- add diagnose info in order to know token usage per agent - revisit agent models per task specialization/skills