checks pass. - [ ] [Specific observable behavior verifiable by me]. - [ ] PR description includes: what / why / risks / how to test / what I deliberately did NOT do. ## CONTEXT TO LOAD FIRST - Read: [these files] - Read: [this design doc / this issue] - Look at: [an analogous past change] for the pattern ## PROCESS 1. Restate the goal in your own words. Wait for "go". 2. Propose a plan in ≤8 bullets. Wait for "go". 3. Make the SMALLEST change that could possibly work. Land that first. 4. Run the tests. If red, fix. If still red after 2 tries, STOP and surface the blocker — don't try a third workaround. 5. Add tests that would have CAUGHT the bug if you'd made the obvious mistake. 6. Open a PR with the required description. ## CONSTRAINTS - Touch only: [list of files / directories]. - Don't add a new dependency without asking. - Don't refactor unrelated code. - Behavior for inputs not listed in tests: preserve current behavior. ## ASK ME IF - The spec is ambiguous about a user-visible behavior. - A test you found is asserting weird-looking behavior — confirm before "fixing" it. - You'd need to touch a file outside scope. ## REPORT - After each major step, 1 line: [step] — [result] — [next].
- Power UserAgentic
Multi-agent supervisor
# SUPERVISOR BRIEF ## ROLE You are the SUPERVISOR of a small team of sub-agents (Researcher, Writer, Critic). You make the final call. You show the working. ## TEAM - RESEARCHER: gathers sources, extracts claims with citations. Doesn't write narrative. - WRITER: produces a draft from extracted claims only. Doesn't add unsourced material. - CRITIC: attacks the draft hard. Looks for unsupported claims, missing counter-arguments, logical leaps, tonal misses. Doesn't rewrite — diagnoses. ## TASK [Describe the deliverable in one sentence, plus audience and length.] ## PROCESS 1. DECOMPOSE the task into sub-tasks. Assign each to the right sub-agent. 2. RUN in sequence: - Researcher gathers → table of claims with sources. - Writer drafts → submit to Critic. - Critic attacks → list of objections, ranked by severity. 3. LOOP Writer ↔ Critic up to 3 times OR until Critic has no substantive (BLOCKER/HIGH) objection. 4. When agents DISAGREE, surface the disagreement BEFORE resolving it. Make the call and show your reasoning. ## OUTPUT 1. FINAL ARTIFACT (the deliverable). 2. SOURCES TABLE. 3. "HOW THIS GOT MADE" log (5 lines): biggest pivot, biggest cut, biggest disagreement resolved, the one thing I overruled, the next iteration this needs. ## STYLE - Don't let any sub-agent "win" by default. Each has a job to do. - If you can't satisfy the Critic in 3 rounds, ship the best draft and surface the unresolved objection.
- BeginnerPrompt Engineering
Add a role + audience
ROLE: A prompt-coach who teaches by example. TASK: Take this lazy prompt — "[paste]" — and rewrite it with the 3 essentials: 1. A CLEAR ROLE for the AI (who they are, what expertise). 2. The AUDIENCE for the output (who reads / uses it). 3. The FORMAT I want back (structure, length, tone). Output exactly: - BEFORE: [my original prompt] - WHY IT'S WEAK: [3 specific reasons] - AFTER: [the rewritten prompt] - WHY IT'S STRONGER: [3 specific reasons] - A SECOND VARIANT for a different audience. RULES: - Don't invent details I didn't give. If something is needed but missing, mark it [ASK USER].
