Taste-Skill Variants Explained: Comparing v1, v2, gpt-taste, soft, minimalist, and brutalist

The six taste-skill variants differ fundamentally in motion engines, configuration systems, and visual constraints—ranging from v1's backward-compatible grid-first approach to brutalist's industrial blueprint aesthetic, each enforced via dedicated SKILL.md files in the Leonxlnx/taste-skill repository.

The Leonxlnx/taste-skill repository ships six distinct aesthetic engines, each residing in its own folder under skills/ and codifying a unique design philosophy through hard-coded rules. These taste-skill variants control everything from animation libraries to typography bans, making the choice between them critical for matching your project's technical stack and visual ambition.

v1 (taste-skill-v1): Legacy High-Agency Baseline

v1 (taste-skill-v1) maintains the original "high-agency" skill set for backward compatibility. According to skills/taste-skill-v1/SKILL.md, it enforces specific numeric dials: DESIGN_VARIANCE = 8, MOTION_INTENSITY = 6, and VISUAL_DENSITY = 4.

v1 relies exclusively on Framer Motion via motion/react—GSAP is prohibited. The variant mandates Tailwind v3/v4 alongside @phosphor-icons/react or @radix-ui/react-icons, and enforces a strict grid-first layout policy over flexbox. Notable constraints include an absolute ban on emojis, prohibition of the h-screen utility class, and strict icon spacing policies.

v2 (taste-skill): Experimental Anti-Slop Rewrite

v2 (taste-skill) shares the same baseline dial values as v1 (8/6/4) but introduces granular dial definitions and additional "anti-slop" guards. As defined in skills/taste-skill/SKILL.md, v2 defaults to Tailwind v4 and optionally integrates shadcn/ui components.

Key differentiators include an anti-center bias, eyebrow limitation (≤1 eyebrow per 3 sections), and stricter gradient policies. While v2 still prefers motion/react for animations, it permits GSAP only when explicitly invoked—unlike v1's total ban. The variant also expands its icon library support to include next/font and maintains the strict "no-emoji" policy.

gpt-taste: Awwwards-Level GSAP Motion

gpt-taste (gpt-tasteskill) forces an AIDA page structure (Attention, Interest, Desire, Action) and mandates GSAP-heavy motion through @gsap/react and ScrollTrigger. Configuration in skills/gpt-tasteskill/SKILL.md eschews numeric dials in favor of a mock Python RNG that randomizes layout, typography, and GSAP patterns per build.

This variant enforces scroll-pinning, stacking, and scrubbing behaviors as mandatory. It bans meta-labels (e.g., "SECTION 01"), emojis, and any grid that isn't gapless (grid-flow-dense). Hero sections must remain concise—limited to 2–3 lines maximum—focusing the visual language on cinematic scroll experiences rather than traditional component density.

soft: Agency-Grade Double-Bezel Architecture

soft (soft-skill) models $150k agency work through a variance engine that selects archetypes rather than numeric values. As specified in skills/soft-skill/SKILL.md, it chooses a Vibe (Ethereal Glass, Editorial Luxury, or Soft Structuralism) and a Layout (Asymmetrical Bento, Z-Axis Cascade, or Editorial Split).

Motion strictly uses Framer Motion with custom cubic-bezier curves—GSAP is never permitted. The variant implements a double-bezel component architecture requiring nested containers for depth. Bans include Inter font, generic icons, gradients, heavy shadows, emojis, and uniform grids, favoring instead rich micro-interactions and bespoke visual rhythms.

minimalist: Warm Monochrome Editorial

minimalist (minimalist-skill) provides an ultra-clean editorial UI with a warm monochrome palette and flat design constraints. The skills/minimalist-skill/SKILL.md hard-codes the visual language without dial systems, specifying muted pastel accents and zero gradients or heavy shadows.

Animation remains minimal, restricted to subtle fade-in effects via IntersectionObserver or motion/react using only transform and opacity properties. The variant bans Inter/Roboto fonts, generic icons, pill shapes, emojis, and bright primary backgrounds, enforcing a restrained, typographic-first approach.

brutalist: Industrial Blueprint Aesthetic

brutalist (brutalist-skill) targets data-heavy dashboards and "declassified blueprint" aesthetics through two visual archetypes: Swiss Industrial Print (light) or Tactical Telemetry (dark). Configuration in skills/brutalist-skill/SKILL.md relies on semantic HTML tags (<data>, <output>, <kbd>) and pure Tailwind with custom CSS.

