I just did a quick search on Alex and WebSockets, and looks like they're actively supported by Alexa/AWS services now. I'll have to look into the limits on data chunk size limits, but I doubt a text pose is going to be greater than what it can handle. Alexa Skills Kit has come a long way in the past few years!
AWS has supported WebSockets in the AWS API Gateway service and similar stuff since... I wanna say December of last year? But I didn't remember anything about Alexa in that particular announcement. If Alexa skills have now got a way to use persistent WebSockets to maintain a connection from request to request, that definitely makes it somewhat easier; it used to be you had to use a proxy server for Alexa to do anything with WebSockets.
(Though every MU* that supports websockets uses a different protocol over that socket; the skills would have to be specific to the server family—like one for PennMUSH, etc.—or else you'd end up using a proxy backend to connect to games anyway. Even if that proxy server used websockets rather than being polled, which makes it marginally less horrible, albeit still a privacy concern for some players since that proxy server could log everything going through it: passwords in the connect command, pages to people, the @mail you read, etc.)
And yeah, I assumed the UX would be utterance based; you would definitely not want to leave the context of the MU* client skill while using it. But even maintaining context, you have that request/response UX cycle to deal with, which is the bigger issue.
And transcription seems the worst bit of all. Leaving aside the 10-second limit on audio transcription, there's the matter of syntax on the average MUSH. Pose and page and look and movement are probably easy enough, but many standard MUSH commands could be absolutely wretched for an interface like this. Imagine trying to tell it how to transcribe +bboard commands to write a post, or to dictate the sort of commands that many WoD games use in their chargen systems.
And I don't even want to think about stuff like Arx's plots or goals system, where the syntax can get particularly convoluted: something like goals/rfp <goal>,<story-beat>=<IC description of goal achivement>/<OOC note to staff about goal achievement> (...which I may even have gotten wrong, because I'm not on the game to check the helpfile right now!) is not the most friendly single-line command syntax to remember even when typing. Trying to dictate that to Alexa could end up being an infuriating experience.
That said, despite all these hassles, it's an interesting project to tackle! If you choose to go forward with it, I wish you luck; it'd be an interesting result to see, and might well be useful to folks out there!
I'm really going to have to look into the time limit and connectivity. It's worth researching now though if there's interest. Now to see if the tech can handle it, and the UX isn't a nightmare. I'm going to have to experiment soon. I have a feeling that I'd have to write some custom backend to make it work.. if it will work. That and not being able to recognize 'original' names might be a breaking problem. We'll see!