An autonomous options agent on Alpaca. Every structure’s maximum loss is computed and set aside before the order is submitted. An order that would not be covered is never sent.
Execution runs entirely through the Alpaca CLI — the tool Alpaca describes
as designed for AI agents, scripts and automation pipelines. Every order is
submitted twice. Once with --dry-run, which renders the exact
request body without sending it. Then again, as the identical command, for real.
That is what lets the interface claim a position was covered before it existed, rather than after. Each structure carries its own receipt, and the receipt is the command.
alpaca order submit
--order-class mleg
--qty 5
--type limit
--limit-price 0.53
--client-order-id vig-3f9a1c22
--legs [
{"symbol":"SPY260903P00751000","side":"sell",…},
{"symbol":"SPY260903P00746000","side":"buy", …},
{"symbol":"SPY260903C00769000","side":"sell",…},
{"symbol":"SPY260903C00774000","side":"buy", …}
]
A structure reaches the broker only after clearing all of them. The refusals are shown alongside the fills, because an agent that only reports what it did is hiding half its judgement.
This is paper trading, and roughly two and a half sessions of it. That is not a sample size from which an edge can be inferred, and none is claimed. Selling premium wins often and loses occasionally larger; that skew is real, and the ceilings bound how much it can cost rather than pretending it away.
What is checkable is narrower. Every position is covered before it exists.