At first, I thought that might get too low-level for me. But then I saw that XML::Rules->new has a handlers => {...} which allows defining handlers for XML::Parser::Expat events. Some experimentation with a dummy callback to handle them all says that Comment and XMLDecl are the events I want during the parsing.

#!perl

use 5.012; # strict, //
use warnings;
use Data::Dump;
use XML::Rules;

my $xml_doc = <<EOXML;
<?xml version="1.0" encoding="UTF-8" ?>
<!-- important instructions to manual editors -->
<root>
    <group name="blah">
        <!-- important instructions for group "blah" -->
        <tag/>
    </group>
    <group name="second">
        <!-- important instructions for group "second" -->
        <differentTag/>
    </group>
</root>
EOXML

my $callback = sub {
    my ($name, $parser, @args) = @_;
    print STDERR "event:", $name//'<undef>', "(";
    print STDERR join ', ', map {defined($_) ? qq("$_") : '<undef>'} @
+args;
    print STDERR ")\n";
};

my %handlers = ();
for my $h ( qw/Comment XMLDecl/ ) { #qw/Start End Char Proc Comment Cd
+ataStart CdataEnd Default Unparsed Notation ExternEnt ExternEntFin En
+tity Element Attlist Doctype DoctypeFin XMLDecl/) {
    $handlers{$h} = sub { $callback->($h => @_) }
}

my $parser = XML::Rules->new(
    stripspaces => 3|4,
    rules => [
        _default => 'raw',
    ],
    handlers => \%handlers,
);

#dd
my $data = $parser->parse($xml_doc);

print
my $out = $parser->ToXML($data, 0, "  ", "") . "\n";

__DATA__

event:XMLDecl("1.0", "UTF-8", <undef>)
event:Comment(" important instructions to manual editors ")
event:Comment(" important instructions for group "blah" ")
event:Comment(" important instructions for group "second" ")
<root>
  <group name="blah">
    <tag/>
  </group>
  <group name="second">
    <differentTag/>
  </group>
</root>
[download]

Now that I've got that far, I should be able to get the prolog and comments into the data object (by returning values, instead of just printing messages). But the harder part will be how to get ->ToXML() to do something on the output. I may have to subclass XML::Rules to get additional outputs for my comment and prolog data items -- if anyone has an easier idea than that, feel free to let me know.

I should be able to get the prolog and comments into the data object

I was optimistic. I misunderstood that the handler return expectations would match the rule return expectations, and whatever was returned from the handler would be added to the data structure, similarly to a custom rule. That didn't happen. And with haukex's advice that round-tripping is difficult, this might not be the right path after all. If anyone has advice for continuing down this path, I will definitely experiment with it, no matter which direction I go from here.