The "leonxlnx/taste-skill" repository provides a collection of portable "Agent Skills" designed to enhance the visual quality and originality of AI-generated user interfaces. Its primary goal is to prevent AI agents from producing bland, generic, or repetitive UI designs—often referred to as "slop"—by equipping them with advanced design sensibilities. Taste Skill achieves this by offering a suite of modular skills that can be easily integrated into AI-driven frontend development workflows, regardless of the underlying framework or coding agent.
The repository centers around the "taste-skill" (install name: design-taste-frontend), which is currently in its experimental v2 iteration. This core skill reads design briefs, infers the appropriate design language, and dynamically adjusts key parameters such as layout variance, motion intensity, and visual density. These parameters are controlled by three dials (DESIGN_VARIANCE, MOTION_INTENSITY, VISUAL_DENSITY), each ranging from 1 to 10, allowing users to fine-tune the resulting UI's style and complexity. The skill also incorporates mechanisms like a hard ban on em-dashes, canonical GSAP animation code skeletons, a redesign-audit protocol, and strict pre-flight checks to ensure high-quality, distinctive outputs.
Beyond the main skill, the repository includes several specialized variants tailored for different use cases. For example, "gpt-taste" enforces stricter layout and motion rules for use with GPT/Codex agents, while "image-to-code-skill" supports workflows where reference images are generated and then translated into code. Other skills focus on redesigning existing projects, enforcing complete output (preventing the AI from leaving work unfinished), or targeting specific visual styles such as minimalist, brutalist, or high-end premium UIs. Each skill is self-contained and can be installed individually, allowing users to select only those relevant to their needs.
A notable feature of Taste Skill is its support for image-generation workflows. Skills like "imagegen-frontend-web," "imagegen-frontend-mobile," and "brandkit" produce reference images for websites, mobile apps, or brand identity boards. These images can be used with ChatGPT Images, Codex image mode, or similar agents to generate visual comps, which can then be handed off to coding agents for implementation. This image-first approach streamlines the process of moving from design ideation to code, ensuring that the final product maintains a high standard of visual quality.
Installation is straightforward via the "npx skills add" CLI, which scans the repository's skills folder and allows users to add either the entire suite or individual skills by name. The repository is framework-agnostic, supporting React, Vue, Svelte, and other major frontend technologies. It also provides comprehensive documentation, examples, and a changelog to help users stay up-to-date with the latest improvements.
Taste Skill is open source under the MIT License and welcomes feedback, contributions, and sponsorships from the community. The project is informed by ongoing research into design best practices, with background materials available in the repository's research directory. Overall, Taste Skill empowers AI agents to produce more creative, polished, and engaging user interfaces, elevating the standard of AI-assisted frontend development.
