GMQuant Natural Language Strategy Engine

Problem

When a trading idea is still expressed in plain language, turning it into runnable GMQuant code still requires handling set_token, symbol, run(), strategy_id, subscription, and order placement. This skill targets that conversion step: it structures Chinese strategy descriptions into runnable .py files and supports backtesting or live/simulation execution. It is not a generic coding assistant; it focuses on the GMQuant SDK strategy generation and execution workflow.

How It Works

The skill first identifies the key strategy variables: instrument pool, frequency, entry signal, exit signal, position sizing, stop loss/take profit, backtest period, initial capital, and execution mode. If the user has not provided a strategy_id, it asks before generation, because this identifier determines whether backtest results are persisted in the GMQuant terminal and shown in performance charts. During generation, it produces a complete strategy file. During execution, it launches scripts/run_strategy.py to run MODE_BACKTEST=2 backtests or MODE_LIVE=1 live/simulation mode. For data-query scenarios, it emphasizes calling set_token first and notes that run() is not required. For trading scenarios, it pays attention to symbol case sensitivity, subscribe, order-function parameters, the 100-share minimum for A-shares, and T+1 constraints.

Boundaries

The skill depends on the GMQuant terminal environment, and some interfaces or premium data may require paid permissions. Behavior of current(), order status, and execution reports can differ across backtest, simulation, and live mode; market orders with insufficient funds may be canceled rather than raising an error. It is suitable for converting explicit strategy intent into runnable code, but it should not invent entry or exit rules when the user’s description is vague, and it does not guarantee trading profits.