AI Zhangle Conditional Stock Selection

Problem

When writing conditional stock-screening logic, users often mix natural-language requests like “baijiu sector, P/E below 20, MACD bottom divergence, gain above 5%, earnings surprise.” Hitting the backend directly can miss OR/AND relationships, relative dates, implied sorting, numeric units, exclusions, and “continuous vs cumulative” semantics. This skill turns a natural-language screening intent into one executable query and returns a stable Markdown candidate list.

How It Works

• Triggering: sector/theme, financial metrics, technical metrics, market data, earnings surprises, combined conditions, and subjective requests such as “recommend a few stocks.”
• First query: pass the user’s original sentence as query to selectStock, without pre-rewriting, to preserve intent.
• Rewrite retry: if the first call returns no result, disambiguate. Multiple values in the same dimension are usually “or”; cross-dimension conditions such as sector, P/E, and gain are usually “and.” Add relative dates, units, sort fields, top-N limits, and exclusion rules.
• Output constraints: return data.result in Markdown without changing its format, and show at most 10 instruments.
• Error handling: distinguish network transient issues from business restrictions by error.category.

Boundaries

Use it for “screening a list of instruments that meet criteria,” not individual stock diagnosis, valuation analysis, portfolio diagnosis, or buy/sell advice. It depends on HT_APIKEY, with optional SELECT_STOCK_SERVICE_URL and SELECT_STOCK_TIMEOUT_SEC. It is designed to execute directly without repeatedly asking follow-up questions, so ambiguous requests are inferred and completed rather than manually confirmed item by item.