referodesign/refero_skill

Research-first design skill for AI agents. 150K+ real app screens and flows via Refero MCP.

What it solves

Refero addresses the tendency of AI models to rely on generic design defaults by forcing them to perform evidence-based research before implementing UI/UX tasks. It ensures that design decisions—such as typography, color, and user flows—are grounded in real-world product examples rather than hallucinations.

How it works

It functions as a design-focused skill and Model Context Protocol (MCP) connection for AI agents. When an agent is tasked with design work, it uses the Refero skill to query a database of over 150,000 app screens and 6,000 user flows. The agent extracts specific patterns and references from this library, applies bundled design guidance, and validates its output against the selected visual direction before finalizing the work.

Who it’s for

This tool is for developers and designers using AI agents (such as Claude Code, Cursor, or ChatGPT) who want to automate UI/UX implementation while maintaining high design standards and consistency with established industry patterns.

Highlights

  • Integrates with major AI agent platforms including Claude Code, Cursor, Gemini, and ChatGPT.
  • Provides access to a massive, curated library of real-world app screens and user flows.
  • Enforces a research-first workflow to prevent generic AI design outputs.
  • Includes built-in design system guidance for typography, motion, spacing, and copywriting.
  • Uses the Model Context Protocol (MCP) for standardized, secure communication between the agent and the design database.

Related

  • Project
    wellingfeng/UltraGameStudio

    An AI coding agent tailored for game development that integrates code generation with the creation of 3D models, sprites, and audio assets in a single workflow.

  • Project
    vercel/next-evals-oss

    Next.js Evals is an open‑source harness that runs AI coding agents against tiny Next.js projects, checks hidden assertions, and publishes pass@4 scores to the Next.js eval leaderboard.

  • Project
    Jianing1117/jianing-skills

    A collection of modular, installable skills for AI agents that automate tasks like content creation, visual design, and personal data reflection.

  • Project
    LukeRenton/explore-claude-code

    A clickable, simulated Claude Code project that teaches every real config file and built-in feature through annotated examples. Instead of reading documentation, you explore a replica repo where each file reveals what it does and how to set it up, so you can copy the scaffolding into your own projects.