Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Comply with Taiwan product-specific regulations — 食品 (食安法), 藥品 (藥事法), 化妝品 (化妝品衛生安全管理法), 保健食品, 酒類 (菸酒管理法), 醫療器材. Covers import permit, product registration, labeling, advertising restrictions, and e-commerce listing compliance. Use when selling any of these categories online. Do NOT use for general consumer protection (`tw-ecom-compliance-consumer`). STATUS: SKELETON — body pending.
.claude/skills/asgard-ai-platform-tw-ecom-compliance-product/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-09 | ✗→✓ | ▲ Improved | -10% | 0% |
| case-05 | ✓→✗ | ▼ Worse | 18% | 0% |
| case-01 | ✓→✓ | ＝ Same ✓ | 11% | 0% |
| case-02 | ✓→✓ | ＝ Same ✓ | 8% | 0% |
| case-03 | ✓→✓ | ＝ Same ✓ | -4% | 0% |
> STATUS: SKELETON — body pending.
tw-ecom-compliance-consumerTODO: category → governing law mapping, pre-market vs post-market obligations.
TODO: category → permit required? labeling required? ad restrictions?
TODO: registration flow, labeling checklist, ad-review process.
TODO: 5-6 pitfalls (薬事法 §69 广告限制, 保健食品 vs 食品 boundary, 化妝品 pre-market change, 酒類 minor verification, parallel-import restrictions).
TODO.
TODO.
tw-healthcare-regulations (for 醫療器材 depth)tw-ecom-compliance-consumer_Last verified: 2026-04_
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | pass→pass | 17,385 | 18,592 | +7% | 1 | 1 | 0% | 2,656 | 2,948 | +11% | 0 | 0 | — |
case-02 | pass→pass | 13,195 | 12,544 | -5% | 1 | 1 | 0% | 2,464 | 2,666 | +8% | 0 | 0 | — |
case-03 | pass→pass | 14,785 | 13,091 | -11% | 1 | 1 | 0% | 2,536 | 2,441 | -4% | 0 | 0 | — |
case-04 | pass→pass | 13,372 | 14,594 | +9% | 1 | 1 | 0% | 2,263 | 2,360 | +4% | 0 | 0 | — |
case-05 | pass→fail | 12,509 | 12,976 | +4% | 1 | 1 | 0% | 2,093 | 2,474 | +18% | 0 | 0 | — |
case-06 | pass→pass | 10,288 | 10,183 | -1% | 1 | 1 | 0% | 1,774 | 1,950 | +10% | 0 | 0 | — |
case-07 | pass→pass | 5,570 | 6,585 | +18% | 1 | 1 | 0% | 928 | 1,297 | +40% | 0 | 0 | — |
case-08 | pass→pass | 15,354 | 14,040 | -9% | 1 | 1 | 0% | 2,270 | 2,619 | +15% | 0 | 0 | — |
case-09 | fail→pass | 20,011 | 15,799 | -21% | 1 | 1 | 0% | 3,259 | 2,922 | -10% | 0 | 0 | — |
case-10 | pass→pass | 12,066 | 17,569 | +46% | 1 | 1 | 0% | 2,147 | 2,978 | +39% | 0 | 0 | — |
case-11 | pass→pass | 13,340 | 11,419 | -14% | 1 | 1 | 0% | 2,030 | 2,238 | +10% | 0 | 0 | — |
case-12 | pass→pass | 13,083 | 13,813 | +6% | 1 | 1 | 0% | 2,018 | 2,543 | +26% | 0 | 0 | — |
case-13 | pass→pass | 15,633 | 14,182 | -9% | 1 | 1 | 0% | 2,739 | 2,921 | +7% | 0 | 0 | — |
case-14 | pass→pass | 15,131 | 14,814 | -2% | 1 | 1 | 0% | 2,591 | 2,854 | +10% | 0 | 0 | — |
case-15 | pass→pass | 15,182 | 13,763 | -9% | 1 | 1 | 0% | 2,110 | 2,635 | +25% | 0 | 0 | — |
case-16 | pass→pass | 23,331 | 25,578 | +10% | 1 | 1 | 0% | 3,162 | 3,907 | +24% | 0 | 0 | — |
case-17 | pass→pass | 12,442 | 10,274 | -17% | 1 | 1 | 0% | 1,913 | 2,030 | +6% | 0 | 0 | — |
case-18 | pass→pass | 13,116 | 12,105 | -8% | 1 | 1 | 0% | 1,849 | 2,407 | +30% | 0 | 0 | — |
case-19 | pass→pass | 10,251 | 7,938 | -23% | 1 | 1 | 0% | 1,649 | 1,773 | +8% | 0 | 0 | — |
case-20 | pass→pass | 11,363 | 10,008 | -12% | 1 | 1 | 0% | 1,860 | 1,996 | +7% | 0 | 0 | — |
case-21 | pass→pass | 9,727 | 11,047 | +14% | 1 | 1 | 0% | 1,742 | 2,256 | +30% | 0 | 0 | — |
case-22 | pass→pass | 18,517 | 18,606 | +0% | 1 | 1 | 0% | 2,949 | 3,495 | +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 0 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.