# The 10 skills as paste-in prompts (ChatGPT, Gemini, or Claude without skills)

No skills support? No problem. Each block below is the skill compressed into one prompt. Paste it, fill the `[BRACKETS]`, then paste your material underneath.

The trade-off: a skill loads itself when it's needed and remembers its files. A pasted prompt you have to paste every time. Tip: save your 3 favorites as a ChatGPT Custom GPT, a Gemini Gem, or a Claude Project's instructions so you paste them once.

Last verified: October 2026.

---

## 1. Voice writer

```
I'm going to give you 3-5 samples of my own writing. First, analyze them and build a short voice profile (max 40 lines): average sentence length, how often I use very short sentences, how I open and close, words and phrases I repeat, contractions, punctuation habits (dashes, exclamation marks, emojis), how confident or hedging I sound, humor, and a "never-list" of words and moves that would sound fake coming from me. Quote a real phrase from my samples as proof for each trait; don't invent traits. Then write a 60-word test paragraph in my voice and ask me what's off. After I confirm, write [WHAT: e.g., a LinkedIn post about X] for [AUDIENCE] in my voice, [LENGTH]. Use only facts I give you; where a personal story would help, write [STORY: ...] instead of inventing one. Before showing the draft, check it against my sentence length, my never-list and these banned words: delve, unlock, game-changer, elevate, seamless, "in today's world", "it's not just X, it's Y". Show only the draft, then one line of notes. My samples: [PASTE 3-5 SAMPLES, LABEL EACH WITH ITS CHANNEL]
```

## 2. Anti-generic strategy

```
Act as a blunt strategy editor. I'll give you a [strategy doc / deck / plan]. Make every claim specific to my business. Before writing, ask me (in one message, max 6 questions) for anything missing from: exact customer, what they use today instead of me, why customers pick me (real quotes), the numbers I have, what I will NOT do, and my main constraint. Then rewrite with these rules: (1) swap test: any sentence a competitor could use unchanged must be rewritten with a fact only I have, or cut; (2) every goal has a number and a date; (3) every pillar states what we give up, as "We will X. This means we won't Y"; (4) replace adjectives (innovative, seamless, best-in-class, robust, leading, scalable, customer-centric) with how it actually works; (5) name the alternative we beat and why it fails; (6) list 3 risks with an early warning signal each; (7) tag important claims [data: source], [customer calls] or [assumption]. Never invent quotes or metrics; write [UNKNOWN: how to find out]. Structure: one-sentence bet, where we are, who it's for/not for, alternative, 3 choices, 3 goals, next 30 days, risks, open questions. End with a short specificity report and the 5 biggest before → after rewrites. My material: [PASTE]
```

## 3. Research brief

```
Research this question so I can make a decision: [QUESTION]. The decision it informs: [DECISION]. Scope: [REGION / BUDGET / TIME WINDOW / MUST-HAVES]. Depth: [quick, about 5 sources / deep, 10-20 sources]. Split it into 3-6 sub-questions, then search. Prefer primary sources (official pages, pricing pages, original studies, regulators) over blog summaries; use forums and reviews only for real user experience. Note the date you saw any price, limit or rule. If you can't browse, say so at the top and label everything "from training data, may be outdated". Output: Bottom line (2-4 sentences, answer first); up to 7 key findings, each with confidence high/medium/low and a source number; a comparison table if comparing options; what's uncertain and how to resolve it; the one next step; a numbered source list with title, publisher, URL, quality (A official, B reputable, C opinion) and date seen. Never invent a source, URL or statistic; write "not found". Under 600 words for quick, excluding sources.
```

## 4. Meeting to actions

```
Here's a [transcript / my notes] from a [client call / internal meeting / sales call] on [DATE]. I am [MY NAME]. Read the whole thing before writing, including the last third. Extract: (1) Decisions: only things actually agreed, with a short quote; suggestions nobody confirmed go to open questions. (2) Action items as a table: action starting with a verb and specific enough for someone else to do, owner (write UNASSIGNED if nobody took it), due date (write "not set" plus a suggestion if none was said), source quote. (3) Open questions and who must answer. (4) Numbers and commitments (prices, dates, promises). (5) Watch-outs with evidence, factual only. Start with a 2-3 sentence TL;DR. Then draft: a recap email to external participants (subject + under 150 words, decisions, their actions, my actions, next meeting), a one-line-per-person internal message, and a suggested next meeting agenda in 3 bullets. Don't invent names, dates or prices. If speaker labels are unclear, say what you inferred. Notes: [PASTE]
```

## 5. Inbox triage

```
Triage these [N] emails/messages. Sort each into: 1 Reply now (needs me, under 5 minutes), 2 Deep work (needs me, more than 5 minutes), 3 Delegate (someone else should handle it; my delegates: [NAMES OR "none"]), 4 Read later (useful, no action), 5 Archive (no value now). Move up: real people, me in "To", deadlines within 72 hours, paying customers, second follow-ups, invoices and payments from known contacts. My rules: always urgent [SENDERS], always archive [SENDERS/TYPES]. Flag anything asking for passwords, gift cards or changed bank details as possible phishing and don't draft a compliant reply. For each "Reply now", draft a reply under 80 words that answers the question in the first sentence, in a [warm, direct] tone, with [BRACKETS] for any fact or commitment you don't know; never guess dates or prices. For Deep work give a one-line summary and the decision needed; for Delegate give a one-line handoff note. End with estimated minutes to clear and 1-2 filter ideas for recurring noise. Don't repeat full email bodies. Messages: [PASTE]
```

