You do not need 40 prompts. You need three artifacts that travel with you: a who-I-am file, a what-I'm-building file, and a how-I-write file. Together they are a cheap operating system.
Step 1 — Who you are (one page)
Write this once. Update it when the job changes, not when you feel uninspired.
# WHO.md Role: What I own this quarter: What I refuse to work on: People I support / people who support me: Tools I already pay for: Decisions I make vs. decisions I only advise on:
The last line matters. If Claude thinks you can fire a vendor you cannot fire, it will give you theater. Be explicit about authority.
Step 2 — What you're building (living, short)
A weekly file, not a manifesto. I rewrite mine every Monday in ten minutes.
# NOW.md This week's one outcome: The three tasks that actually move it: What's blocked, and by whom: What "done" looks like on Friday:
Drop both files into a Claude Project. At the start of a chat: "Read WHO.md and NOW.md. Then help me with X. Do not recap them back to me."
Step 3 — How you write (samples, not adjectives)
Adjectives fail. "Punchy, warm, executive" still produces LinkedIn sludge. Paste three emails you actually sent that you'd send again, and one you regretted. Then:
From these samples, write 12 rules for my voice. Ban phrases I overuse. Give me a 40-word style card I can paste into custom instructions.
That style card is the difference between "this is useful" and "I would send this." Keep the samples in the Project. When a draft feels off, don't rewrite from scratch — point at a sample: "Match the tone of email 2, not email 1."
Why this beats prompt packs
Prompt packs expire the moment your job changes. These three files don't. They also work in ChatGPT, Gemini, and Claude Code. The model is interchangeable. The context is not.
If you only do one thing after this page: create WHO.md today and put it in a Project. That single file will outperform a month of collecting screenshots.