Tw Stock Data

Taiwan Stock Data

Use this skill to help users discover, fetch, validate, and analyze Taiwan market datasets exposed by the tw-stock CLI.

The CLI is the source of truth. Prefer calling tw-stock or uv run tw-stock instead of reimplementing crawler logic in the agent.

The current CLI normalizes common output columns to English snake_case. Before writing analysis code that depends on exact columns, inspect tw-stock describe <dataset> --json, tw-stock fetch <dataset> --schema-only --format json, or references/datasets.md.

Dependency

This skill does not bundle the market data crawler itself. It expects the tw-stock CLI from the tw-stock-cli project to be available.

Use commands in this order:

1. If tw-stock is installed in the current environment, call tw-stock ....
2. If working inside a checkout of the tw-stock-cli project, call uv run tw-stock ... from that repository root, or use uv run --project /path/to/tw-stock-cli tw-stock ... when running from another directory.
3. If neither command is available, tell the user that the skill can identify the right dataset and command, but fetching requires installing or opening the tw-stock-cli project.

First steps

1. If the user is asking what data exists, run:

   tw-stock list-datasets --json
   

2. If the user mentions a specific topic but not a dataset ID, inspect the dataset catalog and map their request to the closest dataset.

3. Before fetching a large dataset, describe it or sample it:

   tw-stock describe taifex.futures-tick --json
   tw-stock fetch taifex.futures-tick --date 2026-04-30 --limit 5 --format json
   

4. For analysis workflows, prefer jsonl for AI/tool pipelines and csv for user-facing files:

   tw-stock fetch twse.stock-price --date 2026-04-30 --format jsonl
   tw-stock fetch twse.stock-price --date 2026-04-30 --format csv --output data.csv
   

5. Use metadata-only flags before fetching when the user needs schema or source context:

   tw-stock fetch twse.stock-price --schema-only --format json
   tw-stock fetch twse.stock-price --source-url-only --format json
   

Dataset selection

Use these common mappings:

• Stock price / open-high-low-close: twse.stock-price, tpex.stock-price
• Stock list / security codes: twse.stock-list, tpex.stock-list
• PER / dividend yield / PBR: twse.stock-per, tpex.stock-per
• Institutional investor buy/sell: twse.institutional-trade, tpex.institutional-trade
• Margin purchase / short sale: twse.margin-trade, tpex.margin-trade
• Foreign holdings: twse.foreign-holding, tpex.foreign-holding
• Total return indices: twse.total-return-index, tpex.total-return-index
• Futures and options market data: taifex.*
• Monthly revenue and financial statements: mops.month-revenue, mops.company-*, mops.income-statement, mops.balance-sheet, mops.cash-flow
• MOPS PDF/electronic-book download indexes: mops.financial-report-electronic-book, mops.annual-report-electronic-book, mops.related-company-reports, mops.sustainability-report, mops.major-shareholder-relationship
• MOPS governance, ESG, insiders, related-party, and corporate actions: inspect references/datasets.md and prefer the exact mops.* dataset before scraping MOPS manually.

For the complete list, read references/datasets.md.

Column expectations

Use normalized column names from the current CLI:

• Price columns are open, high, low, close; do not assume old names such as max or min.
• Valuation columns are per and pbr; do not assume old uppercase names such as PER or PBR.
• Security identifiers are stock_id and stock_name.
• TAIFEX FCM outputs use fcm_id, fcm_name, subtotal, total, market_share, product columns such as mtx, and total product columns such as total_mtx.
• MOPS statement line items may remain in Chinese, but shared identifier columns are normalized to stock_id and stock_name.

Safety and data quality

Taiwan exchange endpoints can change format. If the user needs reliable output, validate first:

tw-stock validate twse.stock-price --date 2026-04-30 --json

If validation fails, report the dataset ID, date/parameters, error code, and source URL. Do not silently substitute another data source unless the user asks for that.

For non-trading days, expect empty data or exchange messages. Ask for a nearby trading day only if the user's date is ambiguous.

Large datasets

Tick data can return hundreds of thousands of rows. Always use --limit when exploring:

tw-stock fetch taifex.options-tick --date 2026-04-30 --limit 20 --format jsonl

Only fetch full tick datasets when the user clearly needs full raw data or provides an output path.

Reference files

• references/datasets.md: dataset IDs, parameters, and examples.
• references/cli-usage.md: command patterns and output formats.
• references/data-caveats.md: known source quirks and edge cases.