# Self Engine Setup — Teach Your AI Who You Are > Hi Claude. The person who handed you this file already built their Second Brain (the Strategy Engine). Now we teach you who they are — their voice, values, story, goals. This is their Self Engine: a digital twin of how they think. > > Read this whole file first. Don't start until they say "go." > > This setup saves its progress as it goes. They can stop any time, come back later in a new session, say "continue," and you'll pick up exactly where they left off — no need to paste anything again. --- ## A quick word before we start (read this out, plainly) This is the second of three engines: - **Strategy Engine (Stage 1)** — your Second Brain. Done. - **Self Engine (Stage 2)** — a digital twin of how you think. We build this now. (This file.) - **Systems Engine (Stage 3)** — the AI workflows you add later. Here's the idea: an AI is only as useful as what it knows about you. Right now I know the shape of your brain, but not *you*. So I'm going to ask you 15 questions — the kind that get at your story, your values, and where you're headed. Then I'll turn your answers into a "twin": a set of notes I read whenever I help you, so everything I write actually sounds like you and fits your life. Two things to know: - You don't have to finish in one sitting. I save after every question. Stop any time. - At the very end there's an optional step to feed me more of your data (old chats, notes, recordings). That makes your twin sharper — but it's optional, and you can do it later. And any time a question feels unclear, just ask me to rephrase it or talk it through — there are no wrong answers here, and I'd rather check than assume. --- ## A note on the heavy thinking step (for the person, quickly) When I pull everything together at the end, that's a big thinking task. **If you're on the $20/month Claude plan, switch me to Sonnet for that step** — it handles long reading well and won't burn through your usage. I'll remind you when we get there. --- ## PHASE 0 — Quick checks before we begin 1. Confirm we're in their Second Brain. Check these exist: - `CLAUDE.md` at the root - `00. System/`, `01. Inbox/`, `02. Projects/`, `03. Areas/`, `04. Resources/`, `05. Archive/` If they're missing, STOP and say: "I can't find your Second Brain here. Run the Strategy Engine Setup first, then come back to this one." 2. Check `03. Areas/Self/digital-twin/`. If it exists, this is a re-run — Phase 3 will update it instead of creating fresh. 3. Check `02. Projects/Digital-Twin/`. If it exists with a `progress.md`, this is a RESUME — read it and offer to continue from the last finished step. 4. Look in `01. Inbox/` for any uploaded chat export (a folder with a `conversations.json` file inside). If one's there, note it — the optional Phase 4 can use it. Don't act on it yet. 5. Report: > "Self Engine — quick check: > - Second Brain: found > - Existing twin: {none, will create fresh / found, will update} > - Saved progress: {none, fresh start / found — resume from step X} > - Uploaded chat history: {none yet / found one, we can use it in the optional step at the end} > > Plan: I'll ask you 15 questions, saving after each, then build your twin. There's an optional 'feed me your data' step at the end you can skip or do later. > > Say 'go' to start (or 'continue' if you're resuming)." 6. Wait for "go" or "continue." --- ## PHASE 1 — Set up the build + turn on resume Print: > "Phase 1 — setting things up so you can stop and resume any time..." ### Step 1.1 — Create the build folder Create `02. Projects/Digital-Twin/` with: ``` 02. Projects/Digital-Twin/ ├── README.md (what this is) ├── progress.md (tracks where we are — this is what lets you resume) └── questions/ (one file per answered question — filled in Phase 2) ``` **`02. Projects/Digital-Twin/README.md`:** ```markdown # Project — Digital Twin This is where my Digital Twin gets built and updated. The twin itself lives at `03. Areas/Self/digital-twin/`. This folder is just the workshop bench. - Started: {TODAY} - To refresh the twin every 6-12 months, run the Self Engine Setup again. It'll update what's here. ``` **`02. Projects/Digital-Twin/progress.md`:** ```markdown --- status: in-progress started: {TODAY} last_updated: {TODAY} steps: phase_1_setup: complete phase_2_questions: in-progress phase_3_synthesis: pending phase_4_feed_data: optional-pending questions: total: 15 answered: 0 last_answered: null --- # Digital Twin — Progress To resume: read this file, find the first step marked `in-progress` or `pending`, continue from there. ``` ### Step 1.2 — Turn on resume in the root manual Open the root `CLAUDE.md`. If it doesn't already have a `## Resuming a setup` section, add this at the end: ```markdown --- ## Resuming a setup If I say "continue," "resume," or "where did I leave off" — check `02. Projects/Digital-Twin/progress.md`. Find the first step marked `pending` or `in-progress` and carry on from there. No need for me to paste anything again. ``` ### Step 1.3 — Make the empty twin folder Create `03. Areas/Self/digital-twin/` with these empty folders (Phase 3 fills them): ``` 03. Areas/Self/digital-twin/ ├── README.md (placeholder for now) ├── 1-voice/ ├── 2-values/ ├── 3-mental-models/ ├── 4-goals-and-vision/ ├── 5-story/ ├── 6-people-and-influences/ └── 7-fears-and-open-loops/ ``` Place a one-line placeholder `README.md` in each ("Filled in when the twin is built."). Phase 3 overwrites these — don't create a separate placeholder file alongside them. Update `progress.md`: `phase_1_setup: complete`, `phase_2_questions: in-progress`. Print: > "Phase 1 done. You can stop any time from here and pick up later by saying 'continue.' Let's start the questions." --- ## PHASE 2 — The 15 Questions Print: > "Phase 2 — 15 questions. Take them at your pace. After each one I save your answer, so you can stop whenever. Want to speak instead of type? If you've got Wispr Flow or dictation on, just talk — it's faster and sounds more like you. > > Say 'start' for question 1." ### How to run each question For every question, in order: **A. Check what you already know.** Search the brain for anything relevant to this question — look in `03. Areas/Self/` and anywhere chat-history notes may have landed. Only count their own words (`#source/user-statement`, `#source/user-belief`, `#source/calibration`). Ignore anything the AI said (`#source/model-statement`). **B. Offer a head start if you found something.** If you found relevant notes, draft a short 2-4 sentence answer from them and show it so they can just confirm or tweak. If you found nothing (normal on a first run), just ask the question fresh. **C. Ask the question** in this shape: ``` ───────────────────────────────────────────── Question {N} of 15 {question text} ───────────────────────────────────────────── {If you found something:} Here's what I already gather from your notes: > {short draft answer} Options: confirm / tweak it / replace it / skip / stop {If you found nothing:} Options: answer / skip / stop ``` **D. Capture their answer.** - confirm → their answer is the draft. - tweak → ask "what would you add or change?" then combine. - replace → ask "what's the real answer?" then use theirs. - answer → use what they type or say. - skip → mark it skipped, move on. - stop → save and stop. Don't ask the next one. **Push gently past surface answers.** If an answer is vague or cliché ("I value freedom"), ask once or twice more, then accept whatever they give: - "Can you give me a real example?" - "Walk me through what actually happened — what did you do or say?" - "What did it cost you, or what did you feel?" Don't grill them more than twice. The point is warmth, not interrogation. **E. Save before moving on (every time, no exceptions).** 1. Write a question file at `02. Projects/Digital-Twin/questions/q{N}-{short-name}.md`: ```markdown --- question: {N} status: answered | skipped saved: {TODAY} --- # Question {N} {question text} ## Their answer {final answer} ``` 2. Update `progress.md`: add 1 to `answered`, set `last_answered: {N}`, update `last_updated`. If a save fails, STOP and tell them — don't carry on half-saved. **F. Next question.** Loop to A with the next one. If they say "stop": "Saved. Question {N} is locked. To pick up later, open Claude here and say 'continue' — I'll start at question {next}." ### The 15 Questions 1. Quick intro to start — your name, what you do, and who you're working to become. 2. Your story in a few chapters — including the hardest or most challenging stretch you've lived through, and what it taught you. 3. Forget your title for a second — who are you, in three sentences? 4. What are you chasing right now? Your real goals for this year. 5. Zoom out — what's the bigger vision? Paint your ideal day 10 years from now: who's there, what you're doing, how it feels. 6. What problem are you uniquely positioned to solve — and why you? 7. What comes easy to you but seems hard to most people? 8. What are 3 to 5 things you refuse to compromise on, even when it costs you? 9. What genuinely makes you angry or pisses you off? (What makes us angry says a lot about what we value.) 10. What's a belief you hold that most smart people would disagree with — and you'd still defend? 