what is codex-dream-skin-workflow-engine fr?

Codex Dream Skin Workflow Engine is a local desktop theming tool for the macOS version of OpenAI's Codex app. It changes the visual appearance of the official Codex desktop application by injecting CSS styles and small DOM markers into the app's interface through Chrome's DevTools Protocol, connecting only to 127.0.0.1 on the local machine. It does not modify the official application files, its code signature, or any user login data. Rather than being a single fixed skin, the project is a full workflow: visual ideas are turned into a structured theme specification, image assets are processed and compressed, the theme is applied to specific named modules of the interface, the result is checked against the app's native layout so nothing important is covered or broken, and the whole theme can be fully reverted at any time. Users do not need a ready made prompt to start, they can begin from a rough idea and refine it interactively with Codex itself. The project targets specific named surfaces in the app, such as the sidebar, the message composer, chat bubbles, and side panels, and treats them differently from independent decorative elements like character graphics, which must stay clear of anything a user needs to read or click. A small set of interactive elements, like a click triggered animation, only load their assets when actually activated and clean up afterward, which keeps the tool from running background processes all the time. Technically, the project is built from a Node.js and Bash based command line installer, a JSON theme specification file, a JSON module policy file controlling which visual pieces load and when, a plain CSS stylesheet for the visual layer, and a JavaScript file that injects and manages the DOM markers safely and repeatedly. Image assets are shipped as PNG, WebP, SVG, sprite sheets, and GIF fallbacks. The project explicitly avoids reading or changing the app's authentication file, modifying its packaged application bundle, or altering API key or model provider settings, and it only ever connects to the local machine. It was built for the OpenAI Build Week 2026 competition in the Developer Tools track.