shioaji

name

shioaji

description

Use this skill whenever the user works with Shioaji, SinoPac (永豐金), or Taiwan financial markets (TWSE/TPEX/TAIFEX) — even if they don't name the library. Covers all access layers: Python binding (sync/async), `shioaji` CLI, HTTP API server with SSE streaming, dashboard with custom-app embedding, and multi-language HTTP clients (JavaScript/TypeScript, Go, C/C++, C#, Rust, Java/Kotlin). Tasks covered: place/modify/cancel stock/futures/options orders (limit, market, ROD/IOC/FOK, margin, short, odd-lot, combo); real-time streaming (tick, bidask, quote) via Python callbacks or HTTP SSE; historical kbars/ticks/snapshots; account balance/margin/positions/P&L; watchlists, scanners, reserve orders; building HTTP/SSE clients in any language against the Shioaji server. Trigger keywords: shioaji, sinopac, 永豐金, 台股, TWSE, TPEX, TAIFEX, 下單, 即時行情, 台灣股票交易, shioaji CLI, shioaji server, SSE streaming. Not for: US/HK markets, Interactive Brokers, generic ta-lib/pandas indicators unless paired with Shioaji data.

Shioaji Trading API

Shioaji is SinoPac's cross-language, cross-platform trading API for Taiwan financial markets (TWSE/TPEX/TAIFEX). It includes Python bindings, a CLI, and an HTTP API server that lets any programming language trade Taiwan markets.

Shioaji 是永豐金證券的跨語言、跨平台交易 API，提供 Python 綁定、CLI 與 HTTP API 伺服器，讓任何程式語言都能透過 HTTP API 交易台灣市場。

Three access layers / 三種存取層：

• Python — native PyO3 binding (import shioaji) for best performance, sync and async
• CLI — shioaji command-line tool for server management, trading, and data queries
• HTTP API + SSE — REST endpoints + real-time streaming at localhost:8080, accessible from JS/TS, Go, C/C++, C#, Rust, Java/Kotlin, or any HTTP client

Reasoning about connections, timeouts, processes, or state? Read CONCEPTS.md first. It is the shared mental model these three layers assume — which timeout applies on which path, why Python needs no server but CLI/HTTP share one daemon, the per-person connection cap, streaming vs request-reply. Answer those questions from CONCEPTS.md instead of guessing. 推理連線、逾時、行程或狀態時,先讀 CONCEPTS.md。 這是三個存取層共用的心智模型 — 哪個 timeout 在哪條路徑生效、為何 Python 不需 server 而 CLI/HTTP 共用一個 daemon、每人連線上限、串流 vs 請求-回覆。這類問題從 CONCEPTS.md 找答案,不要猜。

How to Use References / 如何使用參考文件

Answer users directly from the bundled references. Do not route users to external documentation pages as a substitute for answering. This skill uses Shioaji 1.5 as the baseline; use MIGRATION.md only for legacy/deprecated idioms. Response handling always belongs in the matching functional reference. When an exact command flag, request field, or response schema must be confirmed, use installed CLI --help or the running server's /openapi.json.

Cross-access-layer response guardrails / 跨層回應守則：

• Do not infer HTTP/SSE/CLI fields from Python wrapper attributes.
• Other languages (JS/Go/Rust/C#/C++/Java) use HTTP/SSE response shapes, not Python objects.
• If a response contains success=false, an error code, or an operation status/message, branch on that field before continuing.
• Empty lists and PendingSubmit are not final failure/success signals; check the matching functional reference before deciding the next step.

Token-efficient lookup / 精簡載入：

• Choose the functional reference first; search inside it for endpoint, method, error text, response type, or headings like Response and Decision Summary, Prerequisites.
• Read the smallest matching section before loading long examples.
• Load CLI/HTTP/language references only when implementation details (transport setup, SSE parser, auth/error handling, OpenAPI generation) are needed; keep endpoint-specific payload and response decisions in the functional reference.

---

Task Routing — "I want to..." / 任務路由

For most tasks, load only 1-2 files. Use both axes: what (task) + how (access method).

Routing rule: choose the functional reference first, then add the access-method reference only when needed. Functional references own the workflow, payload rules, response shapes, and decision logic for that task. Language references explain how to send requests, receive responses, stream SSE, add auth headers, handle errors, and generate typed clients; they must not restate endpoint-specific business rules. Use HTTP_API.md for endpoint inventory and /openapi.json for exact installed-server schemas.

