Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Generate UX microcopy in StyleSeed's Toss-inspired voice for buttons, empty states, errors, toasts, confirmations, and form guidance.
.claude/skills/lingxling-ux-copy/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-09 | ✗→✓ | ▲ Improved | 42% | 0% |
| case-16 | ✗→✓ | ▲ Improved | 32% | 0% |
| case-20 | ✗→✓ | ▲ Improved | -57% | 0% |
| case-18 | ✓→✗ | ▼ Worse | 20% | 0% |
| case-02 | ✓→✓ | ＝ Same ✓ | 38% | 0% |
Part of StyleSeed, this skill generates concise product copy for common UI states. It follows the Toss-inspired tone: casual but polite, direct, active, and specific enough to help the user recover or proceed.
Use a short action verb plus object when needed.
Start with a friendly observation, then suggest the next action.
Explain what happened in user-facing language and what to do next. Do not surface raw internal error strings.
Confirm the result quickly. Add an undo action for reversible destructive behavior.
Use clear labels, useful placeholders, specific helper text, and corrective error messages.
State the action in plain language and explain the consequence if the decision is risky or irreversible.
Return:
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-02 | pass→pass | 11,220 | 9,513 | -15% | 1 | 1 | 0% | 1,508 | 2,085 | +38% | 0 | 0 | — |
case-01 | fail→fail | 9,579 | 21,254 | +122% | 1 | 1 | 0% | 1,460 | 1,693 | +16% | 0 | 0 | — |
case-03 | pass→pass | 9,439 | 8,276 | -12% | 1 | 1 | 0% | 1,410 | 1,886 | +34% | 0 | 0 | — |
case-04 | pass→pass | 8,943 | 10,123 | +13% | 1 | 1 | 0% | 1,408 | 1,852 | +32% | 0 | 0 | — |
case-05 | pass→pass | 8,068 | 11,764 | +46% | 1 | 1 | 0% | 1,373 | 2,023 | +47% | 0 | 0 | — |
case-06 | pass→pass | 27,291 | 15,764 | -42% | 1 | 1 | 0% | 2,816 | 2,712 | -4% | 0 | 0 | — |
case-07 | pass→pass | 8,405 | 11,493 | +37% | 1 | 1 | 0% | 1,365 | 2,065 | +51% | 0 | 0 | — |
case-08 | pass→pass | 12,306 | 12,081 | -2% | 1 | 1 | 0% | 1,697 | 2,192 | +29% | 0 | 0 | — |
case-09 | fail→pass | 9,283 | 13,558 | +46% | 1 | 1 | 0% | 1,659 | 2,363 | +42% | 0 | 0 | — |
case-10 | fail→fail | 6,911 | 6,521 | -6% | 1 | 1 | 0% | 1,068 | 1,501 | +41% | 0 | 0 | — |
case-11 | pass→pass | 10,615 | 8,347 | -21% | 1 | 1 | 0% | 1,378 | 1,912 | +39% | 0 | 0 | — |
case-12 | pass→pass | 6,391 | 10,719 | +68% | 1 | 1 | 0% | 1,085 | 1,850 | +71% | 0 | 0 | — |
case-13 | pass→pass | 13,295 | 11,302 | -15% | 1 | 1 | 0% | 1,493 | 2,089 | +40% | 0 | 0 | — |
case-14 | pass→pass | 7,483 | 8,625 | +15% | 1 | 1 | 0% | 1,045 | 1,671 | +60% | 0 | 0 | — |
case-15 | fail→fail | 11,167 | 9,738 | -13% | 1 | 1 | 0% | 1,393 | 1,963 | +41% | 0 | 0 | — |
case-16 | fail→pass | 9,570 | 8,307 | -13% | 1 | 1 | 0% | 1,504 | 1,979 | +32% | 0 | 0 | — |
case-17 | pass→pass | 13,556 | 12,293 | -9% | 1 | 1 | 0% | 1,718 | 1,824 | +6% | 0 | 0 | — |
case-18 | pass→fail | 11,137 | 7,789 | -30% | 1 | 1 | 0% | 1,491 | 1,790 | +20% | 0 | 0 | — |
case-19 | pass→pass | 19,785 | 14,418 | -27% | 1 | 1 | 0% | 2,134 | 2,706 | +27% | 0 | 0 | — |
case-20 | fail→pass | 29,217 | 11,863 | -59% | 1 | 1 | 0% | 5,268 | 2,260 | -57% | 0 | 0 | — |
case-21 | fail→fail | 18,801 | 21,339 | +13% | 1 | 1 | 0% | 4,001 | 3,833 | -4% | 0 | 0 | — |
case-22 | fail→fail | 31,482 | 20,884 | -34% | 1 | 1 | 0% | 4,726 | 3,836 | -19% | 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 +9 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.