Ok so this is a very bizarre problem. I have an audio streaming skill. It's on v1 of the node SDK. When the skill is streaming / has control of the audio player, if I say "Alexa ..." it immediately sends a blank intent to my skill, even before I can finish issuing the command.
The experience is that I'll still be saying the command, but the skill is already responding with its "Unhandled" handler that the V1 node SDK falls back to. Even before I finish saying the invocation name - meaning it will interfere with literally any other skill a user tries to invoke. It wasn't doing this before and I'm not sure when or why this started happening, but the skill is in production.
This doesn't happen at all if the skill doesn't have control of the audio player (CORRECTION: it doesn't happen if the skill isn't currently playing audio, even if it does have control of the player).
I.e., I can say "Alexa, ask [my skill] to [do utterance thing]" and it works just fine if it's not playing audio. This problem only happens when it's streaming audio, which is a significant problem given that that's the central purpose of the skill.