Maybe the transfer mechanism of google-adk subagents is more reasonable.

Under "swam", an agent will receive all messages, but how does the agent know which of these messages are its own and which belong to other agents? Could there be a situation where the agent thinks all messages are its own, imagines itself to be a superman, and then has hallucinations?