Taste Skill
Anti-slop SKILL.md for the front-end. Instead of letting the agent ship yet another generic landing page, Taste Skill reads the brief, infers the design language and tunes three dials (variance, motion, density), with a hard em-dash ban and canonical GSAP skeletons. **It is not a file, it is a family.** The repository ships thirteen separately installable skills, and the name `taste-skill` only covers the first one: a hardened Codex variant, a redesign pass for existing projects, three committed styles (minimalist, brutalist, high-end visual), an image-to-code pipeline, and **brandkit**, which builds a full identity board (logo directions, palette, type pairing, applications) out of an idea that is still vague. **The image branch is worth pulling apart.** `image-to-code` follows a fixed order: it generates site references, analyses them, then writes the front-end to match. `imagegen-frontend-web` and `imagegen-frontend-mobile` stop before code: they only produce design images, to run through a generator such as ChatGPT Images, which helps when you want to settle a direction before paying for a single line of code.
Strengths
- Infers the design language from the brief instead of applying a default theme
- Three explicit dials (variance, motion, density) to frame the output
- Canonical GSAP skeletons provided, motion does not scatter in every direction
Limitations
- Overlaps Impeccable and SkillUI, you pick one taste reference rather than stacking them
- Pairing with GPT Image 2 adds a dependency and a cost for visual references
Best for
- Agents that ship generic front-ends despite a precise brief
- Landing pages and showcase sites where visual taste makes the difference
- Workflows that start from generated visual references before coding the front-end
- A product with no settled visual identity, where brandkit produces the first board before anything gets coded