Ekygai Method — Development skill for Claude Code

The EKYGAI Method

English · **[Français](README.fr.md)**

A way of working with a coding agent: a memory that survives the end of a session, and a protocol that forbids writing code before reading what already exists.

Refined over several years of real development, first on [EKYGAI](https://ekygai.com). This repository publishes the method only — never the product.

---

The problem

A coding agent understands an architecture after two hours of conversation, then the session ends and it all goes. And if nothing is imposed on it, it starts writing as soon as it believes it has understood — which happens very fast. The result compiles, looks right, ignores half the repository's conventions. It survives the sprint. It breaks by the quarter.

Two answers, neither of which holds without the other.

The two parts

Part I — Two Layers of Memory	Automatic capture through hooks, plus an index written by hand. Where facts live, what makes them decay, what it costs — including what it sends off your machine.
Part II — Before the Code, After the Code	The ULTRA PLAN protocol before writing, the GUIDE document after shipping. And the criteria grid everyone must build for themselves.

What's in the repository

en/
├── 01-memory.md           the two memory layers, with diagrams
├── 02-method.md           the protocol and the two documents
├── getting-started.md     the prompt that sets everything up in one conversation
├── skills/
│   ├── ultra-plan/        planning protocol, executable
│   └── delivery-guide/    writing the guide after shipping
└── templates/             memory index, fact sheet, plan, guide

`fr/` mirrors the same content in French.

CONTRIBUTING.md            the criteria grid this repository applies to itself
docs/guides/               including the guide of this repository's own construction
docs/plans/                empty — see that guide, §6

**Thi