Is small cap conversion done for every word that's written in all capitals? The convention that I'm familiar with (I think from The Elements of Typographical Style) is to only set acronyms - that is, pronounceable initialisms like "NASA" or "PETA" - in small caps. All other initialisms like FBI or USA are set as normal-sized caps. I think the reasoning is that small caps are there to preserve the capitals but not disturb the flow of the text, since it's read as a word just like the words around it. Full caps do disturb the flow of the text because they are not words and not pronounced as words either.
Any word in all-caps is 'converted' to small-caps. This behaviour can be changed, or even disabled, by setting a flag in the options object before processing. I prefer to keep all-caps out of my body text.
Back at Flipboard, for our RSS typography processing, we used a rule of thumb that all-capped words 3 letters or longer would get small-caps treatment. It worked pretty well.
I have used a similar tool [1] some time ago in PHP. The project hasn't been kept updated, but the idea of processing text/HTML and converting the right entities is pretty neat. At least the readability improves massively.
This project Typeset.js is slightly different as it adds additional markup and styles. For me, converting here's to here’s was enough for me back then.
I suppose you could write a grunt or gulp task to parse all of the files after jekyll has generated them and run them through typeset. It might be slightly challenging to ensure that only the actual article content gets ran through typeset and not any other content.
Unfortunately, I use Github pages so this wouldn't work for that from what I can tell.
