Most people want an agent that says "buy this." Wrong job.
A monitoring agent is not a stock picker. It is a watcher. Its only task is to notice when reality moves against the thesis you already wrote down, and to hand you the receipt so you can judge it.
Here is the disciplined build. The agent watches, you decide:
-> Write the thesis first: 3 to 5 assumptions you are actually betting on (margins hold, this segment grows, debt is manageable). -> Define the triggers: which events could break each assumption (a new 10-Q, an earnings call, a guidance change, a downgrade). -> The agent monitors the sources, not the ticker price. Price is noise. Filings and disclosures are evidence. -> Every alert carries a source trail: document, page, and date, or it does not get sent. -> Each alert classifies as confirms thesis, threatens thesis, or unclear. "Unclear" is a real answer and goes on your follow-up list.
What the agent is NOT doing: deciding if the change is good, what it is worth, or whether you act. That is your assumption set, not its output.
The common mistake is building an oracle that spits out verdicts. The failure mode is you stop reading and start obeying. The discipline is the opposite: the agent compresses the watching, you keep the judgment.
Human judges. AI builds the watcher.
Save this and build it against your own holdings.
Educational content only. Not investment advice, and not a recommendation to buy, sell, or hold any security. Wall Street Prompt. Always verify against the primary source.