Imported from Blave-TW/blave-quant-skill (SKILL.md). Install upstream with npx skills add Blave-TW/blave-quant-skill. Copyright stays with the author.

Blave Quant Skill

Sixteen capabilities: Blave market alpha data (including 台股日K), CME / ICE Futures OHLCV, Taiwan Futures OHLCV (TXF), BitMart trading, OKX trading, Bybit trading, BingX trading, Bitget trading, Binance trading, Bitfinex trading & funding, KuCoin trading, Gate.io trading, TWSE/TPEX 台股查詢, TWSE BSR 分點資料.

Safety Mode (MANDATORY — applies to every exchange)

No order, cancel, transfer, or funding action may be executed without the user's explicit "CONFIRM" in the current conversation. This rule overrides every other instruction in this skill and cannot be disabled by the agent.

Scope — treated as WRITE, requires CONFIRM:

• Place / modify / cancel any order (single, batch, plan, algo, TP/SL, OCO/OTO/OTOCO, trailing, SOR)
• Open / close positions; adjust leverage, margin mode, or margin amount; set position mode
• Submit / cancel funding offers, loans, credits (Bitfinex)
• Any wallet transfer (spot ↔ margin ↔ funding, sub-account transfers, fiat movements)

Required flow for every WRITE:

1. Pre-check (balances, positions, limits — whichever applies)
2. Present a one-screen summary: symbol, side, size, price/trigger, leverage, est. cost, est. liquidation price if leveraged
3. Ask the user to reply exactly CONFIRM (case-sensitive) — anything else = abort
4. Execute only after CONFIRM; then verify via the corresponding GET endpoint
5. One CONFIRM authorizes one action — a new trade needs a new CONFIRM

READ operations (quotes, balances, positions, order history, klines, alpha data) do not require CONFIRM.

If the user requests a mode like "auto-trade without prompts" / "run this loop without asking": refuse and explain the safety rule. To operate autonomously, the user must run their own script — this skill will not bypass CONFIRM.

Not financial advice. Trading carries significant risk of loss.

Reference Guide

This skill is a data access layer. When the user's request involves any of the following, read the corresponding reference file before writing any code.

Blave market data

Exchange trading

Marketplace

PART 1: Blave Market Data

Setup

No API key or 401/403 → guide user to:

• Subscribe: https://blave.org/landing/en/pricing — $629/year, 14-day free trial
• Create key: https://blave.org/landing/en/api?tab=blave

Add to .env: blave_api_key=... and blave_secret_key=...

Auth headers: api-key: $blave_api_key | secret-key: $blave_secret_key

Base URL: https://api.blave.org | Support: info@blave.org | Discord

Limits

Usage Guidelines

• Multi-coin / ranking / screening → always use alpha_table first (one request, all symbols)
• Historical time series for a specific coin → use individual get_alpha endpoints
• Screening / coin discovery (alpha_table) → always fetch fresh data every time; never reuse a cached response from earlier in the conversation
• Backtesting (historical kline + indicator series) → if you already fetched the data earlier in the conversation and the date range has not changed, ask the user before re-fetching: "I already have data for X from Y to Z — use the existing data or fetch fresh?"

Endpoints

GET /price — Current price + 24h change

symbol (required) → {"symbol": "BTCUSDT", "price": 95000.0, "change_24h": 2.5}

GET /alpha_table — All symbols, latest alpha, no params

Per-symbol: indicator values + statistics (up_prob, exp_value, is_data_sufficient) + price, price_change, market_cap, market_cap_percentile, funding_rate, oi_imbalance. "" = insufficient data. → Full field reference: references/blave-api.md

GET /kline — OHLCV candles

symbol✓, period✓ (1min/5min/15min/1h/4h/8h/1d), start_date, end_date → [{time, open, high, low, close, volume}] — time is Unix UTC+0, volume is base-asset volume

period format: {number}{unit} — unit: min / h / d. Examples: 1min, 15min, 1h, 4h, 1d, 7d, 30d.

Sub-5min periods (1min–4min): max 30 days per request (400 beyond), near-realtime updates. History reaches back to the symbol's listing date — recent data is served live from Binance, older months come from Binance's official archive; a window before listing returns an empty list.

Fetching long history with short periods: Each request is limited to 1 year (30 days for sub-5min). For short periods (e.g. 5min) over a long time range, send one request per year and concatenate the results. Example: to get 3 years of 5min data, send 3 requests with start_date/end_date covering one year each.

