Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Implement Taiwan's e-invoice (電子發票) system including platform integration, B2B vs B2C formats, carrier consolidation, and tax filing reconciliation. Use this skill when the user needs to set up e-invoicing for a Taiwan business, integrate with the MOF platform, understand carrier codes, or troubleshoot invoice issues — even if they say 'set up e-invoice', 'how does 電子發票 work', 'integrate with 財政部', or 'carrier barcode scanning'.
.claude/skills/asgard-ai-platform-tw-einvoice-guide/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | -35% | 0% |
| case-02 | ✗→✓ | ▲ Improved | -33% | 0% |
| case-03 | ✗→✓ | ▲ Improved | -34% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 7% | 0% |
| case-06 | ✓→✓ | = Same ✓ | 69% | 0% |
IRON LAW: E-Invoice Is Mandatory for Most B2C Businesses
Since 2019, businesses using 統一發票 must issue electronic invoices
through the 財政部電子發票整合服務平台. Paper invoices are being phased
out. Non-compliance triggers penalties and may affect 營業稅 filing.Your System (POS/ERP/E-commerce)
↓ API / Turnkey
財政部電子發票整合服務平台 (einvoice.nat.gov.tw)
↓
Consumer (via 載具: 手機條碼 / 自然人憑證 / App)
↓
國稅局 (tax reconciliation)| Method | How It Works | Best For | Complexity | |--------|-------------|----------|-----------| | Turnkey | Install MOF-provided software, batch upload invoices | Traditional businesses, low volume | Medium | | API (加值服務中心) | Connect via 加值中心 API (e.g., 綠界、藍新) | E-commerce, SaaS, high volume | Low-Medium | | Direct API | Connect directly to MOF platform | Large enterprises with IT team | High | | POS integration | POS vendor handles e-invoice natively | Retail, F&B | Low (vendor does it) |
| Aspect | B2B (營業人對營業人) | B2C (營業人對消費者) | |--------|-------------------|-------------------| | Buyer info | Buyer's 統編 required | No 統編 (consumer) | | Format | 三聯式 | 二聯式 | | Tax display | Tax amount shown separately | Tax included in price | | Carrier | N/A | 手機條碼, 自然人憑證, or membership carrier | | Prize eligibility | No | Yes (中獎機制) |
| Carrier | Code Format | Use Case | |---------|-----------|----------| | 手機條碼 | /XXXXXXX (slash + 7 chars) | Most common consumer carrier | | 自然人憑證 | 2 letters + 14 digits | Government ID-linked | | 會員載具 (membership) | Defined by business | Loyalty program integration | | 捐贈碼 | 3-7 digits | Donate invoice to charity |
Phase 1: Registration
Phase 2: Development/Integration
Phase 3: Testing
Phase 4: Go-Live
Issue Invoice:
POST /invoice → { seller_id, buyer_id (optional), items[], amount, tax, carrier_type, carrier_id }
→ Response: { invoice_number, invoice_date, random_code }Void Invoice:
POST /invoice/void → { invoice_number, invoice_date, void_reason }Query Invoice:
