← Back to library

Taste Skill

Turns a design brief into layout, type, motion, and density rules; the default v2 remains experimental.

What problem does it solve

Taste Skill is a set of frontend design rules for coding agents to read. Once you give it the page goal, content structure, and tech stack, the agent first understands the brief, judges which design language fits, then works on layout, type, spacing, motion, and information density, and finally checks whether the result has again turned into a templated interface of centered big headlines, uniform cards, and conservative spacing.

A typical run is to install the skill, trigger it in Codex, Cursor, or Claude Code, and then have the agent generate or adjust frontend code. The difference from simply writing "make it look better" is that it constrains the design and checks the result with explicit rules, instead of relying only on the model picking common, safe styles on the spot.

The default skill offers three knobs — VARIANCE, MOTION, and DENSITY — controlling layout variation, motion intensity, and per-screen information density respectively. The repository also includes specialized skills such as redesign, image-to-code, minimalism, and brutalism, which can be chosen by task without installing all of them.

How to use it

Install the whole repository via the skills CLI, or install only the default skill:

bash
npx skills add https://github.com/Leonxlnx/taste-skill
npx skills add https://github.com/Leonxlnx/taste-skill --skill "design-taste-frontend"

After installation, hand the interface goal, content structure, and tech stack to a coding tool that supports Agent Skills, and trigger the corresponding skill. You can also copy SKILL.md into your project, or paste it directly into a ChatGPT or Codex session.

Input and results

Inputs can be a page brief, an existing frontend project, or reference images. The default skill infers a design language, builds a design system mapping, then implements under the VARIANCE, MOTION, and DENSITY constraints, and checks before delivery.

Implementation skills output frontend code, while image-generation skills output only reference visuals. Whether files are modified, the project is run, or existing styles are overwritten is decided by the agent hosting the skill and the authorization for that run.

Who it's for

It suits people who build landing pages and product interfaces with Codex, Cursor, or Claude Code but often get templated results. When an existing project needs a revamp, you can use the redesign skill in the repository to review the current state first, then adjust hierarchy, spacing, and styles.

Limits of use

The default design-taste-frontend has reached v2, but the repository marks it as experimental, so behavior may still change. Projects that depend on older results can pin design-taste-frontend-v1. The rules are not tied to React, Vue, or Svelte, and final quality still depends on model capability, the brief, and how the host agent executes. The repository uses the MIT License.