It seems you are wanting "actions" to be called based on HTML elements.

I don't know how well it works, but HTML::Parser (H::P) has options to call your "callout function" for the opening and closing tags of specific HTML elements, or all of them. When you specify the callout functions, you tell H::P what you want passed to your function. For example I wanted to see the start/stop/text and non-parsed text (DATA/javascript) so I specified functions for each (mayka is a anon-sub creation routine that included some routine error checks and such).

    $p->parser(HTML::Parser->new("api_version" => 3,
      start_h   => [ mayka($p,6,start_h   => \&_start), 
                      "tag,skipped_text,attr,attrseq,line,text"],
      end_h     => [ mayka($p,4,end_h     => \&_end),  
                      "tagname,skipped_text,line,offset_end"],
      text_h    => [ mayka($p,3, text_h   => \&_text), 
                      "tag,skipped_text,text"], 
      default_h => [ mayka($p,3,default_h => \&_dflt), 
                      "event, skipped_text, text"],
      marked_sections => 1,));
[download]

The last parameter specified what elements I wanted passed to my function, so for tags that had class labels, I could store & nest them.

Note -- I may easily be missing some functionality in WWW::Mechanize, but I didn't see the ability to process class or ID values when they started and ended. It does get a bit hairy trying to keep track of them, since nested elements preempt and assign class+id's to children, and when those elements end, the class+id revert to whatever was in place before you encountered that element (i.e. need to maintain a stack)...

hope this helps...