GET /invoice/{number} → { status, items, amount, carrier }markdown# E-Invoice Implementation Plan: {Business} ## Current State - Business type: B2B / B2C / Both - Current invoicing: Paper / Partial e-invoice / None - Transaction volume: {N}/month ## Integration Method - Method: {Turnkey / API / POS integration} - Provider: {加值中心 name, if applicable} - Rationale: {why this method} ## Implementation Checklist - [ ] MOF platform registration - [ ] AppID obtained - [ ] Integration developed - [ ] Sandbox testing passed - [ ] Carrier scanning implemented - [ ] Donation code support - [ ] Production go-live - [ ] Reconciliation process documented ## Timeline | Phase | Duration | Milestone | |-------|----------|-----------| | Registration | 1-2 weeks | AppID obtained | | Development | 2-4 weeks | Integration complete | | Testing | 1-2 weeks | Sandbox verified | | Go-live | 1 day | First production invoice |
references/einvoice-api.md| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 66,287 | 44,972 | -32% | 1 | 1 | 0% | 6,111 | 3,970 | -35% | 0 | 0 | — |
case-02 | fail→pass | 84,573 | 55,594 | -34% | 1 | 1 | 0% | 7,406 | 4,967 | -33% | 0 | 0 | — |
case-03 | fail→pass | 63,195 | 21,788 | -66% | 1 | 1 | 0% | 6,693 | 4,443 | -34% | 0 | 0 | — |
case-04 | fail→pass | 21,076 | 11,441 | -46% | 1 | 1 | 0% | 2,758 | 2,958 | +7% | 0 | 0 | — |
case-05 | fail→fail | 13,267 | 12,447 | -6% | 1 | 1 | 0% | 2,102 | 2,938 | +40% | 0 | 0 | — |
case-06 | pass→pass | 14,559 | 14,464 | -1% | 1 | 1 | 0% | 2,055 | 3,482 | +69% | 0 | 0 | — |
case-07 | pass→pass | 16,793 | 13,378 | -20% | 1 | 1 | 0% | 2,366 | 3,635 | +54% | 0 | 0 | — |
case-08 | pass→pass | 8,315 | 5,297 | -36% | 1 | 1 | 0% | 1,212 | 2,326 | +92% | 0 | 0 | — |
case-09 | pass→pass | 10,378 | 7,340 | -29% | 1 | 1 | 0% | 1,470 | 2,386 | +62% | 0 | 0 | — |
case-10 | pass→pass | 8,676 | 4,063 | -53% | 1 | 1 | 0% | 1,117 | 2,077 | +86% | 0 | 0 | — |
case-11 | pass→pass | 17,344 | 21,265 | +23% | 1 | 1 | 0% | 2,967 | 4,279 | +44% | 0 | 0 | — |
case-12 | pass→pass | 15,922 | 14,363 | -10% | 1 | 1 | 0% | 2,625 | 3,820 | +46% | 0 | 0 | — |
case-13 | fail→fail | 35,244 | 15,911 | -55% | 1 | 1 | 0% | 2,702 | 3,839 | +42% | 0 | 0 | — |
case-14 | pass→pass | 23,316 | 17,202 | -26% | 1 | 1 | 0% | 3,173 | 4,508 | +42% | 0 | 0 | — |
case-15 | pass→pass | 15,839 | 7,408 | -53% | 1 | 1 | 0% | 2,566 | 2,662 | +4% | 0 | 0 | — |
case-16 | pass→pass | 18,547 | 16,873 | -9% | 1 | 1 | 0% | 2,519 | 4,044 | +61% | 0 | 0 | — |
case-17 | pass→pass | 12,073 | 10,430 | -14% | 1 | 1 | 0% | 1,923 | 3,013 | +57% | 0 | 0 | — |
case-18 | pass→pass | 14,387 | 8,263 | -43% | 1 | 1 | 0% | 1,967 | 2,750 | +40% | 0 | 0 | — |
case-19 | pass→pass | 10,458 | 7,895 | -25% | 1 | 1 | 0% | 1,735 | 2,722 | +57% | 0 | 0 | — |
case-20 | pass→pass | 15,154 | 17,277 | +14% | 1 | 1 | 0% | 2,770 | 4,599 | +66% | 0 | 0 | — |
case-21 | pass→pass | 12,591 | 10,649 | -15% | 1 | 1 | 0% | 1,838 | 3,352 | +82% | 0 | 0 | — |
case-22 | pass→pass | 12,140 | 11,365 | -6% | 1 | 1 | 0% | 1,987 | 3,264 | +64% | 0 | 0 | — |
case-23 | pass→pass | 13,771 | 13,501 | -2% | 1 | 1 | 0% | 2,688 | 3,849 | +43% | 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. 23 cases were attempted. The headline lift of +17 percentage points is the difference between those two pass rates over the 23 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.