I’ve noticed that in both LangGraph v0.6 and v1-alpha, when invoking subgraphs with different state schemas, it’s necessary to define an adapter node in the parent graph. However, this approach negatively affects the visualization, since the parent graph only sees the adapter node. I wonder if it would be possible to add parameters such as InputSchema and adapter_hook to the create_agent method to handle this more seamlessly.
Fine, this could be solved by passing private states between nodes.