Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Void Taiwan e-invoices within the same bimonthly window, or issue allowances (折讓) for cross-period corrections. Use when handling returns / refunds that affect invoice state. Do NOT use for initial issuance — see the 加值中心-specific skills. STATUS: SKELETON — body pending.
.claude/skills/asgard-ai-platform-tw-ecom-invoice-void/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-05 | ✗→✓ | ▲ Improved | 56% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 50% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 36% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -26% | 0% |
| case-11 | ✗→✓ | ▲ Improved | 35% | 0% |
> STATUS: SKELETON — body pending.
tw-tax-basicsTODO: void vs 折讓, bimonthly boundary, accounting implications. NOTE (填充時): 此 skill 只寫「選哪條路」的決策邏輯,不重複 API 呼叫細節。 tw-ecom-invoice-ezpay 已有完整的 void_invoice / issue_allowance / trigger_allowance 步驟; tw-ecom-invoice-universalec 同理。此 skill 的 Implementation guidance 應只提供判斷框架, 並用 "→ see tw-ecom-invoice-ezpay / -universalec" 指向 API 層。
TODO: given refund date vs issuance date → void or 折讓.
TODO: 決策路徑後的 SOP(accounting entries, customer notification, 字軌影響)。 不重複 void_invoice / issue_allowance / trigger_allowance 的 API 參數 — 那些在 ezpay/universalec skill。
TODO: 5-6 pitfalls (boundary off-by-one, 字軌 reservation, lottery invalidation, allowance numbering).
TODO.
TODO.
tw-ecom-invoice-ezpaytw-ecom-payment-dispute_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 | 23,942 | 33,767 | +41% | 1 | 1 | 0% | 3,780 | 5,130 | +36% | 0 | 0 | — |
case-19 | pass→pass | 10,214 | 13,247 | +30% | 1 | 1 | 0% | 1,382 | 2,243 | +62% | 0 | 0 | — |
case-02 | pass→pass | 15,698 | 15,464 | -1% | 1 | 1 | 0% | 2,675 | 3,030 | +13% | 0 | 0 | — |
case-03 | pass→pass | 9,398 | 8,922 | -5% | 1 | 1 | 0% | 1,493 | 2,012 | +35% | 0 | 0 | — |
case-04 | pass→pass | 13,227 | 16,312 | +23% | 1 | 1 | 0% | 2,278 | 2,966 | +30% | 0 | 0 | — |
case-05 | fail→pass | 9,042 | 12,614 | +40% | 1 | 1 | 0% | 1,485 | 2,324 | +56% | 0 | 0 | — |
case-06 | pass→pass | 12,722 | 15,601 | +23% | 1 | 1 | 0% | 2,143 | 2,605 | +22% | 0 | 0 | — |
case-07 | fail→pass | 11,172 | 12,729 | +14% | 1 | 1 | 0% | 1,771 | 2,655 | +50% | 0 | 0 | — |
case-08 | fail→pass | 9,234 | 8,223 | -11% | 1 | 1 | 0% | 1,384 | 1,878 | +36% | 0 | 0 | — |
case-09 | fail→fail | 22,264 | 38,005 | +71% | 1 | 1 | 0% | 3,989 | 6,378 | +60% | 0 | 0 | — |
case-10 | fail→pass | 18,763 | 8,622 | -54% | 1 | 1 | 0% | 2,700 | 1,985 | -26% | 0 | 0 | — |
case-11 | fail→pass | 13,847 | 12,297 | -11% | 1 | 1 | 0% | 1,934 | 2,602 | +35% | 0 | 0 | — |
case-12 | pass→pass | 12,697 | 16,364 | +29% | 1 | 1 | 0% | 2,042 | 3,217 | +58% | 0 | 0 | — |
case-13 | pass→pass | 11,511 | 12,051 | +5% | 1 | 1 | 0% | 2,012 | 2,310 | +15% | 0 | 0 | — |
case-14 | pass→pass | 13,840 | 18,935 | +37% | 1 | 1 | 0% | 2,255 | 3,245 | +44% | 0 | 0 | — |
case-15 | pass→pass | 14,224 | 16,043 | +13% | 1 | 1 | 0% | 2,095 | 2,639 | +26% | 0 | 0 | — |
case-16 | fail→pass | 9,519 | 9,945 | +4% | 1 | 1 | 0% | 1,393 | 1,857 | +33% | 0 | 0 | — |
case-17 | pass→pass | 10,900 | 9,302 | -15% | 1 | 1 | 0% | 1,725 | 1,795 | +4% | 0 | 0 | — |
case-18 | fail→pass | 8,209 | 10,568 | +29% | 1 | 1 | 0% | 1,344 | 2,296 | +71% | 0 | 0 | — |
case-20 | pass→pass | 17,908 | 23,937 | +34% | 1 | 1 | 0% | 3,221 | 4,058 | +26% | 0 | 0 | — |
case-21 | pass→pass | 13,615 | 21,446 | +58% | 1 | 1 | 0% | 2,345 | 3,686 | +57% | 0 | 0 | — |
case-22 | pass→pass | 16,148 | 23,446 | +45% | 1 | 1 | 0% | 2,634 | 3,461 | +31% | 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 +32 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.