Tell Claude about your work

What this does. Gives Claude a permanent set of “instruction files” describing who you are, what you work on, and how you write. Every future Claude Code session reads them automatically — so you don’t spend the first ten minutes of every conversation re-explaining yourself.

This is the single highest-leverage thing you’ll do in this course. It’s also the longest single page, because there’s a fifteen-minute interview in the middle. But Claude is doing the writing; you just answer questions.

Why this matters

Out of the box, Claude is generic. It doesn’t know:

• Whether you write in NZ/UK English or US English.
• That you’re a marine biologist (or a photographer, or running a non-profit, or all three).
• Who your team is.
• What projects you’re juggling.
• What “Tofo” is, or who “Nakia” is, or that “Inhambane” isn’t a typo.

You can re-explain that every session, or you can do it once. Once is what this page is about.

How it works (overview)

You paste one prompt into Claude Code. The AI then:

1. Asks you a handful of questions about yourself, your work, and how you like to write.
2. Writes a personalised set of instruction files: a global one at ~/.claude/CLAUDE.md (cross-cutting preferences), a vault-root one (who you are, what you work on), and optionally one per domain folder if your vault has them.
3. Drops a “Getting Started” note in your vault’s inbox so the next session has somewhere to land.

You answer. Claude writes. You read and tweak afterwards if anything’s off.

Steps

1. Paste the bootstrap prompt

Open Claude Desktop → Code tab → vault selected (from the previous page). Paste this:

Walk me through setting up the system at github.com/marinemegafauna/mmf-claude-code on my machine. I'm new to all of this — please hand-hold me through it, asking me one question at a time.

Claude will download the MMF AI Course materials to your computer, install a starter set of shortcuts (“skills”) so commands like /session-start, /document, /transcribe, /update, and others work, and tell you to quit and restart Claude Desktop so it picks them up. Do that.

2. Run /onboard

After the restart, in the Code tab, type:

/onboard

The interview begins. It covers:

• Who you are — your name, what you do, your organisation.
• How you write — spelling conventions (NZ/UK English vs US), tone you prefer, things you care about (“evidence-based”, “no overstatement”, etc.).
• What you work on — the broad areas you split your time across (research, writing, ops, photography, personal — whatever fits).
• Your task manager — Things 3, Todoist, Apple Reminders, or none. Some optional shortcuts use this.
• Optional per-domain context — for each major folder, Claude offers to write a domain-specific instruction file (e.g. for MMF: species names, staff names, project list, funder conventions).

3. Read what got written

When /onboard finishes, you’ll have a small set of new files:

• ~/.claude/CLAUDE.md — global behavioural defaults (loads for every Claude Code session, anywhere).
• <your-vault>/CLAUDE.md — vault-root instruction file: who you are, what you work on, how you write.
• <your-vault>/<DOMAIN>/CLAUDE.md for each domain folder you opted into.
• <your-vault>/MEMORY.md — starts mostly empty; will fill up over time as you correct Claude on things and ask it to remember.
• <your-vault>/00_INBOX/Getting Started.md — your kickoff note.

Open each in Obsidian. Read them. Edit anything that doesn’t sound right. Claude’s draft is a starting point, not a finished document. If the writing voice is off, fix it. If the project list is missing something important, add it. If a domain instruction file got the team names wrong, fix them — small corrections now save you from compounding mistakes later.

4. Switch to Auto mode (so Claude stops asking permission for every edit)

By default Claude Code asks before each file edit and each shell command. After the first few sessions this is just friction. To stop it:

1. In Claude Desktop’s Code tab, look at the bottom-left of the prompt area. There’s a small badge that cycles between three modes: Default, Accept edits, and Auto mode.
2. Click the badge (or press Shift-Tab) until it shows Auto mode.

To make Auto mode the default for every future session, just ask Claude:

Add Auto mode to my settings as the default.

Claude will edit the settings file and tell you what it changed.

Trade-off. Auto mode means Claude can run any shell command without asking. You’re trusting it to act on your machine. If that makes you uneasy, stay on Accept edits instead — it still prompts before shell commands but lets file edits through. Most people end up on Auto eventually.

What to fix right after /onboard

A few common things to check before moving on:

• Spelling drift. If you told Claude “NZ English” but its draft has “organization” anywhere, search-and-replace.
• Stale projects. Anything you mentioned that’s no longer active — remove it now so Claude doesn’t keep referencing it.
• Voice mismatch. If the “how you write” section sounds wrong, rewrite it in your own words. Even a paragraph or two beats a polished draft that doesn’t sound like you.

What’s next

Record & transcribe meetings — your first real workflow end-to-end. Record a meeting, drop it into Claude Code, get back a clean transcript with speaker labels and action items. After that, the daily-rhythm pages cover how to actually use Claude Code day-to-day.