Transfer Your AI to Claude

I'm moving to Claude as my primary AI. You know me — how I think, how I communicate, what I care about, where I get stuck. Extract that knowledge into three structured outputs Claude can use.

This isn't a summary of our conversations. It's a transfer of *how to work with me*.

Before producing anything, reflect silently:
- What do I consistently care about across topics?
- How do I think? What reasoning style do I need?
- What are my failure modes?
- What communication patterns do I respond to vs. tune out?
- What do I know deeply vs. where are my blind spots?
- What am I building or working toward?
- When have our conversations gone well vs. poorly, and why?

Produce these three outputs. No duplication across them. Every word load-bearing.

FORMATTING: Each output must be inside a code block (triple backticks) so I can copy it with one click. Above each code block, write the exact location where it should be pasted.

---

### OUTPUT 1: STYLE (max 200 words)

**Paste into: Claude → Settings → Profile → Style → Create custom style**

Controls *how Claude writes* — voice, rhythm, formatting, register. Pure presentation layer.

Cover: sentence rhythm and density, formatting preferences, register (match mine), what to avoid in tone, how I like ideas presented.

No reasoning instructions, no facts about me, no behavioral protocols.

---

### OUTPUT 2: PERSONAL PREFERENCES (max 100 words)

**Paste into: Claude → Settings → Profile → Preferences**

Always active, no loading step. The failsafe — if everything else fails, these still shape Claude.

3-5 short imperatives that bias Claude's *thinking* (not writing — that's Style, not facts — that's the next output). Each counters a specific default AI tendency that doesn't serve me.

---

### OUTPUT 3: MEMORY + SKILL (no length limit)

**Paste into: a Claude conversation — Claude will process it automatically**

This output is a single message the user will paste into Claude. Claude will read it and automatically save the memory entries and create the skill file. Structure it exactly like this:

Start with the line:
"I just migrated from ChatGPT. Below is my transfer data. Please: (1) save each memory entry using your memory tools, (2) use your skill-creator skill to create the skill file from the spec below. Do both now."

Then two clearly labeled sections:

**MEMORY ENTRIES**
Max 10 entries, each under 500 characters. Persistent facts: professional context, thinking patterns, failure modes, active projects. Third person ("User works on..."). Each entry standalone. Number them.

**SKILL FILE SPEC**
Deep reasoning protocol. Structure:
1. Reasoning protocol — internal checklist before responding. Never surfaced in output.
2. Behavioral modes — how to respond when I'm presenting ideas, solving, stuck, exploring, or evaluating.
3. Hard stops — behaviors that break the interaction.
4. Background context — intellectual influences, creative patterns, psychology. Applied when relevant, never performed.

Operating manual, not personality profile. Include YAML frontmatter with name: user-context and a description of when to load it.

---

## Constraints:

- No duplication. Style = voice. Preferences = cognitive failsafe. Memory + Skill = facts and reasoning protocol.
- Honest about weaknesses. Operating manual, not portrait.
- Gaps over noise. If you lack signal, say so.
- Every instruction counters a specific default AI behavior. Generic advice is worthless.
- Each output independently useful if it's the only one Claude has.

## IMPORTANT — remind the user:

After producing the three outputs, tell the user: "Once you've pasted all three outputs into Claude, start a new conversation and paste the calibration script to let Claude interview you and refine everything."

I just migrated from ChatGPT and configured your Style, Preferences, Memory, and Skill file based on what it knew about me. Those layers are a starting point — not the full picture.

Run a calibration conversation with me:

1. Start by telling me what you already know — scan my Style, Preferences, Memory, and Skill layers and name 2-3 specific things you see about how I think and work. Be concrete.

2. Interview me with 5-8 targeted questions, one at a time. Probe the gaps — the things an AI learns after hundreds of conversations but that are hard to export:
   - How do I react when I'm wrong?
   - What makes me lose trust in an AI response?
   - What does being stuck actually look like for me?
   - What am I avoiding right now and why?
   - How do I make decisions under uncertainty?
   - What kind of pushback do I respond to vs. resist?

   Adapt based on what you see in my layers. If an answer feels shallow, push deeper. One question at a time.

3. After the interview, propose specific updates:
   - New failure mode → suggest a memory entry
   - Communication mismatch → suggest a style edit
   - Missing heuristic → suggest a skill file addition
   - Harmful preference → suggest removing or rewording

4. Save this for later — after 3-5 real conversations, send me:
   "Now that we've worked together, here's what I've noticed about how you think: [observations]. What am I getting wrong? What am I missing?"

Start now. Tell me what you see in my layers.