Guotai Haitong Lingxi Smart Stock Selection

Problem

A-share stock screening often requires manually comparing market data, financials, valuation metrics, and technical indicators, then validating whether the same rule set would have performed well historically. Doing this with scattered queries and spreadsheets makes it easy to miss sorting logic, data conditions, and date semantics. This skill packages both “screen stocks by conditions” and “backtest those conditions” into Agent-callable MCP tools, so engineers and research users can describe rules in natural language and receive structured results.

How It Works

The core capabilities are:
- Multi-indicator screening: financial-search supports natural-language queries for A-share real-time market data, company information, F10 financial data, and technical indicators.
- Strategy backtesting: backtest evaluates a screening rule over historical data and reports annualized return, maximum drawdown, and win rate.

Key execution steps:
1. Check authorization: read gtht-skill-shared/gtht-entry.json; if the file is missing or expired, trigger the cloud authorization flow.
2. Normalize screening intent: convert natural-language rules such as “gain over 5% and turnover over 3%” into the required query parameter.
3. Call the matching gateway: use financial for data search and backtest for historical validation.
4. Surface important context: preserve the data conditions, date field, and sorting logic in the response, so yesterday’s data is not presented as today’s data.

Boundaries

• It is primarily scoped to A-share base market data and the provided backtesting flow, not information outside the financial and backtest tools.
• Vague requests should be clarified for ticker, condition, or time range before tool calls.
• Execution is restricted to the skill directory and shared configuration directory, with cross-platform commands run through node; Windows PowerShell environments should use ; instead of &&.