Generally speaking, you'll see the same pattern for any problem involving searching a space of potential solutions. In a word, the strength of LP is flexibility when facing the unkown (future feature requirement).
EDIT: you mention parsing. If you're planning anything context-sensitive, you have to check out eDCGs. You probably won't use them, but they'll give you an idea of what's possible.
This feels like a much better path than trying to get it into an AST with like yacc or whatever. But since it isn't actually natural language NLP didn't seem right either. I've got a lot of reading to do but this is very interesting.
Your mud language is probably a "Controlled Natural Language", btw:
https://en.wikipedia.org/wiki/Controlled_natural_language
Game languages often are.
https://github.com/kamahen/edcg , https://occasionallycogent.com/prolog_edcgs/index.html
Curious - I fuzzily recall usually ending up manually threading state (eg, doing a Perl-Compatible Regular Expression engine (apropos "why LP?" - it was just a page of code, with good performance)), but I don't recall what drove that decision...
I guess I think of logic languages as SQL but I actually want to write my whole program in it.
I am kind of half-considering it for mud scripting? Each line from the mud represents a fact or relation so this could be used to query for specific information as needed.
For this case though, it doesn't really seem better than writing a regex-based DSL though, something I'm far more familiar with and that is better supported.