My own aggregator (ikiwiki) provides two ways to do it, either by using the HTML::TreeBuilder module to build a html parse tree and re-emitting with balanced tags, or by calling out to tidy(1), which will go all the way to generate valid html (or in rare cases, give up and declare the post unparseable).

https://www.mirbsd.org/cvs.cgi/www/data/Makefile?rev=HEAD

and:

https://www.mirbsd.org/cvs.cgi/www/Makefile?rev=HEAD

There are validation targets that use xmlstarlet and a
standalone XHTML/1.1 DTD (all I need as everything of
mine uses it) to validate. Of course, the W3C validator
*may* do additional things, but this is a basic test I
use before committing.

There’s some magic, especially in the former (as it con-
verts these *.inc files to XML first) but also in the
latter. The reason for this is that xmlstarlet fails if
there is an URI given in the doctype, which however must
be used in the on-site copy. The www/data/Makefile is a
snippet that can be used well for validating only HTML
snippets instead of whole pages.

HTH, HAND

XHTML has too many problems to mention.