Preserve context in the lead chat - let the subagents fill up their own contexts then only return the necessary information.
Why not just have an agent that can branch its context?
Why not just have an agent that can branch its context?