Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use when generating detailed, technical UI prompts for designers or AI tools (v0, Bolt) that require precise specifications, tokens, and comprehensive state definitions.
.claude/skills/valec3-ui-prompt-generation/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-20 | ✗→✓ | ▲ Improved | 188% | 0% |
| case-13 | ✓→✗ | ▼ Worse | 200% | 0% |
| case-16 | ✓→✗ | ▼ Worse | 33% | 0% |
| case-19 | ✓→✗ | ▼ Worse | -66% | 0% |
| case-21 | ✓→✗ | ▼ Worse | 178% | 0% |
UI Prompt Generation IS the translation of user intent into precise, token-based technical specifications.
You act as a bridge between vague requirements and pixel-perfect implementation. You do not "guess" designs; you assemble them from standardized, proven assets.
Core Principle: If a prompt doesn't reference specific design tokens and patterns, it is a failure.
Use this skill when: Use this skill when:
Don't use when:
ui-design-concepts)Before (Vague): > "Make a login page. It should look clean."
After (Standardized): > Agent quietly generates results/login-prompt.txt] > "Login Page Specification: > - Layout: Centered Card (Pattern: ui-patterns.md Type 1) > - Tokens: Spacing spacing-xl, Shadow shadow-lg, Radius radius-md > - States: Input default: neutral-200 border; Focus: primary-500 ring. > - Accessibility: High contrast text (neutral-900), explicit labels."
CRITICAL RULE: OUTPUT MUST BE A PLAIN TEXT .txt FILE IN results/. NEVER use Markdown syntax (no #, no **, no [ ], no code blocks) inside the .txt file. ALWAYS use clean Plain Text formatting:
CRITICAL RULE: BE EXHAUSTIVE & VERBOSE.
view_file on these files before generating:assets/design-tokens.md (Get exact color/spacing values)assets/ui-patterns.md (Select the structural pattern)assets/component-checklist.md (Ensure all states are covered)assets/prompt-templates.md (Use a base structure if applicable)Build the prompt using the data from the assets. Your prompt must include:
primary-500, not "blue").results/<descriptive-name>.txt.results/<filename>.txt".Bad: "Use a light gray background." Good: "Background: neutral-50." Fix: Always have design-tokens.md open.
Bad: "A button that submits." Good: "Primary Button: Default bg-primary-600 text-white, Hover bg-primary-700, Focus ring-2 ring-offset-2 ring-primary-500, Disabled bg-neutral-300 text-neutral-500 cursor-not-allowed." Fix: If you aren't listing at least 3 states, you are failing. Consult component-checklist.md.
Bad: Printing the prompt in the extensive markdown block in chat. Good: "Saved to results/prompt.txt." Fix: Check your output tool. If it's not write_to_file, stop.
| Asset | Purpose | | :--- | :--- | | design-tokens.md | REQUIRED. Colors, spacing, typography, shadows. | | ui-patterns.md | REQUIRED. Layouts, cards, navigation structures. | | component-checklist.md | REQUIRED. States, accessibility, edge cases. | | prompt-templates.md | Starter skeletons for common pages. |
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 30,385 | 40,563 | +33% | 1 | 1 | 0% | 4,795 | 7,442 | +55% | 0 | 0 | — |
case-02 | fail→fail | 24,804 | 9,107 | -63% | 1 | 1 | 0% | 4,828 | 2,862 | -41% | 0 | 0 | — |
case-03 | fail→fail | 28,700 | 42,972 | +50% | 1 | 1 | 0% | 6,192 | 11,120 | +80% | 0 | 0 | — |
case-04 | pass→pass | 10,952 | 7,001 | -36% | 1 | 1 | 0% | 1,947 | 2,519 | +29% | 0 | 0 | — |
case-05 | pass→pass | 25,797 | 27,558 | +7% | 1 | 1 | 0% | 4,444 | 7,412 | +67% | 0 | 0 | — |
case-06 | pass→pass | 4,244 | 2,948 | -31% | 1 | 1 | 0% | 748 | 1,762 | +136% | 0 | 0 | — |
case-07 | fail→fail | 19,804 | 8,721 | -56% | 1 | 1 | 0% | 4,239 | 1,538 | -64% | 0 | 0 | — |
case-08 | fail→fail | 18,190 | 6,104 | -66% | 1 | 1 | 0% | 3,355 | 1,698 | -49% | 0 | 0 | — |
case-09 | fail→fail | 24,969 | 13,339 | -47% | 1 | 1 | 0% | 4,584 | 1,805 | -61% | 0 | 0 | — |
case-10 | fail→fail | 53,982 | 5,853 | -89% | 1 | 1 | 0% | 5,505 | 1,709 | -69% | 0 | 0 | — |
case-11 | pass→pass | 14,302 | 29,707 | +108% | 1 | 1 | 0% | 2,635 | 7,418 | +182% | 0 | 0 | — |
case-12 | fail→fail | 21,938 | 7,193 | -67% | 1 | 1 | 0% | 2,912 | 1,660 | -43% | 0 | 0 | — |
case-13 | pass→fail | 22,367 | 29,889 | +34% | 1 | 1 | 0% | 2,342 | 7,018 | +200% | 0 | 0 | — |
case-14 | fail→fail | 11,971 | 28,388 | +137% | 1 | 1 | 0% | 2,852 | 1,617 | -43% | 0 | 0 | — |
case-15 | fail→fail | 16,091 | 7,054 | -56% | 1 | 1 | 0% | 2,904 | 1,838 | -37% | 0 | 0 | — |
case-16 | pass→fail | 22,528 | 24,865 | +10% | 1 | 1 | 0% | 4,405 | 5,866 | +33% | 0 | 0 | — |
case-17 | fail→fail | 25,500 | 38,446 | +51% | 1 | 1 | 0% | 3,683 | 7,409 | +101% | 0 | 0 | — |
case-18 | fail→fail | 14,570 | 7,463 | -49% | 1 | 1 | 0% | 2,702 | 1,841 | -32% | 0 | 0 | — |
case-19 | pass→fail | 25,114 | 6,030 | -76% | 1 | 1 | 0% | 5,403 | 1,817 | -66% | 0 | 0 | — |
case-20 | fail→pass | 13,639 | 32,467 | +138% | 1 | 1 | 0% | 2,625 | 7,560 | +188% | 0 | 0 | — |
case-21 | pass→fail | 28,485 | 35,212 | +24% | 1 | 1 | 0% | 2,917 | 8,108 | +178% | 0 | 0 | — |
case-22 | pass→fail | 26,140 | 5,177 | -80% | 1 | 1 | 0% | 5,427 | 1,648 | -70% | 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, and 11 counted toward the lift figure. The other 11 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of -18 percentage points is the difference between those two pass rates over the 11 comparable cases. 10 cases got worse with the skill loaded, and they are 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.