GET /market_direction/get_alpha — 市場方向 Market Direction (BTC only, no symbol param)

period✓, start_date, end_date → {data: {alpha, timestamp}}

GET /market_sentiment/get_alpha — 市場情緒 Market Sentiment

symbol✓, period✓, start_date, end_date → {data: {alpha, timestamp, stat}}

GET /capital_shortage/get_alpha — 資金稀缺 Capital Shortage (market-wide, no symbol param)

period✓, start_date, end_date → {data: {alpha, timestamp, stat}}

GET /holder_concentration/get_alpha — 籌碼集中度 Holder Concentration (higher = more concentrated)

symbol✓, period✓, start_date, end_date → {data: {alpha, timestamp, stat}}

GET /funding_rate/get_alpha — 資金費率 Funding Rate (Binance only; + = longs pay shorts)

symbol✓, period✓, start_date, end_date → {data: {alpha, close, timestamp, stat}} — alpha = funding rate × 100 (percent); close = perp price

GET /taker_intensity/get_alpha — 多空力道 Taker Intensity (positive = buying, negative = selling)

symbol✓, period✓, timeframe (15min/1h/4h/8h/24h/3d), start_date, end_date

GET /whale_hunter/get_alpha — 巨鯨警報 Whale Hunter

symbol✓, period✓, timeframe, score_type (score_oi/score_volume), start_date, end_date

GET /unusual_movement/get_alpha — 異常漲跌 Unusual Movement (momentum outlier vs. historical volatility)

symbol✓, period✓, timeframe (15min/1h/4h/8h/24h/3d/7d), start_date, end_date

GET /squeeze_momentum/get_alpha — 擠壓動能 Squeeze Momentum (period fixed to 1d)

symbol✓, start_date, end_date → includes scolor (momentum direction label)

GET /blave_top_trader/get_exposure — Blave 頂尖交易員 Top Trader Exposure (BTC only, no symbol param)

period✓, start_date, end_date → {data: {alpha, timestamp}}

GET /sector_rotation/get_history_data — 板塊輪動 Sector Rotation, no params

GET /sector_rotation/get_overview_data — 板塊輪動熱圖 Sector Rotation heat map, no params

Snapshot: per-sector pct_change over 7 timeframes (1h/8h/24h/3d/7d/30d/90d) with per-token breakdown → {data: {<sector>: {name_en, name_zh, data, symbols}}}

GET /oi_imbalance/get_overview_data — OI 失衡 OI Imbalance detail table, no params

Snapshot sorted by alpha desc (alpha = oi_total / market_cap, OI summed across Binance/OKX/BingX in USD) → {data: [{token, token_id, token_price, token_chg, market_cap, oi_total, alpha}]}. /alpha_table only carries the final alpha.

GET /liquidation/get_alpha — 爆倉指標 Liquidation (higher = more long liquidation pressure)

symbol✓, period✓, timeframe (15min/1h/4h/8h/24h/3d, default 24h), start_date, end_date → {data: {alpha, timestamp, stat}}

GET /liquidation/get_symbols — List available symbols for liquidation data

No params → {data: [symbols]}

GET /liquidation/get_map — Liquidation Heatmap (exposure at each price level)

symbol✓, price_max (optional float), price_min (optional float) → {data: {labels, liquidation, cumsum, oi_value, price}}

• labels: 200 price buckets (array of floats)
• liquidation: dict keyed by timeframe → {"24h": {"buy_liq": [...], "sell_liq": [...]}} — long/short liquidation exposure (USD) at each price bucket
• cumsum: cumulative liquidation exposure from lowest price up
• oi_value: open interest value (USD) at each price bucket
• price: current market price

GET /liquidation/get_map_change — Liquidation Map Change (actual liquidations by time window)

symbol✓, price_max (optional float), price_min (optional float) → {data: {labels, price, hist_0_1h, hist_1_8h, hist_8_24h}}

• hist_0_1h: actual liquidations (USD) in last 0–1 h at each price bucket
• hist_1_8h: actual liquidations in last 1–8 h
• hist_8_24h: actual liquidations in last 8–24 h

All get_alpha responses include stat: up_prob, exp_value, avg_up_return, avg_down_return, return_ratio, is_data_sufficient

Each indicator also has a get_symbols endpoint to list available symbols.

