Huy's Wiki

Claude Folder — The File That Controls What Claude Does Before You Type

Properties1
tagsclaude, claude-code, project-setup, agents-md, workflow

2 min read

Claude Folder: You Won’t Use Claude The Same Way After Understanding This Folder

Author: Ilia Karelin & Neo Kim — The System Design Newsletter
Source: Substack
Date: June 24, 2026

A map of every file inside, and why it controls what Claude does before you type a word.

Note: This article is paywalled. The notes below are compiled from the public preview and general knowledge about Claude Code project configuration.

What’s the “Claude Folder”?

When using Claude Code (and similar AI coding agents), a special folder of config files in your project root determines how the agent behaves, what context it has, and what rules it follows — before you type a single prompt.

These files include:

CLAUDE.md / AGENTS.md

  • Project-level instructions for the AI agent
  • Architecture decisions, conventions, coding standards
  • What NOT to do (anti-patterns, forbidden approaches)
  • Test setup, build commands, deployment info
  • Agent framework-specific: place it in the project root, Claude Code auto-loads it

.claude/ folder

  • Project-specific instructions and memories
  • Can contain custom rules per project
  • Similar to .cursor/rules/ in Cursor

Benefits

  • Consistent output across sessions
  • No need to re-explain project context every time
  • Team can standardize how AI behaves in their codebase
  • New devs / agents get up to speed instantly

The Big Picture

The “Claude Folder” concept is about shifting from vibe coding to structured AI engineering — setting up guardrails and context files so the AI doesn’t guess. Good prompts + good project context = reliable output.

For Claude Code specifically:

  • CLAUDE.md at project root is the canonical file
  • Multi-file setups via .claude/ directory for complex projects
  • Different agents (Claude Code, Cursor, Copilot) each have their own format

Graph View