Documentation

Getting started.

Install taste-skill, understand how the v2 skill reads a brief, and learn the rules it enforces.

Installation

The default install pulls v2 experimental of the core skill. Works with any agent that supports SKILL.md files.

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

Adds a SKILL.md file to your project. The agent reads it automatically. No extra configuration is required.

Legacy v1

If your workflow depends on v1 behavior, install it explicitly:

$ npx skills add https://github.com/Leonxlnx/taste-skill --skill "design-taste-frontend-v1"

Full bundle

Install every skill in the repo with a single command:

$ npx skills add Leonxlnx/taste-skill

Codex

Install the full skill bundle into the Codex skills directory:

$ npx skills add Leonxlnx/taste-skill -a codex

You can also copy any SKILL.md into your project, or paste it into ChatGPT or Codex conversations.

v2 (experimental)

The default install is now v2 experimental.

v2 is genuinely better than v1 and we recommend it, but it is still iterating. Refinements may land in any pre-release. The install name (design-taste-frontend) stays stable. Section numbering and rule wording may change before v2.0.0 stable cuts.

Compatible Agents

Taste Skill works with every major AI coding agent that supports skill files.

Cursor
Claude Code
Codex
Antigravity
Gemini CLI
AI Studio
v0
Lovable
OpenCode
OpenClaw
Windsurf
Copilot

How v2 thinks

v2 reads the brief first. The agent infers the right layout variance, motion, and density from cues like "minimalist", "editorial", "SaaS", or "portfolio", then picks the design direction.

Brief inference (Section 0)

Reads the page kind, vibe words, audience, references, and brand assets before any code. Declares a one-line design read instead of defaulting to AI-purple gradients and three centered feature cards.

Brief to design system map (Section 2)

When the brief reads as Material, Fluent, Carbon, Polaris, Atlassian, Primer, GOV.UK, USWDS, Bootstrap, Radix, shadcn, or Tailwind, the agent reaches for the official package. Aesthetic-only briefs are labeled honestly as web approximations.

Animation skeletons (Section 5)

Canonical code patterns for sticky-stack, horizontal-pan, and scroll-reveal stagger. Hand-rolled scroll listeners and rAF loops touching React state are out.

Dark mode protocol (Section 8)

Dual-mode by default. WCAG AA contrast, hierarchy parity, brand fidelity in both themes. Off-black and off-white, never pure values.

Big bans (Section 9)

Strict list of the patterns that make generated frontends look the same. The agent enforces these on every output. See the ban list below.

Redesign protocol (Section 11)

Audit-first on existing projects. Greenfield, preserve, or overhaul mode. URL structure, nav labels, and form field names never change silently.

Block library schema (Section 12)

Schema for real, source-backed blocks: hero, feature, social-proof, pricing, CTA, footer, portfolio, transition, navigation. Implementations land iteratively.

Hard pre-flight check (Section 14)

Every checkbox must honestly pass before the agent ships. No box, no ship.

Skill files

The repo bundles multiple SKILL.md files. Enable the ones that match your stack. For release notes, see the changelog.

taste-skillv2 experimental

Substantial 2026 rewrite of the default skill. Reads the brief, infers the right design direction, ships interfaces that do not look templated. Brief inference, design-system map, complete em-dash ban, canonical GSAP code skeletons, redesign-audit protocol, strict pre-flight check. Install name: design-taste-frontend. Actively iterating toward v2.0.0 stable.

taste-skill-v1Legacy

The original v1 of taste-skill, preserved for projects depending on its exact behavior. Install with `--skill "design-taste-frontend-v1"`. Use only if v2 breaks something specific in your workflow.

gpt-tasteskillStable

Stricter taste-skill variant for GPT and Codex: higher layout variance, stronger motion direction, tighter anti-slop checks.

redesign-skillStable

Six-category site audit for upgrading existing projects: finds generic patterns and guides fixes.

soft-skillStable

Calm, polished UI: softer contrast, more whitespace, spring-style motion.

minimalist-skillStable

Editorial product UI inspired by Notion and Linear: restrained palette and tighter hierarchy.

