Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use when you need design-first, spec-driven, skimmable prompts for UI generation. Covers prompt structure, constraints, variations, typography/spacing rules, and iteration workflow for consistent UI outputs.
.claude/skills/mengto-design-first-ui-prompting/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-03 | ✗→✓ | ▲ Improved | 24% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 20% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 14% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 43% | 0% |
| case-16 | ✗→✓ | ▲ Improved | 21% | 0% |
This skill is for design-first prompting: turn fuzzy ideas into a tight spec that produces consistent UI.
Prompt like a design system, not a wish.
Use this skeleton, then fill the blanks.
textGOAL - What are we making? (e.g., landing page hero / onboarding / dashboard / carousel slide) - Who is it for? (persona) - What’s the success criteria? (clarity, conversion, vibe) FORMAT - Size/aspect: (e.g., 1080x1350) - Safe margins: (e.g., 90px) LAYOUT (wireframe in words) - Grid: (e.g., Swiss 6-col) - Placement: (e.g., type-left / image-right) - Hierarchy: H1 → subhead → body → CTA TYPE SYSTEM - Font vibe: (e.g., Söhne / Neue Haas / SF Pro) - Weights: (H1 700, body 400) - Leading: (tight for H1, readable for body) - Tracking: (micro labels wider) COLOR + MATERIAL - Background: (hex or description) - Text: (white/ivory/charcoal) - One accent only: (cyan/lime/purple) - Texture: (subtle grain, no plastic HDR) IMAGERY / UI STYLE - UI style: (minimal / glass / editorial / playful 3D) - If photo: lighting + crop + texture rules - If 3D: materials + lighting + softness COPY (render EXACTLY) - Line 1: - Line 2: - ... CONSTRAINTS (change 1–2 things only) - FONT: ___ - STYLE: ___ - MODE: ___ NEGATIVE PROMPT - No logos, no watermarks - No extra text beyond provided lines - No gibberish typography
If the model keeps misspelling:
1) Generate without text (reserve a clean text-safe area) 2) Typeset in Figma
When you want the model to obey a style:
Example:
textConstraints FONT CANELA STYLE MINIMAL MODE DARK
Don’t ask the model to “remember” taste.
refs/...| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 13,084 | 9,531 | -27% | 1 | 1 | 0% | 2,381 | 2,605 | +9% | 0 | 0 | — |
case-02 | fail→fail | 15,251 | 10,993 | -28% | 1 | 1 | 0% | 2,681 | 2,661 | -1% | 0 | 0 | — |
case-03 | fail→pass | 12,316 | 10,357 | -16% | 1 | 1 | 0% | 2,261 | 2,811 | +24% | 0 | 0 | — |
case-04 | pass→pass | 7,758 | 10,421 | +34% | 1 | 1 | 0% | 1,819 | 3,289 | +81% | 0 | 0 | — |
case-05 | pass→pass | 12,025 | 12,261 | +2% | 1 | 1 | 0% | 2,330 | 3,150 | +35% | 0 | 0 | — |
case-06 | pass→fail | 10,280 | 11,221 | +9% | 1 | 1 | 0% | 1,881 | 2,739 | +46% | 0 | 0 | — |
case-07 | fail→pass | 9,840 | 7,003 | -29% | 1 | 1 | 0% | 1,712 | 2,062 | +20% | 0 | 0 | — |
case-08 | fail→pass | 13,880 | 10,259 | -26% | 1 | 1 | 0% | 2,243 | 2,549 | +14% | 0 | 0 | — |
case-09 | fail→fail | 10,116 | 7,709 | -24% | 1 | 1 | 0% | 1,852 | 2,184 | +18% | 0 | 0 | — |
case-10 | fail→pass | 10,884 | 9,252 | -15% | 1 | 1 | 0% | 1,691 | 2,423 | +43% | 0 | 0 | — |
case-11 | fail→fail | 12,659 | 11,162 | -12% | 1 | 1 | 0% | 2,381 | 2,648 | +11% | 0 | 0 | — |
case-12 | fail→fail | 10,215 | 4,865 | -52% | 1 | 1 | 0% | 1,779 | 1,682 | -5% | 0 | 0 | — |
case-13 | pass→pass | 13,246 | 10,318 | -22% | 1 | 1 | 0% | 2,108 | 2,482 | +18% | 0 | 0 | — |
case-14 | pass→pass | 13,309 | 9,025 | -32% | 1 | 1 | 0% | 2,336 | 2,365 | +1% | 0 | 0 | — |
case-15 | pass→pass | 12,890 | 11,113 | -14% | 1 | 1 | 0% | 2,291 | 2,640 | +15% | 0 | 0 | — |
case-16 | fail→pass | 12,269 | 9,331 | -24% | 1 | 1 | 0% | 2,071 | 2,500 | +21% | 0 | 0 | — |
case-17 | pass→pass | 12,323 | 7,026 | -43% | 1 | 1 | 0% | 2,045 | 2,026 | -1% | 0 | 0 | — |
case-18 | fail→pass | 12,932 | 9,445 | -27% | 1 | 1 | 0% | 2,255 | 2,321 | +3% | 0 | 0 | — |
case-19 | pass→pass | 12,892 | 10,719 | -17% | 1 | 1 | 0% | 2,183 | 2,647 | +21% | 0 | 0 | — |
case-20 | fail→pass | 16,168 | 8,110 | -50% | 1 | 1 | 0% | 2,765 | 2,245 | -19% | 0 | 0 | — |
case-21 | pass→pass | 12,193 | 9,011 | -26% | 1 | 1 | 0% | 2,384 | 2,500 | +5% | 0 | 0 | — |
case-22 | pass→pass | 10,961 | 5,108 | -53% | 1 | 1 | 0% | 1,857 | 1,650 | -11% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted. The headline lift of +27 percentage points is the difference between those two pass rates over the 22 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.