11. The 2 to 3 people you look up to most — who are they, why them, and what did each teach you? 12. What are your biggest fears and concerns right now? 13. What are you avoiding or putting off — something you know you should face but haven't? 14. When are you most alive? What conditions bring out your best work? 15. The frameworks or questions you reach for when a decision is hard — how do you actually think it through? ### When the questions are done When all 15 are answered or skipped (or they say "build it now" early): > "That's the 15. Now I'll pull it all together into your twin." Update `progress.md`: `phase_2_questions: complete`, `phase_3_synthesis: in-progress`. --- ## PHASE 3 — Build the twin Print: > "Phase 3 — building your twin now. > > Reminder: this is the heavy thinking step. If you're on the $20/month plan, switch me to Sonnet first (it reads long things well). Tell me when you're ready." ### What to do Read all 15 question files in `02. Projects/Digital-Twin/questions/`, plus any of their own notes already in the brain (their words only — ignore anything the AI said). Then write their twin into `03. Areas/Self/digital-twin/`, sorting what you learned into these folders: - `1-voice/` — how they talk and write; signature phrases; tone - `2-values/` — what they refuse to compromise on; what makes them angry; what they stand for - `3-mental-models/` — how they think and decide; the frameworks they reach for; contrarian beliefs - `4-goals-and-vision/` — what they're chasing now; the 10-year picture; their ideal day - `5-story/` — origin, hardest chapters, who they are beyond the title, what they've learned - `6-people-and-influences/` — who they look up to and what each taught them - `7-fears-and-open-loops/` — fears and concerns; what they're avoiding; unresolved questions **Write it in this order** (so if you're interrupted, what's done is still useful): 1. **Theme notes first.** Inside each folder, write one short note per clear theme. Each note: a plain-language summary of the pattern, where it shows up, and any exact phrases they used. Link related notes with `[[double brackets]]` — at least 2 links per note. 2. **A summary note per folder.** A `README.md` in each folder listing its theme notes and pointing to related themes in other folders. 3. **The master map LAST** — `03. Areas/Self/digital-twin/README.md`: ```markdown --- type: digital-twin built: {TODAY} based_on: {N} calibration answers{ + any imported data} --- # {Name} — Digital Twin > {One tight paragraph: who this person is, that the AI can read and instantly "get" them.} ## The big threads {3 to 5 patterns that show up across several areas — e.g. a value that drives their goals and their voice. This is what makes the twin feel like a person.} ## Map - [[1-voice/README|Voice]] — {one line} - [[2-values/README|Values]] — {one line} - [[3-mental-models/README|How they think]] — {one line} - [[4-goals-and-vision/README|Goals & vision]] — {one line} - [[5-story/README|Story]] — {one line} - [[6-people-and-influences/README|People who shaped them]] — {one line} - [[7-fears-and-open-loops/README|Fears & open loops]] — {one line} ## How to use this When I help with anything in their voice or about their direction, I read this first, then open the folder that fits. For the sharpest match, they can point me at a specific note: "write this in my voice, using [[1-voice/README]]." ``` **Two rules while writing:** - Don't make things up. If a claim has no source in their answers or notes, drop it. - Use their actual words wherever you can — especially for voice. ### When done Check the master map and each folder README exist. Update `progress.md`: `phase_3_synthesis: complete`. Then: > "Your twin is built, at `03. Areas/Self/digital-twin/`. Open the master map to see yourself through my eyes. > > Want to make it sharper? There's one optional step left — feeding me more of your real data. You can do it now or come back to it any time." --- ## PHASE 4 — Feed your brain (optional, recommended, do it any time) Print: > "Phase 4 — optional. Your twin already works. But if you give me your real history — old chats, notes, call recordings, documents — it gets a lot sharper, because it's grounded in how you actually talk and decide, not just today's answers. > > No pressure. Skip it, or do it now, or come back next week and say 'add this to my brain.' You don't need all of these — start with the one or two that hold the most of your thinking. Here's what you can give me, roughly most useful first:" > 1. **Your ChatGPT history** — usually the richest: months of how you actually think. ChatGPT → Settings → Data Controls → Export data. You'll get an email with a link; download and unzip the `.zip`. The file that matters is `conversations.json`. Drop the folder into `01. Inbox/`. > 2. **Your Claude history** — same idea from your Claude account: Settings → Privacy → Export data (wording may differ slightly). Drop the export into `01. Inbox/`. > 3. **WhatsApp chats** — open a chat → Export Chat → Without Media → send it to yourself so the `.txt` lands on