The Anatomy of a Great CLAUDE.md File (With Template)

CLAUDE.md is the operating system for a Claude Code project. When you open a folder containing CLAUDE.md in VS Code, Claude Code reads the file first and uses it as the master rulebook for every action it takes inside that folder.

The file is plain Markdown. No code. No programming. It's closer to a well-organized employee handbook than to a config file.

1. Project Overview — one paragraph explaining what this folder does, for whom, at what cadence.

2. Folder Map — a bullet list of the sub-folders (/context, /agents, /templates, /data) and what each holds.

3. Routing Rules — at least two "if X, then call Y" rules that describe how sub-agents hand off work.

4. Work Rules — constants the AI should always obey: tone requirements, approval checkpoints, brand guardrails.

Below is a minimal but production-grade CLAUDE.md you can copy as a starting point.

# [Project Name] ## Project Overview [One paragraph: what, for whom, cadence] ## Folder Map - /context — brand & business context - /agents — sub-agents (research, content, review) - /templates — standard output formats - /data — CSV inputs ## Routing Rules 1. If user asks for [X], call research_agent first. 2. If research returns < 3 sources, loop before content_agent. ## Work Rules - All outputs must match /context/brand_guidelines.md tone. - Semi-auto mode: confirm before publishing.

If writing CLAUDE.md from scratch feels intimidating, use the Harness Builder. Answer five questions and you get a complete CLAUDE.md plus sub-agent files as a ZIP.