Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Visual design taste for web UI, distilled from a stronger model's aesthetic - invoke when about to BUILD or STYLE any web page, component, artifact, or UI; picking a visual direction; writing CSS; reviewing how a design looks; or when a design works but feels generic or boring. Named style directions with exact tokens, concrete craft numbers, and a catalog of garbage design to refuse.
.claude/skills/telagod-frontend/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | 167% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -33% | 0% |
| case-15 | ✗→✓ | ▲ Improved | -47% | 0% |
| case-19 | ✗→✓ | ▲ Improved | 4% | 0% |
| case-20 | ✗→✓ | ▲ Improved | -15% | 0% |
Rule content lives in the five files below; this SKILL.md only routes (duplicating rule text here is drift — doctrine/04-maintenance.md governs edits to this bundle too).
| You are about to… | Read (in this folder) | |---|---| | Start any page/component/artifact — brief to shipped | process.md first; it routes onward | | Pick a visual direction, or choose between styles | styles.md | | Understand what makes a design striking vs mediocre; make a boring design memorable | taste.md | | Write actual CSS values — type, space, color, depth, motion, dark mode | craft.md | | Judge a design (yours or anyone's); name precisely why something feels cheap | garbage.md |
Start with the ONE file that matches and follow its cross-references — they are required inputs, not optional reading. Starting from scratch usually means process.md + the two files it points you to.
This bundle is the visual/aesthetic layer only. Server-side design (stack, APIs, data) → the backend bundle; whether to delegate → doctrine; how to think → methods. If the environment provides an artifact-design skill (harness-side), it calibrates how much to invest per request; THIS bundle governs direction and craft once you're building.
Left unguided, every model produces the same page (garbage.md §A names it precisely). What taste means here — and why a clean-but-forgettable page counts as a failure — is taste.md's opening; read it before judging any design.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-11 | fail→fail | 16,310 | 18,758 | +15% | 1 | 1 | 0% | 2,861 | 3,570 | +25% | 0 | 0 | — |
case-04 | pass→pass | 14,119 | 14,214 | +1% | 1 | 1 | 0% | 2,189 | 2,439 | +11% | 0 | 0 | — |
case-01 | fail→fail | 23,085 | 19,978 | -13% | 1 | 1 | 0% | 4,609 | 4,151 | -10% | 0 | 0 | — |
case-02 | fail→pass | 19,969 | 41,080 | +106% | 1 | 1 | 0% | 3,101 | 8,291 | +167% | 0 | 0 | — |
case-03 | pass→pass | 20,535 | 22,932 | +12% | 1 | 1 | 0% | 3,293 | 4,273 | +30% | 0 | 0 | — |
case-05 | pass→pass | 18,277 | 16,958 | -7% | 1 | 1 | 0% | 3,127 | 3,283 | +5% | 0 | 0 | — |
case-06 | fail→pass | 14,645 | 6,833 | -53% | 1 | 1 | 0% | 2,314 | 1,555 | -33% | 0 | 0 | — |
case-07 | fail→fail | 20,197 | 19,885 | -2% | 1 | 1 | 0% | 3,129 | 3,563 | +14% | 0 | 0 | — |
case-08 | fail→fail | 15,354 | 14,933 | -3% | 1 | 1 | 0% | 2,169 | 2,545 | +17% | 0 | 0 | — |
case-09 | fail→fail | 22,723 | 25,437 | +12% | 1 | 1 | 0% | 3,590 | 4,053 | +13% | 0 | 0 | — |
case-10 | fail→fail | 20,415 | 18,415 | -10% | 1 | 1 | 0% | 3,310 | 3,630 | +10% | 0 | 0 | — |
case-12 | fail→fail | 15,811 | 17,305 | +9% | 1 | 1 | 0% | 2,606 | 3,212 | +23% | 0 | 0 | — |
case-13 | fail→fail | 22,149 | 21,577 | -3% | 1 | 1 | 0% | 3,782 | 4,240 | +12% | 0 | 0 | — |
case-14 | fail→fail | 16,964 | 17,147 | +1% | 1 | 1 | 0% | 2,785 | 3,363 | +21% | 0 | 0 | — |
case-15 | fail→pass | 17,348 | 6,195 | -64% | 1 | 1 | 0% | 2,605 | 1,392 | -47% | 0 | 0 | — |
case-16 | pass→pass | 10,509 | 4,528 | -57% | 1 | 1 | 0% | 1,622 | 1,101 | -32% | 0 | 0 | — |
case-17 | pass→pass | 9,639 | 2,506 | -74% | 1 | 1 | 0% | 1,382 | 824 | -40% | 0 | 0 | — |
case-18 | fail→fail | 14,778 | 15,267 | +3% | 1 | 1 | 0% | 2,174 | 2,520 | +16% | 0 | 0 | — |
case-19 | fail→pass | 13,183 | 12,007 | -9% | 1 | 1 | 0% | 1,971 | 2,040 | +4% | 0 | 0 | — |
case-20 | fail→pass | 10,057 | 5,513 | -45% | 1 | 1 | 0% | 1,583 | 1,340 | -15% | 0 | 0 | — |
case-21 | pass→pass | 14,221 | 15,755 | +11% | 1 | 1 | 0% | 2,280 | 2,931 | +29% | 0 | 0 | — |
case-22 | fail→fail | 17,894 | 22,438 | +25% | 1 | 1 | 0% | 3,176 | 4,370 | +38% | 0 | 0 | — |
case-23 | pass→pass | 10,111 | 11,234 | +11% | 1 | 1 | 0% | 1,718 | 1,974 | +15% | 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. 23 cases were attempted. The headline lift of +22 percentage points is the difference between those two pass rates over the 23 comparable cases.
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.