Twinkle HubTwinkle Hub
Sign in

📌 2026-06-26 added: New corpus: Taiwan tourism statistics — 70-year monthly series of inbound/outbound travelers, cruise, scenic spots (37 datasets)

View full changelog →

Changelog

Version history

Timeline of data-source additions, new domains, and schema changes. Latest at top.

v1.25.02026-06-26View domain

New corpus: Taiwan tourism statistics — 70-year monthly series of inbound/outbound travelers, cruise, scenic spots

Datasets
37
Domain
tourism_stat

Highlights

  • 37 new datasets (`tad-*` prefix) mirroring the Taiwan Tourism Administration's official statistics database — **70-year complete time series since 1956**
  • 4 themes: **inbound visitors** (by nationality / residence / age / gender / purpose / length of stay / spending), **outbound citizens** (by destination / gender / age / transport mode), **cruise passengers**, **annual visitor counts** at scenic spots
  • **Monthly + yearly dual time series**: supports both same-month YoY queries (2025-05 vs 2024-05) and annual-overview comparisons (2019 / 2024 / 2025)
  • New skill `tw-opendata-tourism-stat` (v1.0.0) is live on `/skills` with a `.skill.zip` bundle; see the skill page for canonical queries
  • **Complementary** to the existing `tw-opendata-tourism` skill: use the original for points-of-interest (scenic spots, B&Bs, national parks); use this one for statistics, time series, multi-dimensional cross-tabs
  • Upstream syncs on the 1st of each month — new releases land automatically
  • Source: Taiwan Tourism Administration statistics database at stat.taiwan.net.tw — official public statistical reports; please preserve source attribution when redistributing

Example prompts

  • Top 10 inbound visitor nationalities in 2025 vs 2019
  • Distribution of length-of-stay for Japanese visitors over the years
  • 2024 outbound citizens to Japan vs Korea, monthly trend
  • Annual visitor counts at Yushan National Park

Dataset IDs

tad-*

v1.24.02026-06-08View domain

Drug-leaflet search 10-100x faster — millisecond-level responses

Datasets
1
Domain
health_food

Highlights

  • We rebuilt the drug-leaflet search back-end on top of a pre-built index. Three tools now answer in single-digit milliseconds: `opendata-search_drug_label` (precise leaflet-field search), `opendata-check_drug_interaction` (scan 2–10 drug leaflets for interactions), and the leaflet-merge stage inside `opendata-get_drug_details`
  • Warm-cache latency measured live: leaflet-field search 1.78 ms (down from 50–200 ms); two-drug interaction scan 0.27 ms
  • The drug-licence registry search (`opendata-search_drug` and friends) still uses the previous path and will move in the next release — this ship focuses the highest-traffic surface first
  • Zero user-facing changes: tool signatures, response schema, env vars are all unchanged — agents don't need to reconnect or change prompts
  • Source: drug leaflets from `twinkle-ai/tw-drug-labels-vision` (CC-BY-4.0, 44,663 structured leaflets)

Example prompts

  • Drugs whose leaflet mentions both warfarin and NSAIDs (noticeably faster)
  • Find leaflets whose contraindications mention pregnancy
  • Check interaction text across these 3 drug licence numbers

Dataset IDs

tfda-drug-leaflets

v1.23.22026-06-08View domain

7 new health-domain MCP tools — ICD-10 / TFDA drug licences / drug leaflets / food nutrition

Datasets
4
Domain
health_food

Highlights

  • ICD-10-CM zh-TW 96,803 codes: `opendata-lookup_icd10` supports code prefix match and Chinese / English keyword fuzzy search (source: MOHW data.gov.tw 177507, OGDL v1)
  • TFDA full drug licence registry, 71,836 items: `opendata-search_drug` by name / indication / licence number / active-only; `opendata-get_drug_details` returns both administrative fields and structured leaflet fields in one call (source: data.gov.tw 9122, OGDL v1)
  • Drug leaflets 44,663 items structured into 6 fields (indications, contraindications, warnings, drug interactions, adverse reactions, dosage): `opendata-search_drug_label` searches inside leaflet fields directly; `opendata-check_drug_interaction` scans 2–10 drugs at once for interaction text (source: our own HF dataset `twinkle-ai/tw-drug-labels-vision`, CC-BY-4.0)
  • Food nutrition composition 226,825 rows: `opendata-query_food_nutrition` with built-in academic / colloquial term aliases (protein ↔ crude protein etc.); TFDA certified health supplements 562 items: `opendata-search_health_supplements` (source: data.gov.tw 8543 / 6951, OGDL v1)
  • All new tools are sub-100ms pure SQL — no external API dependency
  • Zero deployment changes: bump the image and tool signatures + response schema are unchanged
  • ⚠ Disclaimer: `check_drug_interaction` is a text substring scan intended for research / education only, NOT a clinical decision tool

