Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Ship refrigerated / frozen products in Taiwan — 宅配通 宅配通冷藏 / 黑貓宅急便 低溫, cold-chain CVS pickup limitations, packaging (保冷劑 / 乾冰), and shelf-life SLA. Use for 生鮮 / 冷凍 / 冰品 / 藥品 delivery. Do NOT use for ambient-temperature shipping. STATUS: SKELETON — body pending.
.claude/skills/asgard-ai-platform-tw-ecom-logistics-cold-chain/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-12 | ✗→✓ | ▲ Improved | 43% | 0% |
| case-21 | ✗→✓ | ▲ Improved | 1% | 0% |
| case-01 | ✓→✓ | ＝ Same ✓ | -45% | 0% |
| case-02 | ✓→✓ | ＝ Same ✓ | 17% | 0% |
| case-03 | ✓→✓ | ＝ Same ✓ | 8% | 0% |
> STATUS: SKELETON — body pending.
tw-ecom-logistics-home or -cvstw-ecom-compliance-productTODO: 冷藏 (0-7°C) vs 冷凍 (-18°C), packaging physics, cost structure.
TODO: cold-chain required? carrier choice? package spec?
TODO: vendor onboarding, test shipments, insurance, loss SOP.
TODO: 5-6 pitfalls (Friday / weekend hold risk, 離島 no cold service, CVS cold-chain capacity, insurance claim friction, 乾冰 export restriction).
TODO.
TODO.
tw-ecom-logistics-hometw-ecom-compliance-product_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 | 47,907 | 38,215 | -20% | 1 | 1 | 0% | 2,800 | 1,533 | -45% | 0 | 0 | — |
case-02 | pass→pass | 26,737 | 25,336 | -5% | 1 | 1 | 0% | 3,806 | 4,437 | +17% | 0 | 0 | — |
case-03 | pass→pass | 21,680 | 19,598 | -10% | 1 | 1 | 0% | 2,808 | 3,045 | +8% | 0 | 0 | — |
case-04 | pass→pass | 14,185 | 9,877 | -30% | 1 | 1 | 0% | 2,262 | 1,768 | -22% | 0 | 0 | — |
case-05 | pass→pass | 8,769 | 34,447 | +293% | 1 | 1 | 0% | 1,202 | 1,491 | +24% | 0 | 0 | — |
case-11 | pass→pass | 13,138 | 11,635 | -11% | 1 | 1 | 0% | 1,942 | 2,263 | +17% | 0 | 0 | — |
case-06 | pass→pass | 6,437 | 10,016 | +56% | 1 | 1 | 0% | 999 | 1,605 | +61% | 0 | 0 | — |
case-07 | pass→pass | 9,012 | 11,741 | +30% | 1 | 1 | 0% | 1,405 | 1,854 | +32% | 0 | 0 | — |
case-08 | pass→pass | 24,153 | 23,327 | -3% | 1 | 1 | 0% | 3,082 | 3,313 | +7% | 0 | 0 | — |
case-09 | pass→pass | 7,944 | 11,611 | +46% | 1 | 1 | 0% | 1,230 | 1,781 | +45% | 0 | 0 | — |
case-10 | pass→pass | 11,716 | 22,527 | +92% | 1 | 1 | 0% | 1,418 | 2,038 | +44% | 0 | 0 | — |
case-12 | fail→pass | 11,657 | 18,982 | +63% | 1 | 1 | 0% | 1,834 | 2,620 | +43% | 0 | 0 | — |
case-13 | pass→pass | 15,282 | 12,886 | -16% | 1 | 1 | 0% | 1,942 | 2,328 | +20% | 0 | 0 | — |
case-14 | pass→pass | 12,383 | 14,268 | +15% | 1 | 1 | 0% | 1,901 | 2,182 | +15% | 0 | 0 | — |
case-15 | pass→pass | 8,311 | 9,352 | +13% | 1 | 1 | 0% | 1,119 | 1,797 | +61% | 0 | 0 | — |
case-16 | pass→pass | 12,882 | 11,052 | -14% | 1 | 1 | 0% | 2,109 | 1,939 | -8% | 0 | 0 | — |
case-17 | pass→pass | 8,302 | 9,992 | +20% | 1 | 1 | 0% | 1,303 | 1,550 | +19% | 0 | 0 | — |
case-18 | pass→pass | 26,478 | 9,842 | -63% | 1 | 1 | 0% | 1,122 | 1,636 | +46% | 0 | 0 | — |
case-19 | pass→pass | 22,268 | 19,286 | -13% | 1 | 1 | 0% | 2,895 | 3,197 | +10% | 0 | 0 | — |
case-20 | fail→fail | 13,471 | 18,221 | +35% | 1 | 1 | 0% | 2,088 | 2,817 | +35% | 0 | 0 | — |
case-21 | fail→pass | 9,750 | 8,137 | -17% | 1 | 1 | 0% | 1,499 | 1,514 | +1% | 0 | 0 | — |
case-22 | pass→pass | 10,933 | 16,556 | +51% | 1 | 1 | 0% | 1,447 | 2,514 | +74% | 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.
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.