Does ExpectedInputs / possible_intents only work with "assistant.intent.action.TEXT"?

Question

I am currently trying to write a demo for action on Google using the leisure web service.

At the moment, the user opens the action ("talk to testaction") and receives a welcome message (through the main intention). This initial intention awaits the user's response, and also sets the next expected intention through the possible_intents field in the JSON response

According to the documentation, I should be able to specify the user intention in the possible_interests of my HTTP JSON response.

However, if I use any intention other than "assistant.intent.action.TEXT" as soon as I respond to the initial intention / invitation, I get the following error:

I'm sorry, I did not understand.

And the answer to the initial welcome intention is incorrectly sent to my service.

This does not work:

{
    "response": "...",
    "expectUserResponse": true,
    "conversationToken": "...",
    "audioResponse": "...",
    "debugInfo": {
        "agentToAssistantDebug": {
            "agentToAssistantJson": {
                "conversation_token": "...",
                "expect_user_response": true,
                "expected_inputs": [
                    {
                        "input_prompt": {
                            [...]
                        },
                        "possible_intents": [
                            {
                                "intent": "testintent"
                            }
                        ]
                    }
                ]
            }
        }
    }
}

It works:

{
    "response": "...",
    "expectUserResponse": true,
    "conversationToken": "...",
    "audioResponse": "...",
    "debugInfo": {
        "agentToAssistantDebug": {
            "agentToAssistantJson": {
                "conversation_token": "...",
                "expect_user_response": true,
                "expected_inputs": [
                    {
                        "input_prompt": {
                            [...]
                        },
                        "possible_intents": [
                            {
                                "intent": "assistant.intent.action.TEXT"
                            }
                        ]
                    }
                ]
            }
        }
    }
}

my testintent is correctly defined in the action package and works fine if I call it directly.

Is it possible to use only general TEXT intentions, and then I need to perform all text recognition and intent recognition in the code?

Answer 1

When using the Actions SDK, only the TEXT intent is supported. You must use your own NLU to analyze the raw text input provided by the user.

NLU, API.AI.