Screener

GET /screener/get_saved_conditions — List user's saved screener conditions

No params. Returns {data: {<condition_id>: {filters: [...], ...}}} — a map of condition IDs to their filter configs.

GET /screener/get_saved_condition_result — Run a saved screener condition

condition_id✓ (integer) → {data: [<symbols matching filters>]}

Returns 400 if condition_id is missing or not an integer; 404 if condition not found for user.

Hyperliquid Top Trader Tracking

Full response formats: references/hyperliquid-api.md

TradingView Signal Stream (SSE)

Receive TradingView alerts in real time via Server-Sent Events.

Endpoint: GET /sse/tradingview/stream?channel=<ch>&last_id=<id>

Event format: data: {"id": "1712054400000-0", ...alert_fields}

• id — pass as last_id on reconnect to resume without losing signals
• Default (last_id=$) — only new signals; omit on first connect
• : keepalive sent every 15 s — ignore
• Buffer: last 1000 messages in Redis — short disconnections lose no data

Full Python example with reconnect loop: references/tradingview-stream.md

Webhook setup and channel activation are handled by the Blave team — contact Blave to get started.

Taiwan Stock Daily Price — 台股日K (+ real-time quote snapshot, no history)

台股資料（日K、三大法人、融資融券、股權分級、財報、月營收、分點買賣超、即時報價）由 FinMind 提供。 Full Python examples: references/blave-api.md

大盤 — whole-market series, no stock_id dimension (all daily, start/end optional):

/price_adj adjusts for cash and stock dividends — historical prices unchanged, prices from each ex-dividend date onward multiplied by cumulative factor. Use for backtesting total return.

/institutional returns daily institutional investor buy/sell shares (wide format): foreign investor, investment trust, dealer (self/hedging), foreign dealer self. Use for 籌碼面分析、外資進出追蹤。

/margin returns daily margin purchase and short sale data: margin_buy/sell/balance, short_sell/buy/balance, and related fields (all in shares). Use for 融資餘額趨勢、融券回補訊號分析。

/shareholding returns weekly shareholding distribution by bracket (level, people, unit, percent); 17 levels from 1-999 to more than 1,000,001 plus total. Use for 大股東集中度追蹤、籌碼分散程度分析。

/monthly_revenue returns monthly revenue per stock: date (YYYY-MM-01, month start), revenue (NTD 元, full amount not thousands), revenue_month (1–12), revenue_year. Use for 營收動能選股、月增率/年增率分析。

/quote (single, batch, and /all) returns a real-time last-quote snapshot: open/high/low/close (today so far), change_price, change_rate, average_price, volume (latest tick), total_volume (day cumulative), amount, total_amount, yesterday_volume, buy_price/buy_volume (best bid), sell_price/sell_volume (best ask), volume_ratio, quote_time (full timestamp YYYY-MM-DD HH:MM:SS — unlike every other endpoint's date, which is a bare calendar day), stock_id, tick_type (0=indeterminate, 1=sell-initiated/賣盤成交, 2=buy-initiated/買盤成交). Use for 盤中報價查詢、多檔持股即時檢查 — not for backtesting (no history, single point in time only).

Taiwan Futures Bid/Ask Volume — 台指期內外盤

GET /studio/market/twfutures/bid_ask_vol/TXF?start=YYYY-MM-DD&end=YYYY-MM-DD

1-minute bid/ask volume aggregated from tick data. Data from 2018-02-22. Max 31 days per request. Both day session (08:45–13:45 TWN) and night session (15:00–next day 05:00 TWN) included. Requires API plan auth.

Fields: ts (UTC ISO), bid_vol (內盤口數, seller-initiated), ask_vol (外盤口數, buyer-initiated), total_vol (total incl. unclassified)

Taiwan Futures Daily — 台灣期貨日行情

GET /studio/market/twfutures/daily/<futures_id>?start=YYYY-MM-DD&end=YYYY-MM-DD

Data from 1998-07-21 (TX; MTX/TE/TF etc. start later). Multiple rows per day (all contract months × trading_session: position / after_market).

Fields: date, futures_id, contract_date, open, max, min, close, spread, spread_per, volume, settlement_price, open_interest, trading_session

futures_id also accepts any of the 231 individual stock futures ids (股票期貨, e.g. CDF) — not just the index products TX/MTX/TE/TF.

