Yeah exactly, I wanted something that I could see what the agent was doing. Open Claw was cool but I wanted the controllability and context management that Claude code affords.