CORRECT SKILL ID: amzn1.ask.skill.5e264152-8c70-4da7-838e-08f25729f290
ZOMBIE SKILL ID: amzn1.ask.skill.7556c6b8-5a2f-44c1-a125-a1a6aa876e49
I have a skill that suddenly started failing with INVALID_RESPONSE errors without contacting my server at all, yet somehow I got the end-session request. Other times the skill worked. It's as if the Alexa servers are switching between new and stale copies of some routing table (or in DynamoDB lingo, some kind of broken "eventually consistent" update operation). Today I discovered that when I looked that request JSON when using the simulator, during one of the many INVALID_RESPONSE occurrences, the JSON shows the wrong app ID, one that belongs to a completely different skill.
I've tried to switching Test between off and Development multiple, times, re-logging in and out, a full rebuild of the skill model, changing the invocation name, etc. In a few minutes, out of desperation, I'm going to have to completely delete and recreate the skill from scratch. This is a painful process especially since I have in-skill purchases defined with the skill and worse, I have a contest deadline bearing down on me (Alexa ISP challenge). Amazon, please create a true, comprehensive fix for this soon. Better yet, give us an editor to alter these values ourselves. I tried doing a `ask clone` but the downloaded `skill.json` file did not show any fields with the suitable values I could edit to fix this.
UPDATE: I tried completely deleting the ASK trigger from the target lambda and re-adding it. Did not help.
UPDATE: I had to completely delete the skill and the Lambda function that serviced it. I have created a new skill from scratch and a new Lambda function. It works now as expected. I hope though, that no one thinks this is a satisfactory solution since it is a drastic and time consuming one.
It is noteworthy that all these problems started right after I had gone through one of the Alexa cookbook demos (a wonderful resource). Since I was not adept at AWS permissions, I had several iterations of failed deployments to the Cloudformation platform, complete with several iterations of interstitial "stuck" stack deletion events. At one point, I even had the demo skill "leak" into the skill I just deleted! When I went to do a simulator test, I got a prompt from the successfully deployed demo skill instead of the skill the simulator had loaded.
UPDATE: Broken again! (See next post)