Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Handle Taiwan e-invoice carriers (載具) — 手機條碼, 自然人憑證, 會員載具, plus 捐贈碼 donation flow. Covers carrier validation, scan-to-store UX, member-carrier consolidation, and prize-draw winner notification. Use when designing carrier scanning UX, debugging invalid-carrier rejections, or implementing donation-code flow. STATUS: SKELETON — body pending.
.claude/skills/asgard-ai-platform-tw-ecom-invoice-carrier/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | -3% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 7% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 0% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -8% | 0% |
| case-14 | ✗→✓ | ▲ Improved | 42% | 0% |
> STATUS: SKELETON — body pending.
tw-einvoice-guidetw-ecom-invoice-ezpay / -universalecTODO: carrier types, format specs, validation rules. NOTE (填充時): 格式規格(手機條碼 / + 7 alphanumeric、自然人憑證 2 letters + 14 digits) 已在 tw-ecom-invoice-ezpay Gotchas 中詳細說明。此 skill 不重複格式規格; 改聚焦在 UX 層面:validation 在哪層做、錯誤訊息設計、掃碼元件選型。
TODO: which carrier to default to given context (e.g., 有會員帳號 → 會員載具;無帳號 → 引導填手機條碼;B2B → 不需載具).
TODO: scan widget 選型與整合、client-side validation 流程、member-carrier linking(會員帳號與手機條碼綁定)。 不重複 issue_invoice carrier_type / carrier_num 參數說明 — 那些在 ezpay/universalec skill。
TODO: 5-6 pitfalls.
TODO.
TODO.
tw-einvoice-guide_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-11 | pass→pass | 19,785 | 18,646 | -6% | 1 | 1 | 0% | 2,699 | 3,241 | +20% | 0 | 0 | — |
case-04 | fail→fail | 20,339 | 15,575 | -23% | 1 | 1 | 0% | 3,520 | 3,169 | -10% | 0 | 0 | — |
case-01 | fail→pass | 51,750 | 27,698 | -46% | 1 | 1 | 0% | 4,803 | 4,654 | -3% | 0 | 0 | — |
case-02 | pass→pass | 19,261 | 21,193 | +10% | 1 | 1 | 0% | 2,949 | 3,198 | +8% | 0 | 0 | — |
case-03 | pass→pass | 17,263 | 17,204 | -0% | 1 | 1 | 0% | 2,741 | 3,264 | +19% | 0 | 0 | — |
case-05 | fail→pass | 16,717 | 16,252 | -3% | 1 | 1 | 0% | 2,810 | 2,993 | +7% | 0 | 0 | — |
case-06 | pass→pass | 13,865 | 10,171 | -27% | 1 | 1 | 0% | 2,605 | 2,223 | -15% | 0 | 0 | — |
case-07 | fail→pass | 16,822 | 13,932 | -17% | 1 | 1 | 0% | 2,595 | 2,588 | -0% | 0 | 0 | — |
case-08 | fail→pass | 17,015 | 14,457 | -15% | 1 | 1 | 0% | 2,973 | 2,741 | -8% | 0 | 0 | — |
case-09 | pass→pass | 12,720 | 21,727 | +71% | 1 | 1 | 0% | 2,058 | 4,643 | +126% | 0 | 0 | — |
case-10 | pass→pass | 17,597 | 16,234 | -8% | 1 | 1 | 0% | 2,583 | 3,321 | +29% | 0 | 0 | — |
case-12 | pass→pass | 21,442 | 19,527 | -9% | 1 | 1 | 0% | 3,036 | 3,595 | +18% | 0 | 0 | — |
case-13 | fail→fail | 18,095 | 19,572 | +8% | 1 | 1 | 0% | 3,144 | 3,691 | +17% | 0 | 0 | — |
case-14 | fail→pass | 15,928 | 18,456 | +16% | 1 | 1 | 0% | 2,362 | 3,355 | +42% | 0 | 0 | — |
case-15 | fail→pass | 15,605 | 17,734 | +14% | 1 | 1 | 0% | 2,446 | 3,702 | +51% | 0 | 0 | — |
case-16 | fail→fail | 17,027 | 16,914 | -1% | 1 | 1 | 0% | 2,349 | 3,200 | +36% | 0 | 0 | — |
case-17 | pass→pass | 16,578 | 21,382 | +29% | 1 | 1 | 0% | 2,517 | 3,463 | +38% | 0 | 0 | — |
case-18 | fail→pass | 14,117 | 12,655 | -10% | 1 | 1 | 0% | 2,217 | 2,446 | +10% | 0 | 0 | — |
case-19 | fail→pass | 13,848 | 15,352 | +11% | 1 | 1 | 0% | 2,004 | 2,565 | +28% | 0 | 0 | — |
case-20 | pass→pass | 22,431 | 21,727 | -3% | 1 | 1 | 0% | 3,387 | 3,902 | +15% | 0 | 0 | — |
case-21 | pass→pass | 18,345 | 30,116 | +64% | 1 | 1 | 0% | 2,898 | 5,169 | +78% | 0 | 0 | — |
case-22 | pass→pass | 26,939 | 22,914 | -15% | 1 | 1 | 0% | 4,236 | 3,720 | -12% | 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 +36 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.