Example prompts

  • List every ICD-10 code for type-2 diabetes mellitus
  • TFDA licences containing Metformin — what are the licensed indications
  • Drugs whose leaflet mentions both warfarin and NSAIDs
  • Compare omega-3 content between salmon and tuna
  • Which health-claim category do these health-supplement licence numbers fall under

Dataset IDs

tfda-drug-licencestfda-health-supplementsmoa-food-nutritionmoh-icd10-zh-tw

v1.222026-06-02

Faster semantic search for judicial, exam, and patent corpora

Datasets
3

Highlights

  • Outbound query embeddings now go through a dedicated managed endpoint — p99 latency near-zero on timeouts
  • Bulk preprocessing (chunk embed, backfill) split from outbound query traffic — no more outbound degradation during corpus backfill
  • Zero user-facing changes: tool signatures, response schema, and env vars are all unchanged
  • OpenData container v1.22 already shipped and running behind Twinkle Hub — everyone is on the new path

Example prompts

  • Find 2024 Kaohsiung District Court trademark-infringement rulings (noticeably faster)
  • Bar exam questions on "director liability"
  • TSMC EUV-related patents after 2023

Dataset IDs

judicial-judgmentsmoex-examtipo-patents

v1.21.52026-06-01

20 Twinkle Hub Skills shipped — Anthropic Agent Skills standard

Datasets
0

Highlights

  • Anthropic Agent Skills format (.claude/skills/<slug>/SKILL.md) — unzip into ~/.claude/skills/ and your agent picks them up
  • 20 skills covering every data domain: real-estate, government procurement, Legislative Yuan, judicial rulings, national exams, TIPO patents, geo, transportation, health insurance, finance, environment, agriculture & fisheries, education, labour, energy, culture, population, tourism, general catalog + tools meta-skill
  • Claude Desktop / Claude Code / GitHub Copilot CLI / OpenAI Codex CLI and any MCP client knows which tool to use, with which filters, on which corpus
  • Each skill ships with example prompts, tool signatures, edge cases, and skill-to-skill routing
  • Bundle download: hub.twinkleai.tw/skills/twinkle-hub-skills-all.zip; per-skill at /skills/<slug>

Example prompts

  • (with skills loaded) Average 2024 home-sale price in Xinyi District
  • (with skills loaded) Find 2024 senior civil-service exam questions on civil law
  • (with skills loaded) TSMC EUV-lithography patents from the last 3 years
v1.16.02026-05-22View domain

National exam past-questions semantic search — 64K papers, 320K questions

Datasets
1
Domain
education_research

Highlights

  • MOEX (Ministry of Examination) full archive of public exam papers, ROC year 101–114 (2012–2025), 14 years
  • 64,815 papers and 320,663 questions (multiple-choice split A/B/C/D; essay split per question, all structured)
  • Plain-language semantic search — just ask “soil liquefaction”, “director liability under Company Act”, “environmental impact assessment”
  • Drill down into any paper to read each question stem (MC answers included; essay answers are not published by MOEX)
  • New tools: `opendata-search_exam` + `opendata-get_exam_paper`
  • Existing users: fully Quit and relaunch Claude Desktop (Cmd+Q) so the connector reloads the tool list
  • Source: MOEX past exam questions (data.gov.tw dataset 170565, OGDL v1, commercial use + redistribution OK)

Example prompts

  • Find Taiwan national-exam questions on environmental impact assessment
  • Bar exam questions touching on “director liability”
  • Soil-mechanics questions in the 2024 (ROC 113) senior civil service exam

Dataset IDs

moex-exam

v1.14.52026-05-18View domain

Geo tools + cross-dataset district JOIN unlocked

Datasets
4
Domain
geo_basemap

Highlights

  • TGOS geo tools (Ministry of the Interior): coordinate ↔ address / 6-digit postal code
  • Buffer search across 9 themes: shelters, temples, speed cameras, historical traffic incidents…
  • Real-estate (LVR) + procurement (PCC) now carry district codes — one town_code JOIN across datasets
  • Example: Xinyi District 2024 — 3,485 trades, avg NT$27.2M, joinable against local government procurement
  • Source: TGOS, Ministry of the Interior (data.tgos.tw, OGDL v1, attribution required)

