Your CLAUDE.md brain file

Every Cowork power user interviewed for this course has some version of the same file. JJ Englert calls it a "brain file." The ccforpms.com guide calls it a "project brief." Karo Zieminski uses CLAUDE.md directly. The name doesn't matter. The concept is identical: a markdown file Cowork reads at the start of every session that tells it who you are, how you work, and what you expect.

This is the single most important file you'll create in this course. A good CLAUDE.md turns a generic AI response into something that sounds like it was written in your office by someone who knows your context.

What goes in it

1. Who you are — your name, role, company, team, and current priorities. Not as a formality — as operational context. Claude needs to know whether you're a solo consultant billing by the hour or a VP inside a 500-person company to calibrate its outputs appropriately.

2. How you work — preferred tone, output formats, document conventions. "I want everything in docx" or "bullet points for internal, prose for external" or "never use exclamation points." Concrete preferences, not vague adjectives.

3. Who you work with — key people and their roles. Your manager's name, your client's communication style, your design team's format preferences. When you ask Claude to "draft an update for Sarah," it should know who Sarah is and what she cares about.

4. What good looks like — link to examples of past output you were happy with. A report that hit the mark. A slide deck that nailed the tone. A summary that you forwarded without editing. These examples teach Claude your actual quality bar better than any description.

5. What bad looks like — link to things you hate. AI-tells. Off-brand phrasing. Format choices you find unprofessional. Being explicit about failure modes is just as important as describing success.

6. Standing constraints — rules that apply to every task. "Never delete original files." "Always write outputs to out/." "If unsure about something, create an open-questions.md file instead of guessing." "Ask before publishing anything externally." These constraints don't need to be in every prompt if they're in CLAUDE.md.

Global instructions vs folder CLAUDE.md

Cowork gives you two places to put context, and understanding the difference matters.

Global instructions (Settings → Cowork → Global Instructions) apply to every session, regardless of which folder is mounted. Put high-level preferences here: your role, your general communication style, core safety constraints. These are the things that are true about you in every context.

Folder CLAUDE.md applies only when Claude is working in that folder. This is where project-specific context lives — the stakeholders for this project, the output format for this workflow, the constraints specific to this client. Keep it separate from global instructions so your weekly-report context doesn't bleed into your file-organization tasks.

The test for a good CLAUDE.md

Ask Cowork to read your CLAUDE.md and give you a one-paragraph summary of who you are and what you're working on. Read that summary. Does it sound like how you'd describe yourself to a new colleague? Does it capture your priorities, your constraints, your quality bar?

If the summary sounds generic, your CLAUDE.md is too vague. Add specifics. If the summary sounds accurate, you've built something useful.

The best practice is to tell Claude "read CLAUDE.md before starting" at the top of every task. Or bake it into your global instructions so it happens automatically — "always read CLAUDE.md at the start of every session in any mounted folder."

You'll update this file regularly. That's correct. Every time Claude produces output that gets something wrong — uses the wrong tone, misidentifies a stakeholder, writes in a format you didn't want — that's a signal to add something to CLAUDE.md. The file compounds with use.