Claude Code Folder Structure: The 11 Line Version.

Ryan “RyMac” McKinney · USMC · 15 August 2026 · 6 minute read

Here is the folder structure I run several businesses on. One file that is always read first, one file that says what we are doing now, and four folders holding the material. The map is 11 lines and it does no work at all, which is the part people get wrong.

The whole thing

~/business/
├── CLAUDE.md          <- the map. 11 lines. always read first
├── Context.md         <- the desk. what we're doing today
└── ops/
    ├── clients/
    ├── offers/
    ├── sops/
    └── archive/

That is not a simplified diagram for the article. That is it.

CLAUDE.md, the map

Its only job is to send the agent somewhere else.

It says: before you touch anything, open Context.md and read it. Then it says what Context.md contains and where the material lives.

That is why it is 11 lines. Start putting instructions and examples in the map and it stops being a map.

It becomes one more thing to read on every single request.

You never edit it. That is the point of it.

Context.md, the desk

Four headings, and they are the four things that change:

What we are working on. One job. The real one, not the tidy one.

Already decided. Closed decisions, with the reason. The reason is what stops the same argument happening twice.

Where the material lives. One path.

Next. The single next action, and who does it.

Keep it short. When a section grows, that work has earned its own folder with its own Context.md, and this file just points there.

The four folders

clients/ One folder per client, each with its own Context.md.

offers/ What you sell, in your own words. One file per offer.

sops/ How repeated work gets done. Written once. This is what turns a 40 minute screen recording into "read this file."

archive/ Finished and dead. Out of sight, still readable.

The mistake that makes it rot

Letting Context.md go stale.

If it says you are working on Thursday's proposal and Thursday was last month, the agent is being handed a map to nowhere. It will confidently work on the wrong thing.

An agent that wanders is not confused. It was given stale instructions.

The habit is: update Context.md before you stop working. That is the entire discipline. Everything else is folders.

Where to put it

Somewhere you already look every day. Desktop, Documents, wherever. Name it after your company if you like. Nothing breaks.

Then open your agent in that folder and type: Read CLAUDE.md, then do what it says.

Why this shape and not another

It is not mine. The method is Interpretable Context Methodology, ICM, and Jake Van Clief and David McDermott wrote the paper: folder structure as agent architecture. Jake teaches it himself over at Clief Notes.

Full breakdown: Interpretable Context Methodology in plain English. The concept underneath it: context engineering.