OS and libraries have different roles: an OS manages the sharing of resources between programs, while libraries implement common operations in programs. The abstraction aspect in both really is a side effect.
Back to the topic: I think the real obstacle to this USLS idea is that each language has its own idiomatic way to name things, so a common across languages naming scheme or convention will likely look somewhat alien in each of them, and we all know the importance of source code aesthetics for programmers. This might change only if polyglot programming becomes very common.
Side Comment: If Oracle prevails in copyrighting an API this problem will get worse.
I ran smack into a variant of this problem playing around with ARM chips, libc. Its one of the places where gcc has really let me down, because if you get an ARM cross compiler its expecting anything and everything. The various flavors of C library are all bulky and nicely modularized but somehow less useful. And of course there are things in there which assume an OS (stdio for example) or at least a model that the code adheres to) but embedded systems are sometimes not that easily adapted.
And there are libraries provided by the manufacturers which have copyright notices in them about how this code can only be compiled to run on their (the manufacturer's) brand of ARM and when I look at the code I keep thinking "Hmm, I bet if I dug up the code I wrote for SunOS 4.0 when I was but a wee kernel hacker to make libc work with shared libraries, that a lot of it would be eerily familiar :-)
So what it boils down to are that there are 'things' and the useful things are 'memory abstraction (bcopy/memcpy etc)' string abstractions 'str' and character abstractions 'is' and then io abstractions (open/close/read/write/seek/rewind) and then thread abstractions (fork/sleep/wake/longjmp/yield) and suddenly your USLS is basically an OS API (or it could be) and you're back to square zero.
So rather than a universal library, why not universal VIM code? VIM is already syntax aware based on your language you could add the ability to insert the appropriate code to do what you wanted to do while you were editing. So if you had a variable and you typed if a:upcase == b:upcase it would syntactically awarely conver a:upcase to 'strtoupper(a)' or u"$a" or a->upper() or what ever the language you were writing in needed. Sort of a markdown for library calls. Then you train your fingers to type the markdown and be done with it.