Your voice, on file
Stop retyping who you are
Where: Claude Code · Folder: one folder you keep —
my-contexton your DesktopStart a new Claude Code session
before you begin.
Exercise 1.2 ended on a promise: a markdown file is the cheapest, most powerful unit of work you have. This exercise cashes it in.
So far every exercise threw its folder away. This one doesn't. You're going to make one folder you keep, put a file in it that describes you, and wire it in so every future sessionSessionA single ongoing conversation with Claude Code. Every message, every file Claude has read, every tool result lives inside one session — stored locally on your machine. reads it — without you dragging anything anywhere.
Fifteen minutes. Three prompts.
Step 1 — Create a new folder you won't throw away
In Finder (Mac) or File Explorer (Windows), go to your Desktop and make a folder called my-context. Then point Claude Code at it.
This is the one exception to "fresh exercise → fresh folder." Everything you build here follows you around for the rest of the course.
Step 2 — Write your person.md
Talk about yourself for two minutes, then let the agent write the file.
Prompt:
I want a person.md — a short file that tells you who I am so future sessions
answer me properly.
Interview me first: ask me 5 short questions, one at a time, about my role,
what I'm actually trying to solve at work, how I like to think, and how I
like to be spoken to. Then write the answers into person.md in this folder.
Keep it under 15 lines. No filler.
Answer the questions like a human, not like someone filling in a form. Short and honest beats long and polished.
Want a head start? The person.md page shows a worked example — John Doe, five lines. Steal the shape.
Step 3 — Wire it into your global CLAUDE.md
Right now person.md is a loose file. One prompt turns it into something Claude reads everywhere.
Prompt:
Add an import for this person.md to my global CLAUDE.md at ~/.claude/CLAUDE.md
(create the file if it doesn't exist). Use the absolute path to this folder.
Then add two conditions underneath it for when to read it:
- when you are teaching or explaining something to me, pitch it at my background
- when you write anything I will share outside my team, match my tone
Show me the file when you're done.
Step 4 — Prove it
Start a new Claude Code session in a completely different folder — any empty one.
Prompt:
Who am I, and how should you talk to me?
It knows. You dragged nothing in.
Stuck?Debrief — what should have happened
- A folder on your Desktop called
my-contextwithperson.mdinside it. - A
~/.claude/CLAUDE.mdthat imports it with an@line and two plain-English conditions. - A brand-new session, in a folder that has never seen your name, answering as if it had.
You've now used contextContextEverything Claude can see right now — the conversation so far, the files it has read, the tool results, the system instructions. A big text buffer the model reads end-to-end before every reply. in two different shapes. In 1.2 you pasted the spec into the prompt. Here you filed it once and stopped thinking about it. Same effect, one-thousandth of the effort — that's the entire trick, and every later exercise is a variation on it.
The rule worth stealing: anything you've explained twice belongs in a file. The third time is free.
Stuck?FAQ for this exercise
Q: What is ~/.claude/CLAUDE.md and why is it different from a normal CLAUDE.mdCLAUDE.mdA plain markdown file that teaches Claude how to behave in a project — loaded automatically into context at the start of every session.? A: A CLAUDE.md sitting in a project folder applies to that project. The one in ~/.claude/ is your global one — it applies to every folder you ever open. That's why the file follows you into a fresh folder without dragging anything.
Q: I don't want work context bleeding into every session. A: Then skip the global file and keep person.md loose — drag it in when you want it. You lose the magic trick and keep full control. Both are legitimate; most people start global and never look back.
Q: Can I have more than one? A: Yes. person.md for you, team.md for who you work with, design.md in the next exercise. Import as many as you want; the agent reads them all.
Q: The interview questions felt generic. A: Push back mid-interview — "ask me something harder", "that question is useless, ask me about X instead". The file is only as good as what you told it.
OptionalGo deeper
Make the file argue with you.
Read my person.md and tell me what's missing. What would a colleague who has
worked with me for a year know that this file doesn't say? Ask me about the
three biggest gaps, then rewrite the file.
Voice files rot fast because people write them once, in a hurry, describing who they'd like to be. This prompt is the cheapest maintenance you'll ever do on one.