I know it's not actually a blogroll but I'm toying with the idea of using the Directory to dynamically build a live OPML file and advertise it for download when visiting the directory page.

Creating the file is not an issue and I already have the author names and URLs - I'll just need to grab the feed addresses from each page.

I say just like it's easy!

I'll have to load the remote page then explore the DOM and check for <link> elements with type='application/rss+xml' - hopefully the first one on each page will be the site's main feed.