Is including additional information in the output object a good idea?

I'm experimenting with a Conversation where I would like to modify the output in a couple of different ways:

different output for speech or text
different output depending on the tone of the conversation

It looks like I can add extra output details which make it through to the client ok. For example, adding speech alongside text...

{
  "output": {
    "speech": {
      "Hi. Please see my website for details."
    },
    "link": "http://www.example.com",
    "text": {
      "Hi. Please see http://www.example.com for details."
    }
  }
}

For the tone, I wondered about making up a custom selection policy, unfortunately it seems to treat it the same as a random selection policy. For example...

{
  "output": {
    "text": {
      "values": [
        "Hello. Please see http://www.example.com for more details.",
        "Hi. Please see http://www.example.com for details."
      ]
    },
    "append": false,
    "selection_policy": "tone"
  }
}

I could just add a separate tone-sensitive object to output though so that's not a big problem.

Would there be any issues adding things to output in this way?

Solution

You can definitely use the output field to specify custom variables you want your client app to see with the benefit that these variables will not persist across multiple dialog rounds (which they would if you would add them to the context field).

Now currently there is no "easy" way how to define your custom selection policy (apart from the random and sequential supported by the runtime right now) - but you could still return an array of possible answers to the client app with some attribute telling the client app which selection policy to use and you would implement this policy in the client app.