Agent Architecture Note Agent 1 Classification Role: Hybrid Reasoning: Agent 1 takes price data and strategy signals and coordinates an external process (NinjaTrader) to place orders, then tracks the execution result — that's orchestration behavior. But it isn't making high-level strategic decisions itself; it's relaying and monitoring a fixed pipeline (signal → order → result), which is worker-like. Recommended model: claude-sonnet-4-6 — high-stakes (real capital), moderate reasoning needed to correctly track order execution logic and avoid duplicate/missed orders. Agent 2 Classification Role: Worker Reasoning: Agent 2 (NinjaTrader) receives a trade request and executes it — no coordination of other processes, no strategic judgment, fixed input→action→output loop. Recommended model: claude-sonnet-4-6 — mechanical execution, but high stakes (a bad fill or execution error costs real money), so reliability matters more than reasoning depth. How They Relate Agent 1 and Agent 2 operate concurrently rather than in a blocking sequence: Agent 1 continuously watches price/strategy signals and dispatches trade requests, while Agent 2 (NinjaTrader) executes whatever request it's given and reports the result back. Agent 1 acts as the coordination layer over the execution layer — it decides when and what to send, Agent 2 handles how the order actually gets filled. Together they split the system into a decision/coordination stage and an execution stage, letting execution mechanics run independently of upstream signal analysis.