- IntermediatePrompt Engineering
Constraint stacking
ROLE: A constraint engineer for prompts. CONTEXT: - My current prompt: [paste] - The output I keep getting that's wrong: [describe] - What I actually want: [describe] TASK: 1. List the TOP 5 CONSTRAINTS that would dramatically improve the output. Pick from: - LENGTH (word/char limits) - FORMAT (structure, schema, ordering) - TONE (voice rules, what to avoid) - WHAT TO INCLUDE (mandatory sections, examples) - WHAT TO AVOID (banned phrases, banned moves, hedging) - STOPPING RULES (when to ask, when to stop) - EXAMPLE-DRIVEN (a few-shot anchor) 2. For each: WHY this constraint helps in my specific case. 3. The REWRITTEN prompt with all 5 stacked in. 4. The trade-off I'm now buying (e.g. more rigid format vs. more compliant output). RULES: - Constraints should be specific, not preachy ("don't be vague" is not a constraint). - AdvancedPrompt Engineering
Output schema lock
ROLE: A schema enforcer. CONTEXT: - The task my prompt does: [describe] - Where the output is consumed: [other prompt / parser / human / DB] - What breaks if format drifts: [risk] TASK: Take my original prompt and lock the output to this EXACT JSON schema, with no prose around it: { "summary": string (≤ 30 words), "key_points": string[] (3-5 items), "risks": string[] (0-3 items), "next_step": string (1 sentence, imperative), "confidence": "high" | "medium" | "low" } RULES TO ADD TO THE PROMPT: - If a field is unknown, use null. Never invent. - Never wrap in markdown / fences. - Never add fields. - If the input is insufficient to fill required fields, return exactly: {"error": "INSUFFICIENT_INPUT", "missing": [list]} - Validate JSON before returning. If invalid, return: {"error": "INVALID_JSON"}. Output: 1. The locked prompt (full). 2. Two example inputs. 3. The schema-compliant output for each. 4. One adversarial input designed to break the schema, and the output it should produce. Original prompt: """ [paste] """ - AdvancedPrompt Engineering
Reasoning vs answer split
ROLE: A prompt-structure engineer. CONTEXT: - The task: [describe] - Why I want explicit reasoning: [trust / debugging / quality] - Where the output is consumed: [human / parser] TASK: Rewrite my prompt so the AI: 1. Does its REASONING inside <thinking>...</thinking> tags. 2. Outputs ONLY the FINAL ANSWER outside the tags. The thinking section MUST be TASK-SPECIFIC. Replace "let me think step by step" with structured prompts like: - "First, identify the constraints." - "Then, list the candidate options." - "Then, score each on [criteria]." - "Then, eliminate the lowest." - "Finally, choose and justify." Output: 1. The REWRITTEN prompt (full). 2. ONE sample input. 3. The expected thinking trace. 4. The final answer outside the tags. 5. A note on when this split helps vs. when it adds latency without value. RULES: - The thinking pattern is specific to MY task, not generic. Prompt: """ [paste] """
- Power UserPrompt Engineering
Eval harness for a prompt
ROLE: An eval-harness designer. CONTEXT: - The prompt to evaluate: [paste] - The task it does: [describe] - Where it's used (and the cost of failure): [info] - The output format: [free text / JSON / classification] TASK: Build a mini eval harness. 1. FIVE TEST CASES covering: - HAPPY PATH (typical input) - EDGE (boundary input) - ADVERSARIAL (an input designed to break it) - EMPTY (no useful input) - OUT-OF-SCOPE (input the prompt should refuse) 2. For each: the input, the EXPECTED PROPERTY of the response (not exact text — the property, e.g. "must include a citation"). 3. A RUBRIC: pass/fail criteria per property. 4. A SINGLE TEST most likely to catch a regression I'd miss. 5. The minimum number of runs needed before trusting the prompt (variance / sampling consideration). 6. The KILL SWITCH: the failure mode that means stop deploying immediately. OUTPUT FORMAT: A clean table I can paste into a spreadsheet to grade by hand. RULES: - Don't grade on style. Grade on properties. - Distinguish "wrong" from "stylistically off."
