Multi-Agent Stock Research Roundtable System

Problem

A-share stock questions often get noisy answers: a single model perspective can be overconfident, conclusions can lack traceable evidence, and analysis can drift into trade instructions. This skill splits the question into sectors, valuation, signals, contrarian setups, fundamentals, and short-term flow, then uses independent multi-role analysis to reduce single-point hallucination while keeping decision authority with the user.

How it works

The core loop is an orchestration chain: the editor runs TeamCreate, dispatches six specialist agents, and prevents members from talking to each other. Each role has a fixed method: PE Bands, PEG, sector momentum ranking, signal layers, earnings tracking, and short-term leader / stop-loss checks. The flow first maps the user's question to research dimensions, then dispatches tasks in parallel, requires data to cite web_fetch, online-search, or another source, and finally compiles a conclusion card, expert views, deeper questions, and trigger conditions for later review. It makes the process auditable: who may answer what, what data it relies on, and when the view should be re-evaluated are all visible in the output structure.

Boundaries

It fits A-share research scenarios such as sector direction, valuation comparison, earnings tracking, and short-term signal review. It does not output buy, sell, target prices, or guaranteed-return language, and it is not a substitute for real-time data, licensed advice, or personal decision-making. Policy, undisclosed material events, and insider information remain outside the proper scope.