Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Navigate Taiwan healthcare regulations including NHI system, medical device classification, drug registration, telemedicine rules, and health data protection. Use this skill when the user is building a health tech product for Taiwan, needs to understand NHI, evaluate medical device regulatory pathways, or assess telemedicine compliance — even if they say 'sell a medical device in Taiwan', 'how does NHI work', 'telemedicine regulations', or 'health data privacy in Taiwan'.
.claude/skills/asgard-ai-platform-tw-healthcare-regulations/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-03 | ✗→✓ | ▲ Improved | -20% | 0% |
| case-12 | ✗→✓ | ▲ Improved | 34% | 0% |
| case-01 | ✓→✓ | = Same ✓ | -18% | 0% |
| case-04 | ✓→✓ | = Same ✓ | 31% | 0% |
| case-06 | ✓→✓ | = Same ✓ | 85% | 0% |
IRON LAW: NHI Shapes Everything in Taiwan Healthcare
Taiwan's National Health Insurance (全民健保) covers 99.9% of the
population. Any healthcare product or service strategy in Taiwan must
account for NHI — either by getting NHI reimbursement (volume play)
or by positioning as self-pay/premium (margin play).
Ignoring NHI is like ignoring gravity.| Aspect | Detail | |--------|--------| | Coverage | 99.9% of population (23M+ people) | | Single payer | 衛生福利部中央健康保險署 (NHIA) | | Premium | 5.17% of insured salary (shared: employer 60%, employee 30%, government 10%) | | Co-pay | Outpatient: NT$50-420. Hospitalization: 5-30% (capped) | | Drug pricing | NHIA sets reimbursement prices via Drug Expenditure Target (DET) | | Annual budget | ~NT$800B+ (growing 4-5% annually) |
| Class | Risk | Examples | Approval Path | Timeline | |-------|------|---------|-------------|---------| | Class I | Low | Bandages, tongue depressors | Registration (listing) | 1-2 months | | Class II | Medium | Blood pressure monitors, surgical gloves | Technical review | 6-12 months | | Class III | High | Implants, AI diagnostic software | Full clinical review | 12-24 months | | SaMD (Software as Medical Device) | Varies by intended use | AI diagnosis, clinical decision support | Class II or III depending on risk | 6-24 months |
Regulatory body: 衛生福利部食品藥物管理署 (TFDA)
| Category | Regulation Status | Key Rule | |----------|------------------|---------| | Telemedicine | Expanded post-COVID (通訊診察治療辦法) | Allowed for follow-up visits, chronic disease, remote areas. Initial visits still require in-person for most cases. | | AI diagnostics | SaMD regulation applies | If AI makes/assists clinical decisions, it's a medical device requiring TFDA approval | | Health apps | Unregulated if wellness-only | Crosses into medical device territory if it diagnoses, treats, or monitors a medical condition | | Health data | 個人資料保護法 (PDPA) + 醫療法 | Medical records have stricter protection than general personal data. Patient consent required for data use. | | Electronic medical records | 醫療機構電子病歷製作及管理辦法 | EMR systems must meet MOHW standards. Cloud storage allowed with conditions. |
Does your product diagnose, treat, or monitor a medical condition?
├── NO → Not a medical device. General consumer regulations apply.
└── YES → Medical device (SaMD)
├── Does it provide clinical decision support?
│ ├── Autonomous (AI decides) → Class III
│ └── Assistive (human decides) → Class II
└── Does it monitor vital signs?
