Analysis updated 2026-05-18
Break a design screenshot down into reusable techniques instead of copying it wholesale.
Build a personal library of named design moves you can reference across projects.
Hand a coding agent precise, named techniques instead of a vague style request.
Translate a saved technique into a concrete change for whatever product you are currently building.
| pxx-design/cabinet | datcc8a-max/nhom_4_quan_an_fusion | gavrielp1/salary-2045 | |
|---|---|---|---|
| Stars | 20 | 20 | 20 |
| Language | HTML | HTML | HTML |
| Setup difficulty | moderate | easy | easy |
| Complexity | 2/5 | 2/5 | 2/5 |
| Audience | vibe coder | general | pm founder |
Figures from each repo's GitHub metadata at analysis time.
Requires Claude Code (or a compatible agent), a Chromium based browser, and Python 3 with Pillow installed.
Cabinet is a browser based inspiration library aimed at designers and vibe coders who look at a design and feel something they cannot quite name, then struggle to get an AI coding tool to recreate that feeling. Instead of just saving pretty screenshots to a mood board, Cabinet tries to break a reference image down into named, reusable techniques it calls facets, things like a soft gradient, a photo backed interface, or a specific prop style, so those techniques can be handed to a coding agent as precise instructions rather than asking it to copy the whole image. The tool runs entirely in the browser with no backend server. It works alongside Claude Code or another coding agent that can read its ingest rules file. You paste or drop a reference image into the page, click a button to add it to your library, and the agent reads the image and writes out the techniques it finds, including what the technique is called, where it appears in the image, why it works, and how to describe it to an AI. Later, in any project, you select which techniques you want and the agent reads them alongside the context of what you are actually building, drafts a translation of how to apply them, and waits for your approval before touching any code. The project deliberately avoids full image prompts or pasting exact colors, since the goal is transferring a technique rather than copying a picture outright. Setup involves cloning the repository and running an install script, which sets up the ingest and translate commands for Claude Code, and on macOS also installs a watcher that processes new images automatically. On other systems you trigger ingestion manually. The interface itself needs a Chromium based browser like Chrome or Edge, since it relies on the File System Access API to read and write files on disk, plus Python 3 with the Pillow library for measuring images during ingestion. The project ships with six example reference images so the library is not empty on first use, and it is licensed under MIT for the code and technique text, with a separate notice covering the bundled sample images.
A browser based inspiration library that breaks design references into named, reusable techniques a coding AI agent can apply to your own project.
Mainly HTML. The stack also includes HTML, JavaScript, Python.
The code and technique text are MIT licensed, free to use for any purpose. The bundled sample images have a separate notice covering their use.
Setup difficulty is rated moderate, with roughly 30min to a first successful run.
Mainly vibe coder.
This repo across BitVibe Labs
Verify against the repo before relying on details.