We have a use case to put Alexa devices into many rooms and write custom Alexa skills to serve different responses and perform database updates based on which room they are in. As a result, we need to uniquely identify each Alexa in order to identify the room. We have look into various work around:
1) use "context:System:device:deviceId" which is specific to each skills and changes when the skills disable/enable.
2) use different userid to identify different Alexa but this would mean having many user accounts if the deployment is big
3) when Skill detects unknown "context:System:device:deviceId", expect user to use a certain utterance to provide the room information and register the new id in database.

However, none of these are scalable nor reliable so a real unique device ID for each Alexa would be desirable. Thanks!

Allow users to link their social media accounts (Facebook, Twitter, Instagram, etc) in the Alexa App itself. Then allow skills to query which accounts are linked and request permission to use those links, without requiring each skill to perform account linking separately (and perhaps not even exposing the account details to the skill, for greater privacy!).

This would allow a skill to request permission to post to Facebook, for example, to share their new high score. The skill would request this permission at the time of the interaction, and the approval would be delegated to Amazon's built-in handling (to make sure the skill did not deceive the user into granting permission). Once permitted, the skill would be able to use an Amazon API that would in turn call the social media API.

This would abstract this common interaction away from each skill, provide greater consistency, and encourage organic growth of skills.

I was surprised to find that there is no official Docker image for development with the ASK CLI. The "amazon" user on Docker Hub contains only a couple images for working specifically with ECS and Elastic Beanstalk, and there does not appear to be any official "alexa" user. Upon further investigation, I found the aws-codebuild-docker-images repo in the "aws" GitHub organization, but unfortunately it does not have any sample Dockerfiles that install the ASK CLI either. There are similar issues open at the AWS CLI repo, because there is no official Amazon Docker image for the AWS CLI either, so I imagine that whatever aws-cli image gets created could serve as the base for an ask-cli image.

TL; DR
I believe there should be an official ask-cli Docker image maintained by Amazon and pushed to Docker Hub/Store. Ideally, it should support all the major modern Linux distros, including Alpine Linux. One should be able to run the following command, and have everything just work:

docker run -it alexa/ask-cli

"STOP" and "PAUSE" have two very distinct meanings:

"STOP": cause (an action, process, or event) to come to an end.
"PAUSE": a temporary stop in action or speech.
Different words, different meanings.

Unfortunately, AudioPlayer confuses the two, and even the documentation warns that

"when your skill is not in an active session but is playing audio, or was the skill most recently playing audio, utterances such as 'stop' send your skill an AMAZON.PauseIntent instead of an AMAZON.StopIntent."

This creates a poor user experience, and it doesn't take long to find at least 30 questions, requests relating to this.

For example, when a user is playing audio and they want to stop, they mean stop. As in, cease playing. However, when they "pause", they usually want to do something. In which case, the skill could wait for a command. Then the user could say "how long is this track" or "skip forwards 30 seconds".

This would at least provide a partial fix to the request below, although implement the idea below would provide a far better solution in all cases.

https://alexa.uservoice.com/forums/906892-alexa-skills-developer-voice-and-vote/suggestions/33406042-remove-requirement-for-invocation-name-when-an-aud

Right now, in the test tab for a given skill in the developer console, there's an option to provide some SSML text and hear how Alexa would convert that to speech. This utility uses no contextual information about the skill itself, and there's really no reason at all that it needs to be stuck behind a login gate, other than for the purpose of possibly thwarting abuse of said functionality.

As a developer, I would like the ability to provide this same functionality to my users on a public website I run or to my collaborators in a private tool of my own.

Preferably this would exist in the form of an embeddable widget that I could just drop into my site, but an API that would let me do the same thing would be acceptable.

Some sort of rate-limiting mechanism to prevent abuse of the API would be acceptable (and even expected). The ability to choose Alexa's voice by locale (or even a potential integration with Polly voices) would be a nice bonus feature.

To provide a specific use case - I'm looking at building a skill that can consume a small amount of user-generated content, specifically with regard to online handles. I'd like users to be able to test how Alexa will pronounce their IPA-written names before saving and submitting on a website.