1 · Wealth advisor
Opens the site, types a ticker like F or AAPL, hits send.
2 · Coordinator Agent
Reads the request. Decides what to do. Fires both sub-agents in parallel. When they return, synthesizes a single human-readable briefing. Powered by Claude Sonnet 4.6.
3a · Official Sub-Agent
Strict primary-source analyst. Only reports facts it can cite to a filed document.
4 · Coordinator writes the briefing
Combines both reports into one markdown briefing with clearly separated sections: live snapshot, business overview, top risks (cited), social signal (in its own labeled section with an unverified disclaimer), and any divergences where social claims contradict the filings.
5 · Advisor reads the briefing
Renders in the browser as the agent thinks. Tables for numbers. Bullets for risks. Tool trace shows the work. ~60 seconds end-to-end.
Why two agents, not one?
Trusting the SEC and trusting Reddit are different jobs. One prompt can't encode both well. Each sub-agent has its own skepticism, its own tools, its own output schema.
Why is social labeled "unverified"?
Forum posts include bots, pumps, and rumors. They are useful situational signal, not facts. They live in their own labeled section with a disclaimer, never inlined with the cited filings.
What about buy / sell calls?
The agent declines. That's investment advice, which triggers SEC and FINRA rules. The tool surfaces facts and risks so the advisor can form their own view.
What if a data source is down?
The briefing continues with what it has. If Yahoo's quote API fails, you get an SEC-only brief with a "live quote unavailable" note. No silent failures.