what is winlean-coding-skill fr?

WinLean Coding Skill is a set of instructions you add to an AI coding assistant so it behaves more carefully inside Windows based projects. Instead of repeating the same warnings in every prompt, you install it once and the assistant follows the same discipline every time it touches code. The core idea is to make the assistant slow down before editing. First it checks the operating system, shell, file paths, encoding, and line ending style, since commands and text edits that work fine on Mac or Linux often silently fail or corrupt files on Windows. Next it limits how much of the codebase it reads at once, avoiding generated folders like dependency caches and build output and focusing only on files that matter. Before writing new code, it checks whether the problem is already solved by an existing framework feature, a public library, or something already used in the project, rather than writing a fresh implementation. When it does change something, it aims for the smallest possible edit, careful with encoding and line endings, and reviews the actual diff before calling the task done. The project ships two pieces. One is the skill definition itself, loaded by a Codex style coding agent to change its behavior for a single task. The other is a plain markdown file meant to be copied into a project's own root folder so the same rules keep applying automatically in future sessions. Three modes are available: a light mode for small low risk edits, a full mode running the entire checklist for normal coding work, and an audit mode that only reports on Windows risk and wasted effort without making edits. Installing it means cloning the repository into your Codex skills folder, with no separate package installation required since everything runs on plain Node.js. The project is released under the MIT license and includes a small informal comparison against a similar project, showing lower token usage on one sample task, though the author calls that a single data point rather than a formal benchmark.