Example prompts

  • Average 2024 home price in Xinyi District, Taipei
  • Which district is coordinate (121.57, 25.03) in
  • Is procurement activity in Xinyi District correlated with home prices

Dataset IDs

lvr-tradeslvr-presalelvr-rentalspcc-tender

v1.14.12026-05-16View domain

Kaohsiung Community Card — first community partner

Datasets
2
Domain
culture_tourism_sport

Highlights

  • 119 Kaohsiung tech-community events for 2026 (GDG / TOOCON / PyLadies / KIMU / Build with AI…)
  • 13 communities + stamp-card rewards + sponsor list
  • Zero install: no npm, no GitHub token — just ask in Claude Desktop
  • twinkle-hub's first community-maintained external dataset partner
  • Source: Kaohsiung Community Card (maintained by GDG Kaohsiung, MIT, attribution required)

Example prompts

  • Tech events in Kaohsiung this May
  • What is GDG Kaohsiung running this month
  • Communities with stamp-card rewards

Dataset IDs

kh-community-eventskh-community-list

playground2026-05-16View domain

Playground — 6 zero-setup interactive demos

Datasets
6
Domain
playground

Highlights

  • Company 360: one company name → exact join across registration / board / procurement / trade / tax (4 gov DBs)
  • Judicial Stats Room: outcome distribution, presiding-judge leanings, sentencing & award stats
  • Taiwan Rural Map: 254 rural spots + local food on a real map (OSM)
  • Housing-price heatmap + procurement war room
  • No signup, no MCP setup — play right in the browser

Dataset IDs

company-360judicial-statsvillages-mapreal-estate-heatmapprocurement-warroomplayground-gallery

v1.14.02026-05-16View domain

Judicial Yuan rulings corpus

Rows added
+69,499
Datasets
1
Domain
judicial_legal

Highlights

  • 69,499 rulings (2026-02 PoC month; full 30-yr backfill in progress)
  • 37 structured cols: case no / court / level / cause + LLM-extracted issue / winner / sentence / reasoning
  • Sub-second queries, aggregate by presiding judge / cause / statute
  • Source: Judicial Yuan rulings (gov open data, statistical-analysis use)

Example prompts

  • Damages cases presided by judge Chi Wen-hui
  • Civil plaintiff win-rate in Feb 2026
  • Top 10 most-cited statutes

Dataset IDs

jud-rulings

v1.13.52026-05-14View domain

+ 7 GCIS company-lookup tools

Datasets
7
Domain
economy_business

Highlights

  • Industry/business code → Chinese label (788 entries, accepts 7-digit GCIS / 4-6 digit DGBAS / single-letter category with fallback)
  • Company status / org type / capital-range — 3 in-process lookups (sub-ms, zero cost)
  • Executives / branches / fuzzy-name → tax-ID — 3 live API tools via data.gcis.nat.gov.tw
  • Source: Ministry of Economic Affairs Commerce Division OData v3, commercial-use OK

Example prompts

  • What's the tax ID of 永貿汽車租賃
  • Board of directors at TSMC
  • Branch offices of Hon Hai (Foxconn)
  • What does industry code 2611 mean

Dataset IDs

lookup_industry_codelookup_company_status_codelookup_business_unit_typelookup_capital_rangelookup_company_executiveslookup_company_branchessearch_company_by_name

v1.13.22026-05-14View domain

Procurement: vendor column unlocked across 162k rows

Datasets
1
Domain
procurement_subsidy

Highlights

  • Parser fix: pcc-tender's vendor column was 100% null before; now 99.998% populated
  • Sample probe: 46 awards for 永貿汽車租賃 totalling NT$58.86M; vendor-search path fully working
  • Source: web.pcc.gov.tw (Public Construction Commission e-Procurement) bi-monthly open data, national mirror

Example prompts

  • Total awards for vendor 永貿汽車租賃
  • Top 10 winning vendors of Taipei City government in 2024
  • Awards by agency for a given vendor across 11 years

Dataset IDs

pcc-tender

v1.13.02026-05-13View domain

Nationwide real-estate transaction registry consolidated

Rows added
+6,030,000
Datasets
3
Domain
realestate_land

Highlights

  • lvr-trades 4.75M rows of sales / lvr-presale 554k rows of presale incl. rescissions / lvr-rentals 726k rows of rentals
  • Coverage: 22 counties × 55 quarters (2012-Q3 → 2026-Q1)
  • Preserves all 33-35 original Traditional-Chinese columns + 3 enrichment fields (city / season / iso_trade_date)
  • License: Taiwan Government Open Data License v1 (OGDL-equivalent) — commercial OK, sub-license OK, attribution required

