Handing an AI agent a broker connection and a page of instructions is a genuinely new thing to be doing, and there's no canon for it yet. MomentumHunter is an open notebook of what people are actually trying — momentum, options, risk, and a growing set that parse public filings: STOCK Act disclosures from Pelosi, Crenshaw and Tuberville, 13Fs from Berkshire and Scion. Every prompt keeps full version history. Free to read and fork. Not advice, not signals — nothing here places a trade.
Hey PH 👋
Handing an AI agent a broker connection and a set of written instructions is a genuinely new thing to be doing. There's no canon for it yet — no best practices, no "here's how you write a good one." Most people I talk to don't know it's possible.
So MomentumHunter is less a product than an open notebook. People publish the actual prompts they're giving their agents — momentum, options, mean reversion, risk management, journaling — and every one keeps its version history, so you can watch someone's thinking change over twenty revisions. That history is the interesting part.
The ones people find first are usually the public-filings prompts: walking an agent through Nancy Pelosi's or Dan Crenshaw's STOCK Act disclosures, Tuberville's Senate PTRs, or Scion's and Berkshire's 13Fs. All public data, and the prompts keep the awkward parts in — the 45-day reporting lag, the amended and late filings, disclosed dollar ranges so wide they're nearly useless. They stop at "here's what was filed" instead of guessing why, which I think is the only honest way to do it.
What it isn't: a broker, an advisor, a signal service. Nothing here can place a trade, and upvotes mean people found a prompt interesting to read, not that it made money. Nobody knows what works yet — that's sort of the point.
Free to read, free to publish. If you're experimenting with this, I'd love to see your prompt in here, especially the ones that didn't work. Those are more useful than the wins.