Motion is optional; when implemented, it follows transform-only animation rules. Strict constraints include no gradients, no border-radius, a single mandatory accent red, optional terminal-green, and mandatory visible grid lines. Unlike other variants, brutalist avoids extra component libraries entirely, relying on semantic markup and stark visual hierarchy.

Installing and Switching Between Variants

Each taste-skill variant exposes a specific name field in its SKILL.md frontmatter, invoked via the skill install command. Use these commands to activate specific flavors:

# v1 - legacy high-agency baseline

skill install taste-skill-v1

# v2 - experimental default with granular dials

skill install design-taste-frontend

# gpt-taste - GSAP-driven AIDA structure

skill install gpt-taste

# soft - double-bezel agency aesthetic

skill install high-end-visual-design

# minimalist - warm monochrome flat UI

skill install minimalist-ui

# brutalist - industrial blueprint style

skill install industrial-brutalist-ui

The skill.sh helper script wires these names to their respective configuration files, automatically applying the default tech stacks, motion engines, and visual constraints defined in each variant's SKILL.md.

Key Configuration Files

Authoritative definitions for each taste-skill variant reside in specific markdown files under the skills/ directory:

• skills/taste-skill/SKILL.md — v2 reference with complete dial system and anti-slop rules
• skills/taste-skill-v1/SKILL.md — v1 backward-compatible baseline
• skills/gpt-tasteskill/SKILL.md — GSAP-centric AIDA structure with deterministic randomization
• skills/soft-skill/SKILL.md — Double-bezel architecture and vibe/layout archetypes
• skills/minimalist-skill/SKILL.md — Warm monochrome palette and flat-grid constraints
• skills/brutalist-skill/SKILL.md — Industrial visual system with strict grid and radius bans

Summary

• v1 maintains legacy compatibility with Framer Motion-only animations and strict grid-first layouts
• v2 expands granularity with anti-center bias, eyebrow caps, and optional shadcn/ui integration
• gpt-taste mandates GSAP scroll-pinning and AIDA page flow through randomized layout generation
• soft enforces agency-level double-bezel components with archetype-driven design selection
• minimalist restricts animations to fade-ins while enforcing warm monochrome flat design
• brutalist implements industrial aesthetics through semantic HTML, visible grids, and zero border-radius

Frequently Asked Questions

Which taste-skill variant supports GSAP animations?

Only gpt-taste mandates GSAP through @gsap/react and ScrollTrigger as a core requirement. While v2 permits GSAP if explicitly invoked, v1, soft, minimalist, and brutalist either prohibit it entirely or restrict motion to Framer Motion and CSS transforms.

What are the key differences between v1 and v2 taste-skill?

v2 introduces more granular dial definitions and "anti-slop" guards such as eyebrow limitations (≤1 per 3 sections) and anti-center bias, while v1 maintains the original baseline dials (DESIGN_VARIANCE = 8, MOTION_INTENSITY = 6, VISUAL_DENSITY = 4) with stricter Framer Motion exclusivity. Additionally, v2 defaults to Tailwind v4 and allows optional shadcn/ui components, whereas v1 targets Tailwind v3/v4 without shadcn integration.

Why does the brutalist variant ban border-radius and gradients?

The brutalist variant enforces an industrial "declassified blueprint" aesthetic that relies on hard-edge grids and functional typography to convey data-heavy or technical content. According to skills/brutalist-skill/SKILL.md, removing border-radius and gradients eliminates decorative以人为本" elements, ensuring the interface prioritizes raw information architecture and semantic HTML structure over visual polish.

Can I mix component libraries between different taste-skill variants?

While technically possible, each taste-skill variant defines specific primary component choices and bans in its SKILL.md that conflict with other variants. For example, soft explicitly bans generic icons and requires double-bezel containers, while brutalist forbids extra component libraries entirely. Mixing libraries risks violating the hard-coded constraints that define each variant's visual language and motion behavior.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}