@xim isn’t saying it belongs in the CommonMark spec. She/he is saying that it belongs somewhere, perhaps as a recommendation for parser implementers. Perhaps it makes sense for CommonMark.org to host a Best Practices for Renderer Implementations document.
Some things, such as automatically generated IDs, might belong in the spec but if so should be accompanied with a note about implications for renderers. These could all be collected in the best practices doc. @jgm and @codinghorror: something to consider?
Of course the reference implementations should follow those best practices.