Looking at the repository, the impression I get is that this codebase had a web interface grafted onto it relatively recently. Most of the code looks like it was written with the expectation that it'd be used over a line-mode interface. (A lot of it also looks like it was converted from C to C++ very late in the development process -- outside of the WebSockets library, there's hardly any code which uses even basic STL features like string or vector.)
yes, if you know the history of muds, and diku specifically, you would know that, but where those changes were made it could more easily have been changed to respond either as text, and styled in the browser without adding all of the html elements, or as json. it was a very weird decision to be made given its history and the new goals of the project.
[0] Can't remember if this was before or after PLT Scheme got renamed to Racket, but it was around that time. It turned out to be a great language for it. Wish I had done more with it.
as an aside, I have a nicely working copy of sillymud hanging out in my GitHub, and local copies of phoenix (the successor to sillymud), and epic (another highly customized mud that ran off of goldman.ai.mit.edu).