capital-generation

Retail investors doing securities research face a fragmented data landscape: quotes in one platform, disclosures in another, news scattered across the open web. Pulling, cleaning, and charting all of it by hand is tedious and error-prone. Capital Generation wraps this entire pipeline into DSHs Capital agent mode, so you can simply describe the metric or ticker in natural language and the plugin handles fetching, organizing, and rendering behind the scenes.

Under the hood the plugin runs a multi-subagent architecture. A data collector sub-agent calls roughly 61 structured-data endpoints from Tonghuashun that cover A-share quotes, financial line items, valuation and bidding metrics, indices, and fund data. A data junior sub-agent performs cleaning, descriptive statistics, and pivot summaries to surface the story behind the numbers. A web retriever fetches unstructured content such as corporate announcements and regulatory disclosures via AnySearch and Wind Alice. A visualization specialist generates fully self-contained HTML charts in a one-shot call, powered by Lightweight Charts with zero external requests, so the output works offline. The main agent never touches raw structured rows directly; every slice is extracted on demand by a specialist and returned as a concise message, which keeps the context window lean and reduces hallucination risk. The whole plugin builds on DSHs official infrastructure for subagent orchestration, persona injection, skill loading, context compaction, and credential management rather than reimplementing any of it.

The plugin is aimed at Chinese A-share or fund retail investors who track daily quotes, drill into income statements and balance sheets, and want candlestick or volume-price charts without leaving their DSH session. The project is in an exploratory phase, provides no financial advisory services, makes no return guarantees, and all investment decision risk rests solely with the user.