Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Critique a screen's typography — scale usage, readability, consistency, and token compliance.
.claude/skills/owl-listener-critique-typography/SKILL.md| Model | Eval pass | Runs |
|---|---|---|
| gemini-3.6-flashlowest | 97% | 181 |
| gemini-3.1-pro-preview | 100% | 5 |
| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-14 | ✗→✓ | ▲ Improved | 15% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 8% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 15% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 42% | 0% |
| case-17 | ✗→✓ | ▲ Improved | 11% | 0% |
You are an expert in typographic systems and screen-level type critique.
You audit all typographic decisions on a screen: whether the type scale is applied correctly, whether text is readable at its context, whether type choices are consistent across the view, and whether design tokens are used in place of raw values. You flag problems and provide specific fixes.
Evaluate whether the type scale is applied as a system, not ad hoc.
Evaluate whether text can be read comfortably in its context.
Evaluate whether type decisions are uniform across the screen.
Evaluate whether typography tokens are applied instead of raw values.
For each dimension — Scale, Readability, Consistency, Token Compliance — provide:
Rate each dimension: pass / minor issue / major issue.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-15 | pass→pass | 8,287 | 7,287 | -12% | 1 | 1 | 0% | 1,328 | 1,841 | +39% | 0 | 0 | — |
case-14 | fail→pass | 15,132 | 11,255 | -26% | 1 | 1 | 0% | 2,624 | 3,007 | +15% | 0 | 0 | — |
case-01 | pass→pass | 10,309 | 14,683 | +42% | 1 | 1 | 0% | 2,090 | 3,946 | +89% | 0 | 0 | — |
case-02 | pass→pass | 12,838 | 18,528 | +44% | 1 | 1 | 0% | 2,377 | 4,176 | +76% | 0 | 0 | — |
case-03 | pass→pass | 4,438 | 10,259 | +131% | 1 | 1 | 0% | 855 | 2,636 | +208% | 0 | 0 | — |
case-04 | fail→pass | 13,860 | 10,893 | -21% | 1 | 1 | 0% | 2,419 | 2,620 | +8% | 0 | 0 | — |
case-05 | fail→pass | 14,525 | 13,089 | -10% | 1 | 1 | 0% | 2,575 | 2,966 | +15% | 0 | 0 | — |
case-06 | pass→pass | 9,935 | 12,551 | +26% | 1 | 1 | 0% | 1,806 | 3,138 | +74% | 0 | 0 | — |
case-07 | pass→pass | 8,923 | 9,881 | +11% | 1 | 1 | 0% | 1,888 | 2,962 | +57% | 0 | 0 | — |
case-08 | pass→pass | 10,174 | 15,087 | +48% | 1 | 1 | 0% | 1,800 | 3,176 | +76% | 0 | 0 | — |
case-09 | fail→pass | 11,387 | 11,193 | -2% | 1 | 1 | 0% | 2,013 | 2,856 | +42% | 0 | 0 | — |
case-10 | pass→pass | 15,319 | 11,090 | -28% | 1 | 1 | 0% | 2,802 | 3,020 | +8% | 0 | 0 | — |
case-11 | pass→pass | 7,501 | 10,453 | +39% | 1 | 1 | 0% | 1,641 | 2,870 | +75% | 0 | 0 | — |
case-12 | pass→pass | 9,155 | 12,588 | +37% | 1 | 1 | 0% | 1,769 | 2,953 | +67% | 0 | 0 | — |
case-13 | pass→pass | 8,662 | 11,095 | +28% | 1 | 1 | 0% | 1,725 | 2,935 | +70% | 0 | 0 | — |
case-16 | pass→pass | 7,759 | 11,091 | +43% | 1 | 1 | 0% | 1,343 | 2,811 | +109% | 0 | 0 | — |
case-17 | fail→pass | 19,639 | 17,286 | -12% | 1 | 1 | 0% | 3,497 | 3,889 | +11% | 0 | 0 | — |
case-18 | fail→pass | 17,738 | 16,359 | -8% | 1 | 1 | 0% | 3,093 | 3,921 | +27% | 0 | 0 | — |
case-19 | fail→pass | 22,560 | 13,100 | -42% | 1 | 1 | 0% | 4,007 | 3,241 | -19% | 0 | 0 | — |
case-20 | fail→pass | 14,597 | 14,067 | -4% | 1 | 1 | 0% | 2,424 | 3,160 | +30% | 0 | 0 | — |
case-21 | pass→pass | 7,576 | 8,916 | +18% | 1 | 1 | 0% | 1,449 | 2,453 | +69% | 0 | 0 | — |
case-22 | pass→pass | 6,243 | 10,452 | +67% | 1 | 1 | 0% | 1,284 | 2,952 | +130% | 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 +36 percentage points is the difference between those two pass rates over the 22 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.