Six agents split the trading day between judgment, rules, and one human checkpoint. This is the roster — who each one is, the question it answers, and when it runs. The call-and-handoff diagram (who invokes whom) is next.
“Where is option premium relatively attractive?”
Ranks all 8 underlyings by IV skew each morning; splits the top 3 (sell 15Δ puts) from the bottom 5 (sent to the Analyst).
“What direction does the market appear to favor?”
Reads first-10-min price action, gap %, and 24h news per directional candidate. Returns Direction / Confidence / Reason — never places a trade.
“Can we afford to take this trade?”
Final approve/reject gate before any order — checks buying power, exposure, and daily loss limits. Can override the Analyst.
“Execute and manage the predefined rules.”
Polls open positions roughly every minute; enforces exit rules — 50% TP / 3× SL / EOD for short puts, hard 2:30 ET close for longs.
“What did we learn, and what should we test next?”
Mines the day's trade logs for what to test next — post-mortem, inputs audit, live EOD re-check, then proposals. V1 (n=1 day): process-fix proposals only, no numeric tuning yet.
“Is this proposed change worth testing?”
Approves or rejects the Strategist's proposed change, interactively in a Claude Code session. Reject → discarded, logged. Approve → implemented directly for the next run (V1 skips the formal Backtest/Paper/Risk-Approval pipeline).