Support Human-in-the-loop Requests from Remote Agents

[AK2000]

Describe the Request

We need a mechanism for an agent to ask for approvals or insert human-in-the-loop steps. Currently this could be accomplished by a locally deployed agent that asks for input, but is this the best pattern? Could this be better documented if this is how we want to do things, or maybe made easier with helper-methods?

Sample Code

[benclifford]

"user facing agent" is probably the right pattern here. Otherwise, you introduce something different for remote actions that should only be different if it really is different.

There's no "local" though - it's got to be somewhere where it can interact with the user, whatever that means for the application.