brutalist-skillStable

Swiss typography, raw structure, and sharper contrast.

output-skillStable

Anti-laziness rules for agents: blocks placeholder comments, skipped sections, and half-finished output.

stitch-skillStable

Google Stitch bridge with optional DESIGN.md export.

image-to-code-skillStable

Image-first pipeline: generate reference frames, analyze them, then implement the frontend to match.

imagegen-frontend-webStable

Image-only skill: horizontal website section frames for handoff to Codex, Cursor, or Claude Code.

imagegen-frontend-mobileStable

Image-only skill: mobile screen reference frames for the same handoff workflow.

brandkitStable

Image-only skill: brand identity boards for the same handoff workflow.

The locks (Section 4)

Three rules the agent never relaxes. They keep a generated page visually coherent end-to-end.

Color Consistency Lock

One accent across the whole page. A warm-grey site does not suddenly get a blue CTA in section 7.

Shape Consistency Lock

One corner-radius system per page. All-sharp, all-soft, or all-pill, with a documented exception when mixing.

Page Theme Lock

Light, dark, or auto, picked once at the page level. No mid-page flips into a different theme.

Hero discipline

Hard rules for above-the-fold. The hero must fit the initial viewport.

Headline
Maximum two lines on desktop.
Subtext
Maximum twenty words and four lines.
Primary CTA
Visible without scroll.
Navigation
Single line at desktop, height maximum 80px.

The Anti-Slop Ban System

Section 9 of SKILL.md ships a strict ban list. The agent enforces these on every generation. The full list lives in the skill file.

Em-dashes and en-dashes

Banned everywhere in output copy. Headlines, eyebrows, pills, body, quotes, attribution, captions, button text, alt text. Use a hyphen or restructure the sentence.

Section-numbering eyebrows

00 / INDEX, 001 · Capabilities, 06 · how it works. Eyebrows should name the topic in plain language.

Hero version labels

V0.6, BETA, INVITE-ONLY in the hero is banned unless the brief is explicitly a product launch.

Photo-credit captions as decoration

Field study no. 12 · Ines Caetano under stock images is banned. Real attribution to real photographers only.

Hero decoration text strips

BRAND. MOTION. SPATIAL., TYPE / FORM / MOTION, and other mono-caps strips at the hero bottom are banned.

Pills overlaid on images

No labels or tags floating on top of photos. Add a caption directly below the image if needed.

Version footers on marketing pages

v1.4.2, Build 0048, last sync 4s ago. Devtool fixtures do not belong on landing or portfolio pages.

Locale, city, time, weather strips

Lisbon 14:23 · 18°C and similar atmospheric strips are banned for 99% of briefs.

Scroll cues

Scroll, down-arrow icons, Scroll to explore. The user is at the hero. They know how scroll works.

Decorative status dots

Zero by default. Only when conveying real semantic state, and at most one per page section.

border-t plus border-b on every row

Banned on long lists and spec tables. Reach for cards, tabs, scroll-snap pills, marquee, or carousel.

Div-based fake product UI

No fake task lists, terminals, or dashboards built out of styled divs. Use real screenshots or generated images.

Three-equal-card feature rows

Banned by default. Use 2-column zig-zag, asymmetric grids, or scroll-pinned alternatives.

AI-purple and mesh blob gradients

Banned by default. Neutral bases with one high-contrast accent. The LILA rule overrides only when the brand is explicitly purple.

Hand-rolled decorative SVG illustrations

Strongly discouraged as default. Acceptable only for a simple geometric mark or when the brief explicitly asks for it.

window.addEventListener('scroll')

Banned in JS. Use Motion useScroll, GSAP ScrollTrigger, IntersectionObserver, or CSS scroll-driven animations.

Customization

The SKILL.md file is fully editable. Open it in your project root and modify, add, or remove any rule to match your design direction.

The agent reads the file on each run, so edits apply on the next generation. If you maintain a project-specific style guide, paste it at the top of the SKILL.md and the agent will use it as the dominant source of truth.

Learn More

More material lives in the usage guide, the GitHub repo, and the changelog on GitHub.