capital-generation

Retail investors conducting securities research face a fragmented data landscape: quotes in Tonghuashun, disclosures in Wind, news scattered across the web. Pulling, cleaning, and charting this data manually is tedious and error-prone. Capital Generation wraps this entire pipeline into DSH's Capital mode, turning conversation into a working financial research workflow—describe the metric or the ticker, and the plugin handles fetching, organizing, and rendering behind the scenes.

The plugin uses a multi-subagent architecture: data_collector taps roughly 61 structured-data APIs from Tonghuashun covering A-share quotes, financial line items, valuation and bidding, index, and fund data; data_junior performs cleaning, descriptive statistics, and pivot summaries; web_retriever pulls unstructured content such as announcements and disclosures via AnySearch and Wind Alice; and visualization_specialist generates self-contained HTML charts in a one-shot call. The main agent never touches raw structured data directly—every slice is extracted on demand by a sub-agent and returned as a concise message, protecting the context window and reducing hallucination risk. The entire plugin builds on DSH's official infrastructure for subagent orchestration, persona injection, skill loading, context compaction, and credential management.

This plugin is aimed at Chinese A-share or fund retail investors who track daily quotes, drill into financial statements, and draw candlestick or volume-price charts, and who already use DSH for day-to-day research. The project is in an exploratory phase, offers no financial services or return guarantees, and all investment risk is solely the user's responsibility.