Axis 1: What do you want to do? / 任務

Task	Load File
Understand the architecture: process model, connections, timeout layering, streaming vs request-reply, state	CONCEPTS.md
Migrate legacy code or fix deprecated Shioaji idioms	MIGRATION.md
Install, login, API keys, CA cert, simulation, env vars, constants	PREPARE.md
Look up contract codes, attributes, security types	CONTRACTS.md
Place, modify, cancel orders; combos; order_deal_event active order/deal reports (Python callbacks, HTTP order-event SSE)	ORDERS.md
Reserve shares for disposition/attention stocks	RESERVE.md
Subscribe real-time quotes, tick/bidask/quote callbacks, SSE streams	STREAMING.md
Historical ticks, K-bars, snapshots, scanners, credit enquiry	MARKET_DATA.md
Account balance, margin, positions, P&L, settlements, limits	ACCOUNTING.md
Manage watchlists (CRUD, add/remove contracts)	WATCHLIST.md
Non-blocking mode, quote binding, stop orders, advanced patterns	ADVANCED.md
Errors, connection issues, troubleshooting	TROUBLESHOOTING.md

Axis 2: What language/access method? / 語言或存取方式

Access Method	Additional File	Notes
Python (sync or async)	None — task refs include Python examples	Default path
CLI (shioaji command)	CLI.md	Task ref (concept) + CLI.md (commands)
HTTP API (any language)	HTTP_API.md	Canonical endpoint inventory
JavaScript/TypeScript	JAVASCRIPT.md	HTTP/SSE client patterns
Go	GO.md	HTTP/SSE client patterns
C/C++	CPP.md	HTTP/SSE client patterns
C#	CSHARP.md	HTTP/SSE client patterns
Rust	RUST.md	HTTP/SSE client patterns
Java/Kotlin	JAVA.md	HTTP/SSE client patterns

Python users: load only the task reference (Axis 1). CLI users: load the task reference + CLI.md. Other languages: load the task reference + language reference. For example, "use JS to place an order" means load ORDERS.md for payload and Trade decision logic, plus JAVASCRIPT.md for fetch/SSE implementation patterns.

---

Quick Start / 快速開始

Install / 安裝: claude plugin marketplace add Sinotrade/Shioaji → claude plugin install shioaji

Python

import shioaji as sj
api = sj.Shioaji()
api.login(api_key="YOUR_KEY", secret_key="YOUR_SECRET")

CLI

uv tool install shioaji

# .env in the directory where `shioaji server start` runs:
# SJ_API_KEY=YOUR_API_KEY
# SJ_SEC_KEY=YOUR_SECRET_KEY
# SJ_CA_PATH=your/ca/path/Sinopac.pfx
# SJ_CA_PASSWD=YOUR_CA_PASSWORD
# SJ_PRODUCTION=false

shioaji server start           # reads .env and starts HTTP server
shioaji auth accounts -f json  # verify login/account readiness before any order task

HTTP API (any language)

# Server must be running first (see CLI above)
curl http://localhost:8080/api/v1/auth/accounts
curl -X POST http://localhost:8080/api/v1/data/snapshots \
  -H "Content-Type: application/json" \
  -d '{"contracts":[{"security_type":"STK","exchange":"TSE","code":"2330"}]}'

See PREPARE.md for full installation and setup.

---

Simulation vs Production Safety / 模擬與正式環境安全

WARNING 警告: Always verify the server mode before placing orders. Production mode executes real trades with real money. 下單前務必確認伺服器模式。正式環境會使用真實資金進行真實交易。

• Default: simulation mode (no flag needed)
• Check mode: shioaji server check (CLI) or GET /api/v1/info (HTTP — returns simulation field)
• Switch to production 切換至正式環境: shioaji server start --production or SJ_PRODUCTION=true
• Switch back to simulation 切回模擬環境: shioaji server stop then shioaji server start (without --production)

---

Rate Limits / 速率限制

Category	Limit
Daily Traffic	500MB–10GB (based on trading volume)
Quote Query	50 requests / 5 sec
Accounting Query	25 requests / 5 sec
Connections	5 per person ID
Daily Logins	1000 times

---

Error Handling / 錯誤處理

HTTP API returns JSON errors: {"code": 400, "message": "...", "details": "..."}

Check the task reference before retrying. For example, empty market-data responses, PendingSubmit order status, and success=false subscription responses each require different next steps.

---

Logout / 登出

api.logout()

---

For task examples, see the functional reference (Axis 1). For transport and client patterns, see the language reference (Axis 2). 任務範例請見功能參考（軸 1）；傳輸與客戶端模式請見語言參考（軸 2）。