When agents are interacting as plugins for another agent, the internal dialog isn't exposed. While this is fine for the initial experimental phase, it warrants further refinement.

Also consider a "max-depth" concept for how long this type of conversation is allowed to run.