your computer → drop it in `01. Inbox/`. Do the chats that matter most: clients, your team, your own notes-to-self chat. > 4. **Apple Notes** — use the free "Apple Notes Exporter" tool to turn your notes into text/markdown in bulk → drop the folder in `01. Inbox/`. > 5. **Fathom call recordings** — open a call → Copy Transcript and Copy Summary → paste into a text file → drop in `01. Inbox/`. > 6. **Notion pages** — in Notion, ... menu → Export → Markdown & CSV → unzip → drop in `01. Inbox/`. > 7. **Business documents** — offers, frameworks, sales pages, SOPs, anything that holds your thinking (PDF, Word, markdown, text) → drop in `01. Inbox/`. > "A few honest things: all of this stays on your computer — you're handing it to your own AI, not posting it anywhere. You don't have to be neat; I do the sorting. Start small — one source is enough to feel the difference, and every source you add makes the twin sharper. There's no rush and no finish line. > > Whenever you've added something, just say 'add this to my brain' and I'll sort it in and refresh your twin. If it's a big export and you're on the $20/month plan, tell me to use Sonnet for it." ### When they say any version of "add this to my brain" (Treat any phrasing as the trigger — "add this to my brain," "add my ChatGPT history," "add these chats," and so on. They all mean: import what's new and refresh the twin.) 1. **Find what's new** in `01. Inbox/` (and `01. Inbox/agent-drops/`). 2. **For a chat export** (a folder with `conversations.json`): pull out the meaningful bits as small notes. The golden rule: only THEIR words are their words. When the AI wrote something, that's the AI's — even if they asked it to "write in my voice." Tag each note: - their words → `#source/user-statement` - something they clearly agreed with → `#source/user-belief` - the AI's words they didn't react to → `#source/model-statement` (kept for context, not used as their belief) Skip filler ("thanks," "ok"). One clear fact, opinion, or preference per note. Save notes to `04. Resources/AtomicNotes/{source}-{TODAY}/` (the folder the Strategy Engine made for this). **If this is a big export**, split the work across several helper sub-agents running at once to keep it fast, and give each the same rules above. (On the $20/month plan, use Sonnet for these.) 3. **For other files** (WhatsApp, Notes, Fathom, Notion, docs): read each, pull the meaningful bits into small notes the same way, and file them where they belong — about them → `03. Areas/Self/`, about a project → that project, about an ongoing area → that area. 4. **Watch for sensitive data.** If a file has passwords, full card numbers, ID numbers, or private medical details, don't copy those into notes — flag the file to them instead. 5. **Refresh the twin.** Re-run Phase 3's writing step, now with the richer notes folded in. Tell them what changed. 6. Update `progress.md`: `phase_4_feed_data: complete (last run {TODAY})`. --- ## Done Print: > "Your Self Engine is in. > > - Your twin lives at `03. Areas/Self/digital-twin/` — voice, values, how you think, your goals, story, the people who shaped you, and what you're wrestling with. > - I read it whenever I help you, so what I write fits you. > {if they fed data:} - Grounded in your real history, not just today's answers. > > Try me: 'Write a short post in my voice about {something you care about}, using my twin.' See if it sounds like you. > > Refresh this every 6-12 months — just run this file again and say 'continue.' Your AI workflows (the Systems Engine) come in a later session." Update `progress.md`: `status: complete`. --- ## How resuming works (for you, Claude) If they quit and come back later in a new session and say "continue": 1. The root `CLAUDE.md` points you to `02. Projects/Digital-Twin/progress.md`. 2. Read it, find the first step marked `pending` or `in-progress`, and carry on. 3. Tell them where you're picking up. They never re-paste anything. --- ## Safeguards - Don't go past Phase 0 if the Second Brain isn't there. - Save after every single question, and after each batch of imported notes. Never advance on half-saved state. - All progress lives on disk (in `02. Projects/Digital-Twin/` and `03. Areas/Self/`), not in the chat — so it survives closing the app. - Re-running is safe: it finds existing progress and offers to resume or refresh rather than wiping anything. - Their words only count as their beliefs. The AI's words never do — even "write in my voice" output. - Never invent answers or quotes. If there's no source, leave it out. - If you're unsure what they mean, ask a short clarifying question rather than guessing. - Phase 4 is optional. If they skip it, the twin still works. --- That's the Self Engine. When they're ready, run Phase 0 and wait for "go."