Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Run Taiwan e-commerce promotional campaigns — 雙11, 618, 年中慶, 雙12, 週年慶, 母親節. Covers promo calendar, stacking rules across platforms, 檔期 pricing strategy, inventory allocation, and post-campaign 退貨 handling. Use when planning a specific campaign or building a year-round 檔期 schedule. Do NOT use for ad ROI measurement (use `ecom-promo-roi`). STATUS: SKELETON — body pending.
.claude/skills/asgard-ai-platform-tw-ecom-operations-promotion/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-08 | ✗→✓ | ▲ Improved | 24% | 0% |
| case-11 | ✗→✓ | ▲ Improved | 30% | 0% |
| case-06 | ✓→✗ | ▼ Worse | 50% | 0% |
| case-13 | ✓→✓ | ＝ Same ✓ | 39% | 0% |
| case-04 | ✓→✓ | ＝ Same ✓ | 13% | 0% |
> STATUS: SKELETON — body pending.
ecom-promo-roitw-ecom-operations-pricingTODO: TW promo calendar, stacking hierarchy, DTC vs marketplace behavior during peaks.
TODO: which 檔期 to participate in, channel mix, discount depth.
TODO: 檔期 plan template, inventory buffer, support staffing.
TODO: 5-6 pitfalls (stacking over-discount, platform fee waiver traps, inventory over-commit, return spike 2 weeks after peak, marketplace cancelation penalties).
TODO.
TODO.
ecom-promo-roitw-ecom-operations-pricing_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 | fail→fail | 22,391 | 55,906 | +150% | 1 | 1 | 0% | 3,283 | 4,027 | +23% | 0 | 0 | — |
case-02 | fail→fail | 50,357 | 25,130 | -50% | 1 | 1 | 0% | 3,200 | 4,411 | +38% | 0 | 0 | — |
case-13 | pass→pass | 17,740 | 18,794 | +6% | 1 | 1 | 0% | 2,331 | 3,229 | +39% | 0 | 0 | — |
case-03 | fail→fail | 31,435 | 44,938 | +43% | 1 | 1 | 0% | 4,558 | 7,003 | +54% | 0 | 0 | — |
case-04 | pass→pass | 25,169 | 23,576 | -6% | 1 | 1 | 0% | 3,301 | 3,734 | +13% | 0 | 0 | — |
case-05 | pass→pass | 12,881 | 16,949 | +32% | 1 | 1 | 0% | 1,783 | 2,730 | +53% | 0 | 0 | — |
case-06 | pass→fail | 23,476 | 31,059 | +32% | 1 | 1 | 0% | 2,890 | 4,329 | +50% | 0 | 0 | — |
case-07 | pass→pass | 24,792 | 41,642 | +68% | 1 | 1 | 0% | 3,416 | 5,798 | +70% | 0 | 0 | — |
case-08 | fail→pass | 19,419 | 17,705 | -9% | 1 | 1 | 0% | 2,516 | 3,118 | +24% | 0 | 0 | — |
case-09 | fail→fail | 16,836 | 30,407 | +81% | 1 | 1 | 0% | 2,628 | 3,706 | +41% | 0 | 0 | — |
case-10 | pass→pass | 17,165 | 23,886 | +39% | 1 | 1 | 0% | 2,671 | 3,890 | +46% | 0 | 0 | — |
case-11 | fail→pass | 17,593 | 24,204 | +38% | 1 | 1 | 0% | 2,758 | 3,593 | +30% | 0 | 0 | — |
case-12 | fail→fail | 27,873 | 37,092 | +33% | 1 | 1 | 0% | 3,722 | 4,817 | +29% | 0 | 0 | — |
case-14 | pass→pass | 18,507 | 22,631 | +22% | 1 | 1 | 0% | 2,443 | 3,635 | +49% | 0 | 0 | — |
case-15 | pass→pass | 12,540 | 14,565 | +16% | 1 | 1 | 0% | 1,919 | 2,420 | +26% | 0 | 0 | — |
case-16 | pass→pass | 10,650 | 15,031 | +41% | 1 | 1 | 0% | 1,966 | 2,394 | +22% | 0 | 0 | — |
case-17 | pass→pass | 29,094 | 30,048 | +3% | 1 | 1 | 0% | 4,112 | 5,232 | +27% | 0 | 0 | — |
case-18 | fail→fail | 21,132 | 31,596 | +50% | 1 | 1 | 0% | 3,288 | 4,857 | +48% | 0 | 0 | — |
case-19 | pass→pass | 16,806 | 20,178 | +20% | 1 | 1 | 0% | 2,609 | 2,843 | +9% | 0 | 0 | — |
case-20 | pass→pass | 15,469 | 13,272 | -14% | 1 | 1 | 0% | 2,182 | 2,528 | +16% | 0 | 0 | — |
case-21 | pass→pass | 16,705 | 23,207 | +39% | 1 | 1 | 0% | 2,630 | 4,039 | +54% | 0 | 0 | — |
case-22 | pass→pass | 23,683 | 25,971 | +10% | 1 | 1 | 0% | 3,169 | 3,744 | +18% | 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. 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.