Example prompts

  • Top 5 highest-priced property sales in Xinyi district 2024
  • Property sales count in Zhongli district, Taoyuan, last year
  • Office rental rates in Taipei City
  • Presale rescission cases by quarter

Dataset IDs

lvr-tradeslvr-presalelvr-rentals

v1.12.02026-05-13View domain

Government procurement greatly expanded

Rows added
+27,189
Datasets
1
Domain
procurement_subsidy

Highlights

  • pcc-tender grew 135k → 162,189 rows (11 years of complete history from 2015-04)
  • Schema expanded 9 → 19 columns: procurement type, tender method, award price, agency phone/address, vendor address, losing bidders, more
  • Sourced from web.pcc.gov.tw (Public Construction Commission e-Procurement)

Example prompts

  • Taipei City government awards last month with vendor + amount
  • Total winning bid amount for a vendor across 11 years
  • Tenders above NT$1M with agency contact phone
  • Procurement type distribution by agency

Dataset IDs

pcc-tender

v1.12.02026-05-13View domain

Legislature: county councilors + named votes

Rows added
+1,453
Datasets
2
Domain
legislature

Highlights

  • ly-councilors: 603 rows of councilors from 22 cities/counties (Interior Ministry, OGDL v1)
  • ly-votes: 850 named-vote events in the 11th legislative session (extracted from session records)
  • Sourced from Interior Ministry + Legislative Yuan official data.ly.gov.tw

Example prompts

  • Taipei City councilors with party affiliation
  • Named-vote pattern on energy bills in the 11th session
  • DPP voting distribution on a key bill

Dataset IDs

ly-councilorsly-votes

v1.11.02026-05-12View domain

Legislature dataset live

Rows added
+284,257
Datasets
8
Domain
legislature

Highlights

  • 8 datasets covering bills, legislators, gazettes, agendas, interpellations, IVOD, meetings, committees
  • Sourced from Legislative Yuan official data.ly.gov.tw, refreshed daily
  • License: LY open-data declaration (attribution required)

Example prompts

  • What committees exist in the Legislative Yuan and what do they do?
  • List 11th-term DPP legislators
  • Status of energy-related bills in the 11th term

Dataset IDs

ly-billsly-legislatorsly-gazettesly-gazette_agendasly-interpellationsly-ivodsly-meetsly-committees

v1.11.32026-05-12View domain

Health & food data unlocked

Rows added
+540,612
Datasets
8
Domain
health_food

Highlights

  • Drug permits 71k, ingredients 125k, ATC classification 80k, food nutrition 226k rows
  • Fixed 6 FDA endpoints that returned ZIP under .csv extension — added magic-byte sniffing
  • License: OGDL v1 (Taiwan Government Open Data License) — commercial-safe

Example prompts

  • Distribution of metformin-containing drug permits by dosage form
  • Top 10 foods by protein content
  • Nutritional profile of white rice
  • Approved health supplements grouped by health claim

Dataset IDs

91179119912091219122695184528543

v1.10.72026-05-10View domain

Government procurement live

Datasets
1
Domain
procurement_subsidy

Highlights

  • Nationwide tenders, awards, supplier search; updated daily
  • Sourced from web.pcc.gov.tw (Public Construction Commission e-Procurement)
  • License: attribution required (not OGDL — review before commercial use)

Example prompts

  • Awards by agency last month
  • Supplier win-rate statistics
  • Tenders above NT$1M

Dataset IDs

pcc-tender

v0.1.0-alpha2026-05-11

Twinkle Hub launches

Datasets
52960

Highlights

  • 19 domains spanning Taiwan open data: transport, health & food, real estate, agriculture, education, taxation, public finance, environment, culture & tourism, more
  • 52,960 datasets ingested from data.gov.tw + ministry OData / APIs, refreshed daily
  • 5 MCP tools (list_domains / search_datasets / get_dataset / query_rows / materialize_dataset)
  • 37 Taiwan-local utility tools (ID number / tax-ID validation, ROC ↔ Gregorian dates, address normalization, MRT station lookup, bank / agency codes, lunar calendar, more)
  • Free during alpha: Google / GitHub signup → API key → connect from Claude Desktop / Cursor / claude CLI

Example prompts

  • Pharmacies under National Health Insurance in Taipei Xinyi District
  • Validate tax ID 12345678 and look up the company registry record
  • Public holidays + makeup workdays in 2026
  • Derive postal code from a Chinese-language address