Taste is a pipeline for generating reusable SKILL.md files from a curated set of reference images. It targets users, including AI agents, who need to translate visual design principles into concrete, actionable instructions for AI-driven tasks. The primary benefit is the automated creation of detailed, style-specific AI skills from visual examples, enhancing consistency and capability.

How It Works

The pipeline follows a five-step process: curating reference images, indexing them while removing duplicates, analyzing each image independently using configured vision models (e.g., OpenAI GPT-5.5, Anthropic Claude Sonnet 4.6) to extract visual properties like layout, color, and hierarchy. These analyses are then synthesized and chunked into rule sets. Finally, a skill writer translates these rules into specific, objective instructions, avoiding vague aesthetic labels and producing a SKILL.md artifact. This approach aims to provide a structured method for AI to learn and replicate visual styles.

Quick Start & Requirements

To use the local pipeline, first install dependencies with npm install. Then, copy .env.example to .env.local and configure your API keys (e.g., OPENAI_API_KEY, ANTHROPIC_API_KEY, or gateway keys like OPENROUTER_API_KEY). Place your reference images (JPG, PNG, WebP) in the reference-images/ directory and run the pipeline using npm run taste. The generated skill will be saved to .taste/runs/<run-id>/SKILL.md. The system requires Node.js and API access to specified AI providers.

Highlighted Details

• Automated Skill Generation: Transforms visual references into concrete, machine-readable SKILL.md files with specific instructions (e.g., "Use neutral sans-serif typography").
• Local-First Pipeline: The scripts/ directory provides a local runner that avoids external dependencies like Postgres, Vercel Blob, or hosted authentication, facilitating private use.
• Agent Comparison: Demonstrated effectiveness in a frontend design comparison against base models and other AI skills, producing standalone HTML files based on prompts.

Maintenance & Community

No specific details regarding maintainers, community channels (like Discord/Slack), or project roadmap were found in the provided README excerpt.

Licensing & Compatibility

The provided README excerpt does not specify a software license. Users should verify licensing terms before adoption, especially for commercial use or integration into closed-source projects.

Limitations & Caveats

The pipeline has a current production limit of 20 reference images per run. It relies on external API keys for vision model access, incurring potential costs and dependencies. The absence of a stated license requires careful consideration for usage.