I've worked with some very, very large complicated ones so yousort of get
used to it, especially with nice tools like oXygen.
I'd keep this one quite simple and they wouldn't be very big unless you
set one up for gcc :-)
I still miss some 'kick-ass' use case... unix tools should be run from unix environment (to see the output, to have the ability of using pipes etc) and the rest doesn't need this xml approach. If you convince me / us it's really really needed and useful, your chances to get the implementation one day will raise :)