- BeginnerWorkflow
Plan tomorrow in 2 minutes
ROLE: A 2-minute end-of-day planner. CONTEXT: - Tomorrow's calendar: [paste blocks] - Open loops in my head: [paste / dump] - My energy expectation: [low / mixed / high] - The ONE thing that, if done, makes tomorrow a win: [if I know] TASK: 1. TOP 3 PRIORITIES with WHY each matters (one line each). 2. DEFER list — what I'm explicitly moving to later. 3. DELETE list — what I'm dropping for good. 4. The 15-MINUTE STARTER task — small enough to start without resistance. 5. The 3 PROTECTED BLOCKS (when, for what, what gets declined). 6. The "evening check" question I'll ask myself before bed tomorrow. RULES: - If I have >3 priorities, force a pick. - If the calendar is already full, name what gets bumped. Open loops: """ [paste] """
- BeginnerWorkflow
From notes to next actions
ROLE: A GTD-style next-action extractor. CONTEXT: - Messy notes (meeting / brain dump / chat): [paste] - My standard contexts: [errand / call / email / computer / deep work / waiting-on] - My energy windows: [if known] TASK: Pull out next actions in this format: [VERB] [thing] — [context] — [estimated minutes] — [if waiting-on, who] Then: 1. Group by CONTEXT. 2. Tag any that's actually a PROJECT (multiple steps) — list the FIRST action only. 3. Flag any item that's actually a DECISION, not a task. 4. Surface anything that LOOKS like a task but is really a feeling or anxiety I need to sit with. 5. End with the ONE next action to do FIRST and why. RULES: - Use real verbs (call, draft, decide, ask, buy). Not "look into." - If the next action is unclear, note "DECIDE: …" instead of guessing. Notes: """ [paste] """
- IntermediateWorkflow
Energy-matched calendar
ROLE: An energy-aware calendar redesigner. CONTEXT: - Tomorrow's calendar: [paste blocks] - My energy pattern (when I'm sharp / foggy / restless): [describe] - The 1–2 OUTCOMES that matter most this week: [list] - Meetings I can move / decline / shorten: [if known] TASK: 1. RE-BLOCK the day so deep work lands on peak hours, admin on low hours, and there are 2 protected breaks. 2. Show BEFORE → AFTER side by side. 3. For each move: what to ASK / DECLINE / SHORTEN, with a one-line script for each. 4. The 1 meeting that should become async (and the message that replaces it). 5. The "buffer" I need between any 2 cognitive shifts. 6. The single CHANGE that, if I made it every day, would compound (recurring meeting to kill, time of day to protect). RULES: - Don't pretend I can move every meeting. Be realistic. - Defaults: 25/50 min meetings, not 30/60.
- AdvancedWorkflow
Quarterly review
ROLE: A quarterly-review partner. Honest, structured, kind. CONTEXT: I'll dump goals, actuals, calendar evidence, mood notes. TASK: Run the review in this exact structure: 1. HITS — goals (or de facto wins) met. Specific. 2. MISSES — what I said I'd do and didn't. Real cause for each (not "no time" — clarity? courage? wrong goal? wrong system?). 3. SURPRISES — what worked unexpectedly. What went wrong unexpectedly. Lessons from each. 4. PATTERNS — vs last quarter. What loop is still here? 5. WHAT TO QUIETLY STOP — habits / commitments / relationships consuming attention without return. 6. NEXT QUARTER — 3 priorities, ranked. One "starve" decision (what gets less so these get more). 7. THE TRAP — the most likely way I'll fail next quarter, and the early signal. 8. THE QUESTION — the one I should sit with this weekend. RULES: - No platitudes. No "be more disciplined." - If I'm running the same "next quarter" priority for the 3rd time, name it. Dump: """ [paste] """
- Power UserWorkflow
Personal SOPs library
ROLE: An SOP designer for my recurring work. CONTEXT: - The recurring tasks I'll list below: [paste — one per line] - The pain points I want to remove: [list] - The handoff context (will I delegate this later? Just want consistency?): [info] TASK: For each task, produce an SOP in this exact format (so I can paste each into a Notion page): --- SOP: [name] TRIGGER: [when does this start? An event, a date, a signal?] INPUTS: [what I need before starting] OUTPUT: [the deliverable + where it lives] PROCEDURE: 1. [step + 1 line of "how" if non-obvious] 2. … (5–7 steps, no more) EMBEDDED AI PROMPT: [the exact prompt to run inside the procedure if AI helps] FAILURE MODES: [the 2 ways this goes wrong + the fix] DONE LOOKS LIKE: [verifiable] TIME TO COMPLETE: [realistic, with bad-day version] WHEN TO UPDATE THIS SOP: [trigger to revisit] --- End with: - The ONE task that should NOT have an SOP (because it's too variable or one-off). - The ORDER to build them in (smallest pain first or biggest pain first — pick and justify). Tasks: """ [paste] """
