Type them into the Code tab in Claude Desktop, in this order, when the room does. Or tap copy.
A chatbot answers your question. An agent knows your situation.
The prompts
P1 · ACT I · WHO AM I?
Read my-desk.html in this folder first. Then interview me, one question at a time, asking only what is still missing, until you know: who I am, my role, what I produce every week, what I read every week to produce it, and how I like things done. Then write CLAUDE.md in this folder, in my own words, with those as its sections. Then make three sub-folders if they are not there yet: inbox, outbox and notes.
Why: this builds your handbook - the desk's rulebook for who you are.
P2 · ACT II · WHAT WAS I DOING?
What were we working on just now, and what's left to do?
Why: asked in a brand-new session, this shows you goldfish mode.
Typed twice: once in Act II in a new session, once again after the break.
ADDING SKILL-CREATOR · ACT II · BEFORE P3
1. Open Claude Desktop, go to the Claude Code tab. 2. Open Customize. 3. Open Skills. 4. Search "skill-creator" and add it.
P3 · ACT II · WHAT WAS I DOING?
Build me a skill called handoff. Use the skill-creator skill if you have it; if not, create the skill yourself as .claude/skills/handoff/SKILL.md. Its description must say: use this whenever I say "wrap up". The skill: update HANDOFF.md with what we did, any decisions we made, and what's next, then confirm in one line.
Why: Builds your first skill. If skill-creator is not there, the agent writes it itself.
P4 · ACT II · WHAT WAS I DOING?
wrap up
Why: Two words. Your handover writes itself. Take the break.
P5 · ACT III · HOW DO WE DO THINGS HERE?
What do you know about how I like things done? Tell me in three lines. I will correct one thing; when I do, add the corrected rule to CLAUDE.md.
Why: You correct one thing. The rule lands in your handbook. That is how it grows.
BONUS · ACT III · IF YOU FINISH EARLY
Read CLAUDE.md and HANDOFF.md. Tell me one thing you would add to each so you work better for me next week. Then add them.
Why: this is how your agent grows further. One more pass, in its own words.
P6 · ACT IV · WHAT'S MY ROUTINE?
Here is one thing I do every week: I turn the file in my inbox into a one-page brief for my manager, with three key points and one recommended action. Turn this into a skill called weekly-brief that reads the file in my inbox folder, writes the brief into my outbox folder, and updates HANDOFF.md when it's done. Then run it now on the file in my inbox.
If your real weekly task is different, replace the first sentence with yours. Keep the rest.
Why: your first real output lands in your outbox.
P7 · ACT IV · WHAT'S MY ROUTINE?
Schedule the weekly-brief skill to run every Monday at 9am. If you can't schedule it here, tell me, and write a one-line reminder into HANDOFF.md to run it by hand every Monday.
Why: your first routine. It runs while you do something else.
The data rule
Own notes and non-client material only. Nothing classified, nothing with customer data.
Bring one real input file you use every week into C:\AI-Harness\inbox before the day. Non-client material only. Make it a Word, PDF or notes file. If it is a spreadsheet, save a PDF copy next to it.
Stuck?
Raise your hand. Help comes to you. Nobody goes to the stage.
What's the last thing you did, and what do you need from me to continue?
Your harness, four parts
Reins - instructions. Your handbook, CLAUDE.md.
Saddlebag - memory. Your handover, HANDOFF.md.
Training - skills. What you built in Act II.
Route - routine. What you scheduled in Act IV.
You're the rider. You hold the reins.
This week
Run "wrap up" at the end of each day.
Put one real weekly task through your routine.
Every morning, ask your agent "what were we doing?"
When it gets something wrong, tell it the rule and say: add that to CLAUDE.md.