vs-design-diverge

This skill guides the creation of distinctive, production-grade frontend interfaces by explicitly mitigating "Mode Collapse" (the tendency to produce generic AI-slop). It uses Verbalized Sampling logic to unlock LLM creativity and deliver high-entropy, unforgettable designs.

## Phase 0: Context Discovery (MANDATORY)

BEFORE any design analysis, you MUST gather deep context from the user. Use the `AskUserQuestion` tool to probe the following dimensions:

### Discovery Dimensions

1. **Emotional Tone**: What feeling or atmosphere should the design evoke? (e.g., "trustworthy", "edgy", "playful", "luxurious")
2. **Target Audience**: Who will see this? What are their expectations and technical sophistication?
3. **Reference Points / Anti-References**: What does the user want to emulate? What do they explicitly want to avoid?
4. **Business Context**: What problem does this UI solve? What's the usage scenario?

### Context Signals

- **Analyze existing code** if present—extract style patterns, color schemes, component conventions
- **Parse user prompt** for implicit context keywords (landing page, dashboard, portfolio, SaaS, etc.)
- **Ask follow-up questions** even if the initial prompt seems simple—surface the full vision in the user's mind

**CRITICAL**: Do not proceed to Phase 1 until you have sufficient context. A vague prompt requires MORE questions, not assumptions.

---

## Phase 1: Identify the Mode (The Generic Baseline)

- Verbalize the most predictable, high-probability (P ≈ 0.95) design for this request.
- Identify "AI-slop" markers based on the context. Common examples include:
  - Inter/Roboto/System fonts
  - Rounded blue/purple buttons
  - Standard F-pattern layouts
  - White backgrounds with gradient accents
  - Generic hero sections with stock-style imagery
- **CRITICAL**: You are forbidden from choosing this baseline.

---

## Phase 2: Sample the Long-Tail (Probability Mapping)

Generate three distinct aesthetic directions and assign a **"Typicality Sco