Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Create or edit HTML designs inside an active iPolloWork Design Studio session while preserving its template, selection, theme tokens, and project boundaries.
.claude/skills/devin-axis-ipollowork-design-studio/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-15 | ✗→✓ | ▲ Improved | -44% | 0% |
| case-16 | ✗→✓ | ▲ Improved | -3% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -18% | 0% |
| case-09 | ✗→✓ | ▲ Improved | -38% | 0% |
| case-13 | ✗→✓ | ▲ Improved | -33% | 0% |
Use this Skill only for a design project already owned by the active iPolloWork session. The built-in Design Studio, templates, editor, undo history, and exports exist independently of this Skill.
design-tokens.css before editing.design/<session-id>/ project.design-tokens.css link.--ipw-* theme tokens for themeable colors, typography, spacing, radii, shadows, surfaces, and page dimensions so the Design System controls keep working.If the active session provides stricter instructions, those instructions take precedence.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-15 | fail→pass | 15,173 | 9,473 | -38% | 1 | 1 | 0% | 1,948 | 1,085 | -44% | 0 | 0 | — |
case-16 | fail→pass | 15,860 | 12,302 | -22% | 1 | 1 | 0% | 1,508 | 1,456 | -3% | 0 | 0 | — |
case-01 | fail→fail | 7,418 | 16,340 | +120% | 1 | 1 | 0% | 384 | 2,305 | +500% | 0 | 0 | — |
case-02 | fail→fail | 22,790 | 15,068 | -34% | 1 | 1 | 0% | 3,701 | 654 | -82% | 0 | 0 | — |
case-03 | fail→fail | 13,168 | 14,818 | +13% | 1 | 1 | 0% | 1,233 | 619 | -50% | 0 | 0 | — |
case-04 | fail→pass | 15,777 | 11,122 | -30% | 1 | 1 | 0% | 1,681 | 1,386 | -18% | 0 | 0 | — |
case-05 | fail→fail | 11,061 | 7,057 | -36% | 1 | 1 | 0% | 944 | 747 | -21% | 0 | 0 | — |
case-06 | fail→fail | 13,351 | 9,167 | -31% | 1 | 1 | 0% | 1,387 | 1,194 | -14% | 0 | 0 | — |
case-07 | pass→pass | 11,937 | 9,533 | -20% | 1 | 1 | 0% | 1,027 | 1,110 | +8% | 0 | 0 | — |
case-08 | pass→pass | 10,939 | 8,842 | -19% | 1 | 1 | 0% | 980 | 966 | -1% | 0 | 0 | — |
case-09 | fail→pass | 16,511 | 8,923 | -46% | 1 | 1 | 0% | 1,754 | 1,083 | -38% | 0 | 0 | — |
case-10 | fail→fail | 12,589 | 10,702 | -15% | 1 | 1 | 0% | 1,396 | 1,363 | -2% | 0 | 0 | — |
case-11 | fail→fail | 14,199 | 8,127 | -43% | 1 | 1 | 0% | 1,412 | 876 | -38% | 0 | 0 | — |
case-12 | pass→pass | 11,527 | 7,970 | -31% | 1 | 1 | 0% | 947 | 887 | -6% | 0 | 0 | — |
case-13 | fail→pass | 14,351 | 9,013 | -37% | 1 | 1 | 0% | 1,571 | 1,047 | -33% | 0 | 0 | — |
case-14 | fail→pass | 16,605 | 9,418 | -43% | 1 | 1 | 0% | 1,834 | 1,017 | -45% | 0 | 0 | — |
case-17 | fail→pass | 18,279 | 11,920 | -35% | 1 | 1 | 0% | 2,207 | 1,074 | -51% | 0 | 0 | — |
case-18 | fail→fail | 11,399 | 9,510 | -17% | 1 | 1 | 0% | 937 | 1,013 | +8% | 0 | 0 | — |
case-19 | fail→pass | 14,393 | 9,263 | -36% | 1 | 1 | 0% | 1,428 | 1,058 | -26% | 0 | 0 | — |
case-20 | pass→pass | 11,366 | 7,796 | -31% | 1 | 1 | 0% | 1,125 | 844 | -25% | 0 | 0 | — |
case-21 | fail→fail | 10,509 | 8,839 | -16% | 1 | 1 | 0% | 935 | 1,042 | +11% | 0 | 0 | — |
case-22 | pass→pass | 8,760 | 8,163 | -7% | 1 | 1 | 0% | 597 | 883 | +48% | 0 | 0 | — |
case-23 | fail→pass | 11,680 | 11,551 | -1% | 1 | 1 | 0% | 1,105 | 1,596 | +44% | 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, and 21 counted toward the lift figure. The other 2 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 +39 percentage points is the difference between those two pass rates over the 21 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.