TradingAgents A-Share Multi-Agent Investment Research Framework

Problem

When a single model analyzes an A-share ticker in one pass, it often compresses price action, sentiment, fundamentals, policy, and capital-flow evidence into a single narrative. That makes it hard to cross-check conflicting signals or explicitly apply trading constraints such as T+1, price-limit bands, and minimum order size. This skill targets a reproducible multi-agent research workflow: given a ticker and a date, it produces a Markdown report with bull/bear debate, risk debate, and a trade plan.

How It Works

The skill adapts the six-layer TradingAgents-Astock pipeline for WorkBuddy using Team/Agent orchestration:
- Parallel analysts: seven agents cover market, social, news, fundamentals, policy, hot-money, and lockup views, using westock-data or wb-finance-skill for candles, announcements, financials, dragon-tiger lists, lockups, and pledges.
- Dual LLM tiers: fast models handle parallel analysis and debate, while reasoning models synthesize research and final decisions.
- Adversarial debate: Bull and Bear researchers challenge each other, then Aggressive, Conservative, and Neutral risk debaters weigh execution risk and downside.
- A-share constraints: the Trader stage explicitly considers T+1, 10%/20%/5% price limits, 100-share lot size, trading hours, and ST flags.

Boundaries

The output is research reference, not investment advice. Reports can be affected by stale data, inconsistent definitions, or reasoning gaps. It is better used to test A-share research workflows and generate institution-style reports, not as direct position commands. Confirm the ticker, analysis date, and availability of WorkBuddy finance tools before use.