codeswithroh/tastemaker
A Claude Code skill that grounds AI-generated UI in real reference images and a persistent per-developer taste profile, instead of generic AI-slop defaults.
What it solves
AI-generated user interfaces often suffer from a "monoculture" of generic designs—typically characterized by indigo-to-purple gradients and soft-shadow cards—because models lack grounded design taste and memory of user preferences. Tastemaker provides coding agents with a structured design system and a set of computational constraints to ensure UIs look professional, unique, and accessible.
How it works
Tastemaker acts as a skill for coding agents (such as Claude Code, Windsurf, and Gemini CLI) that replaces generic defaults with a rigorous design workflow:
- Computational Palette Generation: Instead of picking from a list, it uses a Python script to generate a fresh, contrast-verified palette based on a project's "mood" (e.g., premium, technical, playful) using OKLCH color space.
- Pixel-Based Extraction: It can read actual pixel values from reference images via scripts rather than relying on the model's text descriptions of a "vibe."
- Enforced Constraints: A contrast matrix script (
check_contrast.py) mathematically verifies that color pairings meet WCAG accessibility standards, treating readability as a hard constraint rather than a suggestion. - Anti-Slop Scanning: A dedicated script (
anti_slop_scan.py) flags common AI design tropes like generic gradients,h-screenusage, and placeholder copy. - Persistent Memory: Design decisions are locked in local files (
.tastemaker/style-lock.md) and personal preferences are stored in a global profile (~/.tastemaker/profile.md) to maintain consistency across sessions and projects.
Who it’s for
Developers using AI coding agents who want to build high-quality, visually distinct landing pages and app screens without manually managing a design system or fighting the model's default aesthetic tendencies.
Highlights
- Dynamic Palette Generator: Creates unique, legible color schemes per project based on mood and harmony rules.
- Real Asset Sourcing: Integrates with Openverse, Iconify, and unDraw to pull in attribution-free photos, icons, and illustrations.
- ** Firestore-free Local Execution:** Runs entirely on the user's machine with no hosted backend or API keys required.
- Motion Integration: Implements GSAP and ScrollTrigger for professional motion by default, with an audit script to prevent common animation errors.
- Design Lock: Prevents "style drift" by recording decisions in a project-specific log and a durable user profile.
Related
- Project
- Project
- Project
- Project