Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Operate a Shopee Taiwan store — listings, promotions, flash sales, SIP (Shopee Supported Program) cross-border, ads (蝦皮廣告), and reputation/review management. Use when setting up or running Shopee TW operations, participating in platform campaigns (雙11, 618), or managing seller-center workflows. Do NOT use for Shopee API integration specifics (no official Asgard MCP yet) or DTC platforms. STATUS: SKELETON — body pending.
.claude/skills/asgard-ai-platform-tw-ecom-marketplace-shopee/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-06 | ✗→✓ | ▲ Improved | 27% | 0% |
| case-15 | ✗→✓ | ▲ Improved | 37% | 0% |
| case-16 | ✗→✓ | ▲ Improved | 53% | 0% |
| case-14 | ✓→✗ | ▼ Worse | 34% | 0% |
| case-19 | ✓→✗ | ▼ Worse | 45% | 0% |
> STATUS: SKELETON — body pending.
algo-ad-bidding, algo-ad-gsp, algo-ad-ctrTODO: Shopee TW vs Shopee SEA differences, seller tiers and fee tiers, 蝦皮購物 vs 蝦皮商城.
TODO: when Shopee is the right channel, when to tier up.
TODO: listing optimization, campaign participation, ad setup, dispute handling.
TODO: 5-6 pitfalls (fee surprises on 免運券, campaign eligibility rules, search algo behavior, review-manipulation penalties).
TODO.
TODO.
tw-ecom-channel-strategytw-ecom-marketplace-momotw-ecom-operations-promotion_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 | 27,436 | 70,533 | +157% | 1 | 1 | 0% | 4,517 | 5,623 | +24% | 0 | 0 | — |
case-02 | pass→pass | 46,222 | 28,601 | -38% | 1 | 1 | 0% | 3,666 | 3,662 | -0% | 0 | 0 | — |
case-03 | pass→pass | 32,225 | 23,281 | -28% | 1 | 1 | 0% | 5,206 | 4,206 | -19% | 0 | 0 | — |
case-04 | fail→fail | 46,259 | 53,303 | +15% | 1 | 1 | 0% | 2,333 | 3,408 | +46% | 0 | 0 | — |
case-05 | pass→pass | 15,286 | 17,784 | +16% | 1 | 1 | 0% | 2,510 | 2,827 | +13% | 0 | 0 | — |
case-06 | fail→pass | 16,153 | 23,334 | +44% | 1 | 1 | 0% | 2,884 | 3,669 | +27% | 0 | 0 | — |
case-07 | fail→fail | 5,505 | 8,326 | +51% | 1 | 1 | 0% | 979 | 1,723 | +76% | 0 | 0 | — |
case-08 | pass→pass | 13,916 | 10,255 | -26% | 1 | 1 | 0% | 1,946 | 2,016 | +4% | 0 | 0 | — |
case-09 | pass→pass | 7,049 | 5,546 | -21% | 1 | 1 | 0% | 927 | 1,299 | +40% | 0 | 0 | — |
case-10 | pass→pass | 7,346 | 7,108 | -3% | 1 | 1 | 0% | 1,102 | 1,274 | +16% | 0 | 0 | — |
case-11 | pass→pass | 12,935 | 20,599 | +59% | 1 | 1 | 0% | 2,053 | 3,055 | +49% | 0 | 0 | — |
case-12 | pass→pass | 14,544 | 14,968 | +3% | 1 | 1 | 0% | 2,127 | 2,892 | +36% | 0 | 0 | — |
case-13 | pass→pass | 10,205 | 13,292 | +30% | 1 | 1 | 0% | 1,684 | 2,304 | +37% | 0 | 0 | — |
case-14 | pass→fail | 8,653 | 10,090 | +17% | 1 | 1 | 0% | 1,292 | 1,736 | +34% | 0 | 0 | — |
case-15 | fail→pass | 14,749 | 17,476 | +18% | 1 | 1 | 0% | 2,334 | 3,208 | +37% | 0 | 0 | — |
case-16 | fail→pass | 11,843 | 16,716 | +41% | 1 | 1 | 0% | 1,984 | 3,027 | +53% | 0 | 0 | — |
case-17 | pass→pass | 17,077 | 18,928 | +11% | 1 | 1 | 0% | 2,648 | 3,055 | +15% | 0 | 0 | — |
case-18 | pass→pass | 5,106 | 8,507 | +67% | 1 | 1 | 0% | 898 | 1,517 | +69% | 0 | 0 | — |
case-19 | pass→fail | 14,586 | 15,211 | +4% | 1 | 1 | 0% | 1,936 | 2,802 | +45% | 0 | 0 | — |
case-20 | pass→pass | 13,405 | 17,508 | +31% | 1 | 1 | 0% | 2,180 | 2,763 | +27% | 0 | 0 | — |
case-21 | pass→pass | 5,380 | 5,453 | +1% | 1 | 1 | 0% | 868 | 1,246 | +44% | 0 | 0 | — |
case-22 | pass→pass | 9,931 | 9,887 | -0% | 1 | 1 | 0% | 1,599 | 1,924 | +20% | 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 +5 percentage points is the difference between those two pass rates over the 22 comparable cases. 2 cases got worse with the skill loaded, and they are 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.