## 6. Weekly review

```
Run my weekly review as an honest chief of staff. First, ask me these 6 questions in one message and wait: what I got done; what I planned but didn't do (my top 3 last week were: [TOP 3 OR "none"]); what took more time or energy than it should; any numbers that matter; what's coming next week; anything on my mind not written down. After I answer: mark last week's top 3 done/partial/not done with the reason (underestimated, interrupted, avoided, deprioritized, blocked); call out anything that slipped 2+ weeks; turn worries into an action, a scheduled slot, or "let go"; pick next week's top 3 as outcomes not activities ("Send proposal to Acme", not "work on proposal"), at least one tied to my goal [GOAL], sized to my real calendar; anything that slipped twice is either in the top 3 or explicitly dropped. Output one page: scorecard, wins, what got in the way, stop/start/keep, next week's top 3 with a day for each, up to 5 "also", a "not doing" list, and calendar block suggestions. Under 400 words. Don't invent numbers.
```

## 7. Decision memo

```
Help me decide: [SHOULD I A OR B?]. Include "wait / do nothing" if realistic. First classify it: reversible (decide fast, ~70% confidence is enough) or hard to reverse (slow down, find the 1-2 facts that would change the answer). Ask me up to 5 questions in one message: deadline, budget/time I can commit, my criteria ranked (money, time, risk, learning, reputation, happiness, other), what I've ruled out, and which I'd regret more. Then for each option give: best realistic case, worst realistic case, cost in money and time, how to undo it, and "what you'd need to believe" for it to be right. Score options 1-5 against my weighted criteria in a table, and say the score is a thinking aid. Run a pre-mortem on the recommended option and a 10/10/10 check. Output one page: recommendation with confidence and "would change if...", options table, score table, pre-mortem, a kill criterion ("If [signal] by [date], we stop/switch"), and the next 3 steps. If options are within ~10%, say it's close and pick the more reversible one. For legal, tax, medical or big financial risks, remind me to check with a professional.
```

## 8. Content repurposer

```
Turn the piece below into 10 platform-native posts. First mine it for an angle bank of 10-15 items tagged [claim], [number], [story], [how-to], [list], [quote], [contrarian], using only what's in the source; never invent numbers, stories or quotes. Then write one post per angle (no angle twice): 1 X single post (under 280 characters), 2 X thread of 5-8 posts (each under 280), 3 LinkedIn post (hook line alone, short paragraphs, 120-250 words, one question or CTA, max 3 hashtags), 4 LinkedIn carousel outline (6-8 slides, max ~25 words each), 5 Instagram carousel (slide 1 max 8 words) + caption (under 2,200 characters), 6 a 30-45 second short video script with [0-3s] hook, body, payoff and CTA (~2.5 words per second), 7 Threads post (under 500 characters, ends with a question), 8 newsletter blurb (subject, preview, 80-150 words, [LINK]), 9 a quote-card line under 15 words, 10 a contrarian take backed by the source. Every hook starts from the reader's problem or desire, not "I wrote an article". One CTA max per post: [CTA]. Voice: [DESCRIBE OR PASTE VOICE PROFILE]. Add character counts for X and Threads, then a 7-10 day posting plan. Source: [PASTE]
```

## 9. Code reviewer

```
Review this [diff / PR / file] like a senior engineer who'll be on call for it. The change is meant to: [INTENT, or "infer it and tell me"]. Stack: [STACK]. Check correctness (conditions, null/empty, async errors, race conditions, dates, money as floats, edge cases), security (auth AND authorization on new endpoints, injection in SQL/shell/HTML, secrets in code or logs, webhook signatures, input validation), data (migrations, missing indexes, N+1, unbounded queries, transactions), failure modes (timeouts, retries, swallowed errors), and tests (does a test fail without this change?). Skip pure style. Rank findings: Blocker (breaks prod, loses data, security hole), Should fix, Consider. For each: title, file:line, what's wrong, the concrete failure scenario, and a fix as a diff. Then: tests to add, areas checked and fine, and a verdict (ready / merge after fixes / needs rework). Precision over volume: if unsure, say "possible issue, verify by..." Never invent library behavior. Code: [PASTE]
```

## 10. Prompt sharpener

```
Improve this prompt before running it: [PASTE YOUR PROMPT]. Check it for 7 parts and mark each present, implied or missing: goal and what good looks like, audience, context the model can't guess, the specific task, inputs clearly separated from instructions, output format, quality bar (examples, checks). If 3+ are missing, ask me up to 4 questions in one message, each with a default I can accept by saying "defaults". Then rewrite it with sections: role and goal, context (give the reason behind each constraint), inputs (long material first, in tags like <document>), task (numbered steps if multi-stage), output format (say what to do, not just what to avoid), and a short self-check. Keep my intent. Keep it as short as possible. No fake pressure or roles. Show: the diagnosis, the sharpened prompt in a code block, and 3-5 bullet lessons on what changed and why. Then ask: run it now or save it?
```