@zwol: The reason for the # in there is that it also supports classes, by preceding the word with a .. It’s CSS syntax, basically.
My own experience with auto-genning IDs suggests that it’s often not a great idea. If you try to gen them from text, you get some terrible anchors, and they’re not stable against wording changes; if you gen them from a counter or something similar, they’re not stable against reordering or adding/deleting sections above them. You can make it slightly better if you pay attention to the outline generated by the headings, as then sub-headings before you won’t matter, but still.
In my spec-generation tool that accepts some Markdown, I auto-gen IDs from text but log a warning that the author should be providing an explicit ID.
If we do end up supporting autogenned IDs on headings, however, I do suggest it be based on the outline level of the heading, like “heading-2-1” for the first h2 after the second h1. As I said above, it makes it slightly more robust against document changes; only ancestors and preceding siblings of ancestors (in the outline tree) can affect your ID, rather than all preceding headings like what you get if you just number them sequentially.