[digiltd]

 When you say "Alexa, tell Kodi to play Supernatural" that might be because she hears "super natural" (check your logs or the app to see what she is sending).

 The problem is that that is a short phrase, with more than half of it being a slot she has to figure out.  Alexa matches on a confidence threshold, and if no utterance meets the minimum threshold, she will fall through to the last intent in the model.  According to Amazon, this is by design, even though it's shitty design.

In the model, it just happens that the fall-through intent is the one to play recently added songs.  If I had to guess, sahara doesn't have any songs in his/her library, which is why it's producing an error instead. 

Not sure if correcting me or adding to the explanation... but isn't that what I said? (albeit in a different way)

Also you are using the generic "play" command, which is always going to make things trickier for her. She doesn't know if "supernatural" it is a show, a movie, a video playlist, an audio playlist, a band, an album, a song etc. But when you say "play the next episode of the supernatural" she then knows it is a tv show because she heard to word "episode".