Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Taiwan e-commerce customer service — LINE / Messenger / email / phone channel mix, 消保法鑑賞期 scripts, PTT/Dcard reputation monitoring, 負評 response, and SLA targets for the TW market. Use when designing CS SOP, training CS agents, or responding to viral 負評. Do NOT use for generic chatbot design (use `cs-chatbot-design`). STATUS: SKELETON — body pending.
.claude/skills/asgard-ai-platform-tw-ecom-operations-customer-service/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-05 | ✗→✓ | ▲ Improved | 58% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 14% | 0% |
| case-13 | ✗→✓ | ▲ Improved | 58% | 0% |
| case-16 | ✗→✓ | ▲ Improved | 19% | 0% |
| case-18 | ✓→✗ | ▼ Worse | 36% | 0% |
> STATUS: SKELETON — body pending.
cs-chatbot-designpr-crisis-responseTODO: channel mix, TW consumer complaint culture, PTT / Dcard etiquette.
TODO: complaint → channel → script template.
TODO: SOP template, escalation ladder, monitoring setup.
TODO: 5-6 pitfalls (鑑賞期 over-promise, PTT Streisand effect, LINE 1-to-1 burnout, negative-review response template pitfalls, 消保官 tone).
TODO.
TODO.
cs-soptw-ecom-compliance-consumerpr-crisis-response_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 | 68,978 | 55,493 | -20% | 1 | 1 | 0% | 7,286 | 4,676 | -36% | 0 | 0 | — |
case-02 | pass→pass | 47,001 | 89,940 | +91% | 1 | 1 | 0% | 2,216 | 2,293 | +3% | 0 | 0 | — |
case-03 | fail→fail | 50,759 | 23,842 | -53% | 1 | 1 | 0% | 2,779 | 3,445 | +24% | 0 | 0 | — |
case-04 | pass→pass | 16,137 | 20,377 | +26% | 1 | 1 | 0% | 2,196 | 2,961 | +35% | 0 | 0 | — |
case-05 | fail→pass | 18,558 | 59,089 | +218% | 1 | 1 | 0% | 2,354 | 3,713 | +58% | 0 | 0 | — |
case-06 | fail→pass | 16,609 | 21,723 | +31% | 1 | 1 | 0% | 2,497 | 2,840 | +14% | 0 | 0 | — |
case-07 | pass→pass | 18,703 | 23,275 | +24% | 1 | 1 | 0% | 2,764 | 3,227 | +17% | 0 | 0 | — |
case-08 | pass→pass | 17,057 | 25,499 | +49% | 1 | 1 | 0% | 2,683 | 3,820 | +42% | 0 | 0 | — |
case-09 | pass→pass | 15,807 | 29,424 | +86% | 1 | 1 | 0% | 2,413 | 4,158 | +72% | 0 | 0 | — |
case-10 | pass→pass | 24,488 | 33,305 | +36% | 1 | 1 | 0% | 3,262 | 4,733 | +45% | 0 | 0 | — |
case-11 | pass→pass | 13,095 | 16,394 | +25% | 1 | 1 | 0% | 2,190 | 2,980 | +36% | 0 | 0 | — |
case-12 | pass→pass | 19,704 | 20,718 | +5% | 1 | 1 | 0% | 2,837 | 3,208 | +13% | 0 | 0 | — |
case-13 | fail→pass | 17,659 | 31,635 | +79% | 1 | 1 | 0% | 2,731 | 4,316 | +58% | 0 | 0 | — |
case-14 | pass→pass | 19,137 | 22,484 | +17% | 1 | 1 | 0% | 3,285 | 3,928 | +20% | 0 | 0 | — |
case-15 | pass→pass | 13,641 | 18,318 | +34% | 1 | 1 | 0% | 2,112 | 2,925 | +38% | 0 | 0 | — |
case-16 | fail→pass | 18,243 | 20,640 | +13% | 1 | 1 | 0% | 2,667 | 3,180 | +19% | 0 | 0 | — |
case-17 | pass→pass | 15,100 | 21,264 | +41% | 1 | 1 | 0% | 2,227 | 2,959 | +33% | 0 | 0 | — |
case-18 | pass→fail | 21,949 | 24,956 | +14% | 1 | 1 | 0% | 2,989 | 4,074 | +36% | 0 | 0 | — |
case-19 | pass→pass | 19,826 | 29,383 | +48% | 1 | 1 | 0% | 3,079 | 4,630 | +50% | 0 | 0 | — |
case-20 | pass→pass | 15,812 | 21,349 | +35% | 1 | 1 | 0% | 2,248 | 3,125 | +39% | 0 | 0 | — |
case-21 | pass→pass | 27,686 | 30,965 | +12% | 1 | 1 | 0% | 3,679 | 4,937 | +34% | 0 | 0 | — |
case-22 | pass→pass | 19,639 | 39,349 | +100% | 1 | 1 | 0% | 3,181 | 4,829 | +52% | 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 +14 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.