Probably because of bike shedding. It’s not exactly hard to write some simple code from scratch to read the input files, pass them through various parsers, templates, processors, and write out the result. The heavy lifting is done by the processors, which you typically don’t need to write yourself.

So an SSG is an attempt to put an abstraction on top of this, which may have some benefits but also introduces a lot of relatively uncalled for complexity and fragile dependency chains. And because the nature of the task is very heterogeneous, the API will either have to be very allowing (and less useful as an abstraction) or impose various assumptions and restrictions on how you do things. That’s very fertile ground for a lot of people to want to create their own take on it as they run into limitations of existing SSGs.