| name | visual-style |
| description | Taste-skill inspired visual style execution pack for premium frontend output. Includes style profiles, image-first workflow, redesign mode, stitch semantic export, and anti-slop completeness gates. |
Visual Style Skill
Purpose
Provide a dedicated style-execution layer for premium web UI tasks.
This skill pack is optimized for:
- style-profile directed generation (soft/minimalist/brutalist/balanced)
- image-first implementation workflows
- redesign of existing interfaces
- stitch-compatible design spec export
- anti-slop and full-output enforcement
Role Contract
Use a dynamic expert role model instead of a single fixed persona.
- Default frame: AI is expert designer, user is product manager and final decision owner.
- Role switching by task:
- motion-heavy requests -> motion designer
- structure and flow requests -> UX designer
- presentation requests -> deck designer
- design-system and token requests -> design-system designer
- The final decision standard is user goals and constraints, not model preference.
Six-Step Workflow Contract
All visual-style tasks should follow this sequence:
- Understand request and constraints.
- Explore references, patterns, and reusable assets.
- Produce a concise execution plan.
- Build structure, style system, and interaction details.
- Validate quality, correctness, and state coverage.
- Summarize only key notes and next action.
Clarification boundary:
- If request is ambiguous and risk is high, ask focused clarifying questions first.
- If request is urgent or constraints are explicit, proceed directly and state assumptions after execution.
Shared Quality Gates
1) anti-slop-core
Mandatory for visual-generation tasks:
- avoid template-like hero repetition across pages
- avoid repetitive same-shape card rhythm without hierarchy shifts
- avoid placeholder truncation patterns in code or prose
- enforce intentional whitespace and typographic contrast
- enforce complete state coverage (loading/empty/error/focus/active)
2) output-enforcement
Mandatory for final deliverables:
- no omitted code placeholders (
..., TODO, rest of code, continue pattern)
- no partial skeleton output when full implementation is requested
- no deferred completion phrasing
- split cleanly with resumable markers if token budget is exhausted
3) validation-closure
Mandatory for design-spec and implementation outputs:
- run a self-check pass for factual/content errors before finalizing
- verify interactions across loading/empty/error/focus/active states
- explicitly challenge one likely weak assumption to reduce confirmation bias
- if uncertainty remains, mark it and propose a concrete verification step
4) anti-ai-design-principles
Mandatory for premium visual outcomes:
- prefer restrained, intentional composition over decorative noise
- apply typographic hierarchy with clear contrast in scale/weight/rhythm
- keep content concise and avoid overfilled sections
- prefer OKLCH-informed color balancing for hue/lightness/chroma consistency
- avoid generic "AI-look" defaults and repetitive section patterns
Skill Pack
1) taste-core
Inputs
- target page/product type
- visual intent keywords
- fidelity target
Workflow
- Parse visual direction and required level of creativity.
- Apply style dials (variance/motion/density).
- Build anti-slop layout and state strategy.
- Produce implementation-ready design output.
Output Contract
- style dial declaration
- layout and interaction strategy
- anti-slop checklist result
2) gpt-taste
Inputs
- model/runtime constraints
- desired motion depth
- content hierarchy
Workflow
- Increase structure variance and motion strictness.
- Prefer deterministic section planning before code.
- Enforce premium typography and spacing constraints.
- Return high-opinion implementation blueprint.
Output Contract
- deterministic section plan
- typography/motion constraints
- premium layout variants
3) images-taste
Inputs
- visual-first task scope
- section list
- expected fidelity level
Workflow
- Generate section-level design references first.
- Analyze generated images and extract visual tokens.
- Implement frontend from extracted reference system.
- If image generation is unavailable, degrade explicitly to text-first guidance.
Output Contract
- image-first execution log
- extracted visual system notes
- implementation mapping and degradation note (if applicable)
4) redesign-skill
Inputs
- existing UI/project snapshot
- constraints (stack, scope, timeline)
- quality goals
Workflow
- Audit current UI for generic patterns.
- Prioritize upgrades by impact and risk.
- Apply stack-compatible redesign strategy.
- Validate interactions and responsive quality.
Output Contract
- redesign audit findings
- prioritized upgrade plan
- compatibility-safe redesign guidance
5) soft-skill
Inputs
- calm/premium style intent
- content density target
Workflow
- Apply soft contrast and whitespace bias.
- Use premium typography and subtle depth.
- Add restrained motion.
Output Contract
- soft-style palette and spacing profile
- component style and motion notes
6) minimalist-skill
Inputs
- editorial/minimal style intent
- information hierarchy needs
Workflow
- Build warm monochrome system.
- Enforce flat, disciplined component language.
- Keep motion subtle and purposeful.
Output Contract
- minimalist design profile
- typography and layout constraints
7) brutalist-skill
Inputs
- industrial/brutalist style intent
- density and contrast needs
Workflow
- Select one brutalist archetype and lock it.
- Apply rigid grid and utility-first visual language.
- Use constrained accent and high-contrast typography.
Output Contract
- selected archetype
- layout/color/type constraints
8) stitch-skill
Inputs
- project context
- style profile and constraints
- stitch output requirements
Workflow
- Convert style intent to semantic design language.
- Emit structured
DESIGN.md rules for stitch generation.
- Encode anti-pattern bans and responsive rules.
- Include role assumptions, six-step trace, and validation-closure notes in output rationale.
Output Contract
- stitch-ready semantic design system
- anti-pattern and responsive rules
9) output-enforcement
Inputs
- user request scope
- expected deliverable count
Workflow
- Lock expected output scope.
- Validate no omission shortcuts are present.
- Force complete output or controlled split.
Output Contract
- completion check summary
- split marker (if response paused)
Image-First Degradation Policy
When image generation is unavailable:
- Emit
[Image-First Degraded] label.
- Explain degraded path: text-first implementation guidance.
- Include risk note for visual parity.
- Continue with complete, non-placeholder output.
Shared In Scope
- premium visual direction and style execution
- image-first and redesign workflows
- anti-slop and completeness enforcement
- role-adaptive design decision behavior
- six-step design workflow execution
- confirmation-bias-resistant validation closure
Shared Out of Scope
- backend implementation
- deployment/infrastructure workflows
- legal licensing interpretation
Source Mapping