dsh-trading-toolkit

Doing A-share or US-stock strategy research inside a DeepSeek Harness agent used to mean manually wiring up market-data APIs, juggling API keys, coding ADX calculations and backtest loops from scratch—all with the ever-present risk of accidentally placing a live order. dsh-trading-toolkit compresses that entire pipeline into four read-only tools: market_quote auto-detects six-digit A-share codes, sh/sz prefixes, Chinese stock names, US tickers, and major indices, pulling data from EastMoney's domestic endpoints with no key required; kline_history fetches OHLCV bars across eight timeframes from 1-minute to monthly with optional forward, backward, or no adjustment; regime_signal applies the same ADX thresholds used in production (≥25 trend, 15–25 oscillating, <15 noise) to classify market state, including ±DI direction and 200-EMA bias; and backtest_run executes a simple cross-state long/short backtest reporting total return, max drawdown, win rate, and trade count.
Every tool is strictly read-only—no order placement, no credential access, no credential storage. The backtest module is explicitly labeled as an educational preview rather than a production-grade system. The plugin opens up the core signal logic of the author's live PrinciplesV2 strategy stack in a lightweight, safe form, making it a natural fit for quant researchers, strategy hobbyists, and anyone who wants to go "fetch quotes, classify the regime, preview a trade" in a single agent conversation without ever risking real capital.