what is claude-code-for-non-coders fr?

Claude Code for Non-Coders is a governance system, not a code library. It is written for someone who directs an AI assistant like Claude Code to build real software without being able to read the code it produces. The author built the method while creating a real family mobile app using React Native and Firebase, with around thirty server functions and security rules tested in continuous integration, despite not being a professional developer. The project explains its own history in three stages. At first, fear of not being able to check the AI's output led to heavy safeguards: six automatic blocking hooks, mandatory planning for tiny edits, isolated environments and numbered rituals. This created friction and false alarms while the real danger, mistakes in judgment, still slipped through. After an outside audit and a few incidents, the method shifted to lighten don't harden, removing four of the six hooks and making the heavier process optional, moving trust toward the barrier that actually deserves it: automated tests run in continuous integration. The result is fewer, better placed checks: always loaded discipline rules, method skills you can call on demand, a written policy for the most sensitive parts of a project, deterministic CI tests, and a session ritual that lets a memoryless AI resume work cleanly. The repository installs by pasting a single instruction into Claude Code, which reads an install guide, asks a few questions about your project, and sets itself up, or by hand, copying a global rules file and three skill folders into your Claude configuration. Included are templates for a session handoff journal, a minimal project rules file, and a ready to paste session opener, plus guardrail patterns, two example hooks, and field notes drawn from real incidents. The method rests on five principles: putting effort into a precise brief before work starts, requiring an actual question before any AI changes scope, defining a written zone where mistakes are costly, trusting deterministic tests above cosmetic checks, and treating verification as proof rather than narrative. It is aimed at non-technical founders, designers, product managers and side project builders acting as a technical director over an AI team, though developers are told they will find useful engineering discipline here too. The project is released under the MIT license.