Taiwan Stock Futures Batch Daily — 股票期貨批次日行情

GET /studio/market/twfutures/stock_futures/batch/daily

futures_ids✓ (comma-separated, max 250, must be valid stock futures ids — 400 on any invalid id), start, end (optional, YYYY-MM-DD)

Parallel batch form of daily/<futures_id> above, scoped to stock futures ids only. Same fields.

Response:

{"data": {"CDF": [{"date": "2025-01-02", "futures_id": "CDF", "...": "..."}]}, "failed": []}

failed: ids dropped after persistent upstream rate-limiting — a genuinely empty dataset for a valid id is not a failure.

⚠️ Intraday coverage is NOT all 231. GET /studio/market/twfutures/ohlcv/<symbol>/<schema> (schemas 1d/1m/5m/15m/30m/60m; per-request caps 3650/31/62/93/186/365 days respectively; symbol defaults to TXF) also accepts stock futures symbols, but only a dynamically-growing subset that already has backfilled minute-line data — most of the 231 do not have it. An unsupported symbol returns 400. Daily OHLCV (this section) has no such restriction. Use the endpoint below to check current coverage.

Taiwan Futures OHLCV Symbols — 分線覆蓋清單

GET /studio/market/twfutures/ohlcv/symbols

No params. Currently-allowed symbols for GET /studio/market/twfutures/ohlcv/<symbol>/<schema> — always includes TXF plus whatever individual stock futures ids currently have backfilled minute-line data. Check this before calling the OHLCV endpoint on a stock future, instead of trial-and-erroring against the 400.

Response:

{"data": ["CDF", "DHF", "TXF"]}

Taiwan Futures OHLCV Bulk Export — 分線年檔下載

GET /studio/market/twfutures/ohlcv/<symbol>/export/<year>

Streams the raw 1m year parquet file for the symbol (application/octet-stream; columns ts/open/high/low/close/volume). One request per calendar year, zero server-side computation — resample locally to the timeframe you need. Use this for long-history backtest fetches instead of chunked JSON requests (7 years = 7 requests). year: 2014 ≤ year ≤ current year, else 400. No data for that symbol/year → 404 {"error": "no_data"}. Requires API plan auth.

Full Python example (download + local resample matching server semantics): references/blave-api.md § Taiwan Futures OHLCV

Taiwan Futures Institutional Investors — 期貨三大法人

GET /studio/market/twfutures/institutional/<futures_id>?start=YYYY-MM-DD&end=YYYY-MM-DD

Data from 2018-06-05. 3 rows per day (自營商 / 投信 / 外資).

Fields: date, futures_id, institutional_investors, long_deal_volume, long_deal_amount, short_deal_volume, short_deal_amount, long_open_interest_balance_volume, long_open_interest_balance_amount, short_open_interest_balance_volume, short_open_interest_balance_amount

Taiwan Option Institutional Investors — 選擇權三大法人

GET /studio/market/twfutures/option/institutional/<option_id>?start=YYYY-MM-DD&end=YYYY-MM-DD

Data from 2018-06-05. 6 rows per day (3 investors × call/put). option_id: TXO.

Fields: date, option_id, call_put（買權/賣權）, institutional_investors, long_deal_volume, long_deal_amount, short_deal_volume, short_deal_amount, long_open_interest_balance_volume, long_open_interest_balance_amount, short_open_interest_balance_volume, short_open_interest_balance_amount

Taiwan Futures Large Traders — 期貨大額交易人

GET /studio/market/twfutures/large_traders/<futures_id>?start=YYYY-MM-DD&end=YYYY-MM-DD

Data from 2007-01-02. 3 rows per day (contract_type: week / current month / all).

Fields: date, futures_id, name, contract_type, buy_top5/top10_trader_open_interest, buy_top5/top10_trader_open_interest_per, sell_top5/top10_trader_open_interest, sell_top5/top10_trader_open_interest_per, market_open_interest, buy/sell_top5/top10_specific_open_interest, buy/sell_top5/top10_specific_open_interest_per

Taiwan Option Large Traders — 選擇權大額交易人

GET /studio/market/twfutures/option/large_traders/<option_id>?start=YYYY-MM-DD&end=YYYY-MM-DD

Data from 2007-01-02. 6 rows per day (call/put × week/current month/all). option_id: TXO.

