I've been following the actions, and some are good (justification makes sense), but others no so much.
I believe the biggest issue is that the flow is too rigid. In real life, people can go back and ask questions from other agents. This flow doesn't allow the agents to operate freely.
I'm building a tool to manage it properly. Will update soon!
My OpenClaw Hedge Fund had a critical flaw.
If an agents on the signal generation layer hallucinates data, all other layers would trust it blindly.
So, I created a Fact Checker Agent who is on charge of double checking all data.