“AI Disruption” Publication 10,500 Subscriptions 20% Discount Offer Link.
To be honest, after two years of using Claude Code, I always thought its configuration boiled down to three things: write conventions in CLAUDE.md, set permissions in settings.json, and maybe add an MCP if needed.
Then a couple of days ago I casually opened the .claude folder in the project root — and it turns out it can hold twice as much as I thought. Rules, skills, agents, workflows, agent-memory: all names I had seen in the docs, but I never realized they lived under the same roof.
Worse, I checked against the official directory docs and found I had been putting several configs in the wrong place. For example, I once stuffed a .mcp.json into .claude/, when it actually belongs in the project root.
This post unpacks that “directory map”: what each file does, whether it should be committed or ignored, and who overrides whom. Save it and use it as a checklist next time you set up an environment.
