W.
Interactive Guide

CLAUDE.md for
Product Managers

A short, interactive course on how CLAUDE.md files work — and how to use the memory hierarchy to make Claude show up already knowing your product.

Chapter 01

What is CLAUDE.md?

Every Claude Code session starts from zero. Unless you tell it not to.

CLAUDE.md files are persistent instructions Claude reads at the start of every session. You write them once. They load automatically. Claude shows up already knowing your product, your terminology, and your constraints.

Think of it less like a settings file, and more like briefing a new team member before their first day. You wrote it once, but they carry it into every meeting.
New session
Draft a prioritization framework for our onboarding improvements.
I’d be happy to help! Could you tell me a bit more about your product and your users? What industry are you in, and what does your onboarding flow currently look like?
We use RICE scoring, B2B SaaS, enterprise buyers, our users are called “workspace admins”…
Got it! So for B2B SaaS with enterprise buyers, I’d recommend starting with…

Every session. Every time. Repeating yourself.

Chapter 02

The Memory Hierarchy

Five files. Each one loads at a different scope.

Claude loads CLAUDE.md files in a specific order — from broadest to most specific. Each level you unlock below stacks on top of the ones before it.

Set by your IT or DevOps team. These are company-wide policies — security rules, compliance constraints, approved tooling. You can't override these, and they're always present if your org has configured them.

What lives here
  • Security rules and data handling policies
  • Compliance constraints (SOC2, HIPAA, etc.)
  • Approved tools and services
  • Anything that applies to every employee
# Managed policy — deployed by IT · read-only

## Data handling
- Never log PII or customer data to console
- Do not commit API keys — use environment variables
- Customer data must stay in approved cloud regions

## Compliance (SOC 2)
- Audit logging must not be disabled
- Security review required for new third-party integrations
- Incident response: page #security-oncall in Slack
Most PMs won't interact with this directly. If your org uses Claude Code, ask IT.
Quick check

Where does it go?

Three scenarios. Pick the right file level.

Question 1 of 3

You use RICE scoring on every product you've ever worked on and want Claude to know this across all projects — even ones you haven't started yet. Where does this instruction go?

Question 2 of 3

Your team has a strict rule: always say "workspace," never "project." This applies to everyone on the team, needs to be version-controlled, and should carry over when a new PM joins. Where does this go?

Question 3 of 3

You're currently debugging a specific user interview pipeline and want Claude to remember the staging URL and your current hypothesis — just for you, just for this sprint. Where does this go?

Course map

That’s the whole course.

Jump back to any chapter, or pick the one that’s most useful for what you’re setting up right now.

Chapter 01
What is CLAUDE.md?
The mental model — briefing a team member before day one.
Jump to chapter →
Chapter 02
The Memory Hierarchy
Five levels. What each scope means. What goes where.
Jump to chapter →
Chapter 03
Organizing Your Workspace
File structure for a PM team, conflicts, .claudeignore.
Jump to chapter →
Chapter 04
Connecting Live Tools
MCP servers, design system tokens, engineering collab.
Jump to chapter →
Chapter 05
Keeping It Fresh
/loop, scheduled tasks, hooks, the cheat sheet.
Jump to chapter →
← Back to resources