I would bet decent money that between Dublin Core, indieweb and various other XML standards of the day, something like that is already defined / describable, it's just not used because the model you suggest doesn't scale well - once comments are in the thousands (which happens, on FB or twitter), single files become unwieldy and the clients get too slow. If you steer away from single files, now you have a protocol that also defines URLs that clients and servers must agree on, increasing overall complexity. It has nothing to do with RSS, a format that has been extendable through XML schemas since 1.0 at least; it's that nobody could agree on such schemas in numbers high enough to make them a de-facto standards, because most actors had an interest in lock-in on their own platforms.
I wish you luck, but this space is hardly new. The challenge is not technical, it's entirely a political one - a lot of people have to agree and implement a standard and then reach some sort of critical mass without triggering the search for lock-in.
Today all of those are dead now in favor of ActivityPub. AP loses a lot of the simplicity that could make RSS ubiquitous (you need a home server and na application server to handle api calls) but the expected interactions are all there, and it's built to be extensible.
In fact, I think reusing the "everything is a file" is definitely cool and makes things easier to try, but you would definitely benefit from using the whole JSON-LD taxonomy. Things are already defined and have been used for some time in production, so you can expect some work has been thrown at it. It'd also make collaboration easier.