Your problem is that

<META http-equiv="...
[download]

is not a link tag and hence WWW::Mechanize does not report it. Especially as the content attribute does not contain an URL but the timeout and an URL. I suggest you extract the target URL manually via a regular expression if the website you're parsing is fairly static, or using HTML::TokeParser or HTML::HeadParser if you have to deal with many types of pages.

my $c = $agent->content;
# my $target
if ($c =~ m!<meta\s+http-equiv=['"]refresh["']\s+content="\d+;([^"]+)"
+!si) {
    warn "Refresh found (-> $1), following";
    my $target = $1;
    $c->get($target);
};
[download]

A quick Google search for www mechanize refresh turns up WWW::Mechanize follow meta refreshes...