LLM trigger lists of values similar to intents logic

Hi,
For LLM scripts, like triggering Music Assistant and others, I am facing the issue of fancy device names or media etc, for which I would need to translate the inputs from “in” to “out” values, similarly like for intents. Something like the below:

        radio_trigger:
          name: Radio trigger
          description: The trigger sentences to request a specific radio station.
          selector:
            text:
              multiline: false
              multiple: true
          default:
          - (play|listen to) [the ]((radio station)|(radio)|(station)) {media_name}
            [(in|on|using) [the ]{area_or_player_name}]
          lists:
            media_name:
              values:
                - in: "radio (fool|full|fall)"
                  out: "radio foul"

A very common case is creative radio names that LLM will have no chance to guess. I want to avoid solution based on prompt tweaking also.
Any tip?
(for media, sicne entities exist, I may try with entity aliases)
edit: I realized that I could simply create manual url for radios and named them by what stt piper server will predict most of the times… not nice but okayish