After discussions today we're starting with the following approach.

• User is responsible for creating a markdown version of their knowledge doc(s). We can support this with documentation on how to use things like pandoc.
• User can experiment with knowledge in the cli (gen, train, serve) with support Kai and company have been building
• The PR submission for knowledge will contain a qna.yaml (just like a skill submission) with some differences. In addition to the PR, the user will also create a new git repo, and place the markdown file(s) in the new repo. The qna.yaml will contain a pointer to markdown file used

Backend

• The triager can do a "quick" visual check of the markdown (and associated optional public link)
• If accepted, the markdown will have to be pulled from the user repo to the right COS(?) storage