jaytel0

jaytel0 / taste

Public

Taste skill creation pipeline

taste.jaytel.com
47
4
85% credibility
Found May 24, 2026 at 75 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
TypeScript
AI Summary

Taste is a web tool that transforms a collection of reference images into a reusable design skill for AI agents. You upload screenshots of designs you admire, and the tool automatically analyzes them to extract visual patterns, design rules, and aesthetic principles. The output is a SKILL.md file that can guide AI coding assistants to produce interfaces matching your preferred style. The hosted version runs entirely in the browser with OpenRouter authentication, while a local command-line version lets you run the pipeline privately with your own AI service keys.

How It Works

1
💡 You discover Taste

You hear about a tool that can teach an AI your design taste by looking at examples you provide.

2
🔑 You connect your AI service

You sign in through OpenRouter to connect your AI account so the tool can analyze your images.

3
📸 You drop in your reference images

You select screenshots of designs you love—dashboards, apps, interfaces—and upload them.

4
✨ The magic happens automatically

The pipeline studies each image, extracts the visual patterns, and distills them into concrete design rules.

5
⏳ You watch the progress

A progress bar shows you each stage: analyzing images, finding patterns, building the rule set.

🎉 You get your design skill

Download your finished SKILL.md file—a reusable set of instructions that makes any AI agent design in your taste.

Sign up to see the full architecture

4 more

Sign Up Free

Star Growth

See how this repo grew from 75 to 47 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is taste?

Taste is a pipeline that analyzes your reference images and generates a reusable SKILL.md file for AI coding agents. Drop in screenshots of designs you like, and it produces concrete visual rules like "use neutral sans-serif typography" or "keep accent color localized" instead of vague directives like "make it premium." The project ships as a Next.js web app with OpenRouter OAuth, plus a standalone local runner that works with just your own API keys and a folder of images.

Why is it gaining traction?

The hook is specificity. Most AI design prompts produce generic output because the model has no visual reference beyond text. Taste solves this by fusing multi-model vision analysis (GPT and Claude) into concrete, actionable rules. The project includes benchmark comparisons showing the generated skill outperforming both base models and Anthropic's own frontend-design skill. The Skill Lab feature lets you run an autoresearch loop where you pick the best AI-generated UI variant and the system iteratively improves the underlying skill.

Who should use this?

Frontend developers who want AI agents to match their design language without constant manual correction. Teams with established design systems who want to encode visual rules for reuse across projects. Developers building tools that generate UI and need consistent aesthetic output. Not suitable for teams without clear visual references or those who prefer ad-hoc prompting.

Verdict

Taste solves a real problem with a thoughtful pipeline, but the 47-star count and 0.85% credibility score signal early-stage software. The documentation is solid and the local runner makes experimentation low-risk. Worth trying if you have reference images gathering dust and want AI output that actually looks like your brand.

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.