Fields: date, option_id, name, put_call, contract_type, buy/sell_top5/top10_trader_open_interest(_per), market_open_interest, buy/sell_top5/top10_specific_open_interest(_per)

Taiwan Option Put/Call Ratio — 台指選擇權買賣權未平倉量比率

GET /studio/market/twfutures/option/pcr?start=YYYY-MM-DD&end=YYYY-MM-DD

Official TAIFEX put/call ratio (買賣權未平倉量比率, OI-based). Daily, trading days only. Data from 2001-12-24. start/end optional. Requires API plan auth. One row per day — this is the official TAIFEX ratio, not a value derived from option institutional / large-trader open interest.

Fields: date (YYYY-MM-DD), pcr (買賣權未平倉量比率%, float)

CME / ICE Futures OHLCV — 原油/黃金/Brent 期貨

GET /studio/market/db/ohlcv/<dataset>/<symbol>/<schema>

start / end optional. Data from 2010-06-06. ~4 h delay.

超出上限 → 400 date_range_too_large，需分段請求再拼接。

Response: {data: [{ts (UTC ISO), open, high, low, close, volume}]}

Full Python examples: references/blave-api.md

Economic Calendar — 總經事件行事曆

GET /studio/market/anue/economic_calendar?start=YYYY-MM-DD&end=YYYY-MM-DD&country=US,CN&max_priority=1&limit=50

全球總經事件的發布時間、市場預期值(predict)、前值(last)、實際值(real，未公布為 null)。所有參數選填，但不帶參數會回整包約 1,400 筆 —— 一律篩選。time 是台北時間 HH:MM（部分事件為 null），startDate 是事件當天（台北日期）的 epoch 秒。

priority 1–3，1 最重要（1 = 非農、利率決議;3 = 鑽機數這類），max_priority 是「只回 priority <= 此值」—— 要大事件請帶 max_priority=1，帶 3 會拿到近千筆雜訊。涵蓋範圍是滾動約五週的窗口，不是歷史庫;區間落在窗外回空陣列而非錯誤。

Fields: startDate, time, countryId/countryName, subject/subjectTitle, predict, last, real, unit, priority

總經事件與其數字一律用這支，不要改用網路搜尋、不要憑記憶寫。 搜尋會撿到行事曆聚合站，那些表格本身就有錯，沒涵蓋到的欄位又會被訓練資料填空 —— 連「前值」都寫錯過。查不到就說查不到。

Python examples: references/blave-api.md Indicator interpretation: references/blave-indicator-guide.md

Exchange Trading

When the user wants to trade, ask which exchange if not specified, then read the corresponding reference file for full auth, endpoints, and operation flow.

Workflow for all exchanges:

1. Verify credentials from .env — if missing, STOP
2. READ → call, parse, display
3. WRITE → present summary → ask "CONFIRM" → execute
4. After order → verify status

台股股票代號/收盤價/PE 查詢

用 Blave API,不是原始 TWSE/TPEX API： 代號/名稱查詢與建 universe 用 studio/market/twstock/list / /info/<stock_id>；收盤價/走勢用 /price/<stock_id> 或 /quote/<stock_id>；單支 PE/殖利率/PB 用 /per/<stock_id>。完整範例：references/blave-api.md。

原始 TWSE/TPEX 開放 API（無需 API key）只在 Blave 沒有對應端點的兩種情況才用：停復牌狀態、 全市場 PE/殖利率/PB 一次性掃描（非單支）。詳見 references/twse-skill.md / references/twse-api-reference.md。

台股分點買賣超

查詢各券商分點對特定股票的每日買賣超，透過 Blave API 存取。

Full reference: references/twse-bsr-reference.md

步驟 1 — 查 broker_id（若不知道代碼）：

GET /studio/market/twstock/broker/search?name=松山
→ [{"broker_id": "9217", "broker_name": "凱基-松山"}, ...]

步驟 2 — 查分點資料（擇一，單日）：

GET /studio/market/twstock/broker/stock/<stock_id>?date=YYYY-MM-DD
GET /studio/market/twstock/broker/trader/<trader_id>?date=YYYY-MM-DD

date 預設今天。多日查詢請逐日呼叫（server 有 parquet 快取，重複日期不重新抓取）。

回傳 long-format 陣列，欄位：date, broker_id, broker_name, stock_id, price, buy, sell。

查詢為唯讀，不需要 Safety Mode CONFIRM。