Built an AI agent that monitors options flow — shares 9-wave morning briefing architecture

[tellmefrankie]

For agent builders — sharing a production financial agent that's run autonomously every morning for 6 months.

Core pattern: 9 sequential agents, each with a single responsibility

Rather than one monolithic 'analyze my portfolio' prompt, I split it into 9 waves that pass structured output to each other:

Wave 1: Macro context (futures, VIX, FX)
Wave 2: Options flow scan (sector ETFs, lottery-call filtered)
Wave 3: News sentiment (50+ RSS → Claude scoring)
Wave 4: Technical setup
Wave 5: Portfolio Greeks
Wave 6: Opportunity scan
Wave 7: Alert triggers
Wave 8: Trade ideas (structured: entry/stop/target)
Wave 9: Execution checklist

Each wave outputs a fixed schema. The next wave reads it as context. This threading is what makes the system useful — macro context from Wave 1 changes how Wave 2 interprets an anomaly.

Real catch: Wave 2 flagged XLI put/call at 5.32 (5.96x above its own 30-session baseline) while SPY was 0.44. Sector divergence signal. Wave 3 confirmed no fundamental catalyst. Wave 7 triggered MONITOR. Wave 9 said don't trade.

I didn't trade. The system worked.

Free agents: https://github.com/tellmefrankie/ai-investment-skills