├── Clinical grade → Class II-III
└── Wellness/fitness → Likely not regulated (but verify with TFDA)markdown# Healthcare Regulatory Assessment: {Product} ## Product Classification - Type: Medical device / Wellness / SaMD / Telemedicine - Risk class: I / II / III - Regulatory body: TFDA / NHIA / None ## Regulatory Pathway | Step | Action | Timeline | Cost | |------|--------|----------|------| | 1 | {regulatory step} | {months} | NT${X} | ## NHI Strategy - NHI reimbursement: Pursuing / Not pursuing - If pursuing: {reimbursement category, pricing strategy} - If not: {self-pay positioning, target market} ## Compliance Checklist - [ ] TFDA classification confirmed - [ ] Clinical data requirements identified - [ ] Data privacy (PDPA + medical records) compliant - [ ] NHI reimbursement strategy decided
references/tfda-submission.mdreferences/nhi-reimbursement.md| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | pass→pass | 34,208 | 47,108 | +38% | 1 | 1 | 0% | 4,860 | 3,961 | -18% | 0 | 0 | — |
case-02 | fail→fail | 43,425 | 46,476 | +7% | 1 | 1 | 0% | 7,156 | 4,642 | -35% | 0 | 0 | — |
case-03 | fail→pass | 36,995 | 17,178 | -54% | 1 | 1 | 0% | 5,033 | 4,023 | -20% | 0 | 0 | — |
case-04 | pass→pass | 17,682 | 12,698 | -28% | 1 | 1 | 0% | 2,599 | 3,400 | +31% | 0 | 0 | — |
case-05 | fail→fail | 20,334 | 19,511 | -4% | 1 | 1 | 0% | 2,935 | 3,990 | +36% | 0 | 0 | — |
case-06 | pass→pass | 46,622 | 22,890 | -51% | 1 | 1 | 0% | 2,303 | 4,262 | +85% | 0 | 0 | — |
case-07 | pass→pass | 13,879 | 9,214 | -34% | 1 | 1 | 0% | 2,287 | 2,676 | +17% | 0 | 0 | — |
case-08 | pass→pass | 17,473 | 12,480 | -29% | 1 | 1 | 0% | 2,376 | 3,217 | +35% | 0 | 0 | — |
case-09 | pass→pass | 15,688 | 15,834 | +1% | 1 | 1 | 0% | 2,377 | 3,229 | +36% | 0 | 0 | — |
case-10 | pass→pass | 14,875 | 13,994 | -6% | 1 | 1 | 0% | 2,479 | 3,596 | +45% | 0 | 0 | — |
case-11 | fail→fail | 26,664 | 16,937 | -36% | 1 | 1 | 0% | 3,842 | 4,004 | +4% | 0 | 0 | — |
case-12 | fail→pass | 15,855 | 14,318 | -10% | 1 | 1 | 0% | 2,544 | 3,413 | +34% | 0 | 0 | — |
case-13 | pass→pass | 14,469 | 17,215 | +19% | 1 | 1 | 0% | 2,390 | 4,218 | +76% | 0 | 0 | — |
case-14 | pass→pass | 16,956 | 17,942 | +6% | 1 | 1 | 0% | 2,810 | 3,857 | +37% | 0 | 0 | — |
case-15 | pass→pass | 17,254 | 12,599 | -27% | 1 | 1 | 0% | 3,161 | 3,536 | +12% | 0 | 0 | — |
case-16 | pass→pass | 13,299 | 5,420 | -59% | 1 | 1 | 0% | 1,879 | 1,877 | -0% | 0 | 0 | — |
case-17 | pass→pass | 11,961 | 12,515 | +5% | 1 | 1 | 0% | 1,938 | 3,214 | +66% | 0 | 0 | — |
case-18 | pass→pass | 25,984 | 13,824 | -47% | 1 | 1 | 0% | 1,988 | 3,268 | +64% | 0 | 0 | — |
case-19 | pass→pass | 19,266 | 20,760 | +8% | 1 | 1 | 0% | 2,928 | 4,093 | +40% | 0 | 0 | — |
case-20 | pass→pass | 14,044 | 16,782 | +19% | 1 | 1 | 0% | 2,463 | 3,993 | +62% | 0 | 0 | — |
case-21 | pass→pass | 22,932 | 29,698 | +30% | 1 | 1 | 0% | 3,922 | 5,809 | +48% | 0 | 0 | — |
case-22 | pass→pass | 19,921 | 24,647 | +24% | 1 | 1 | 0% | 2,967 | 4,056 | +37% | 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 +9 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.