I'm quite curious about the file reading code. With a simple directory browse from the user couldn't you start to walk the file tree and read whatever you liked? Nothing I see to stop something like this then firing off Ajax calls to send up everything it finds. I think some digging through the code is in order.
It seems quite clear that you could package up all files and send them to the server. I've read through the code in this example though and I don't see anything that could be doing that here. Since all of the code is being served off of github I think it'd be pretty hard to switch in some malicious code at times, but I would be pretty wary of granting access to a webkitdirectory chooser type from a less transparent site.
