Issue with ApprovalRequiredAIFunction in sub-agent

[hobbsax1]

I am using ApprovalRequiredAIFunction to try and perform human in the loop actions.

I have the following setup:

TriageAgent
UserRoleAgent

TriageAgent using UserRoleAgent as a tool.

UserRoleAgent has several tools, one of which is AddUserToRole that is encapsulated in ApprovalRequiredAIFunction

When I try to get the approval, the TriageAgent seems to be eating the FunctionApprovalRequestContent content coming back. I have middleware on both agents and can see where the UserRoleAgent has the last message as a FunctionApprovalRequestContent with the proper argument calls, but I cannot seem to bubble that up so that the triage agent returns it to the caller.

We are using a custom ASP.NET Core with restful endpoints so we have an endpoint /agent that takes (approved takes an optional approved status if the user was asked to approve a tool call):

public class Request 
{
    public string Message { get; set; }

    public bool? Approved { get; set; }
}

And then responds with either the messages from chat or info needed for the approval (the message is generated from the function call arguments and name in the backend):

public class Response
{
    public string? Message { get; set; }

    public bool ApprovalRequired { get; set; }

    public string? ApprovalMessage { get; set; }
}

Anyone know of a way to bubble up approval requests from sub agents that are acting as tools themselves?

[moonbox3]

Adding @markwallace-microsoft to help route appropriately.