Assuming that you need to figure out the columns dynamically, here's one way to do it:

use warnings;
use strict;
use XML::LibXML;
use Text::CSV;

my $dom = XML::LibXML->load_xml(location => 'example.xml');

my %items;
for my $product ($dom->findnodes('/products/product')) {
    my $sku = $product->findvalue('sku');
    for my $downloads ($product->findnodes('downloads')) {
        for my $group ($downloads->findnodes('group')) {
            for my $group_items ($group->findnodes('group_items')) {
                for my $item ($group_items->findnodes('item')) {
                    my $name = $item->findvalue('name');
                    my $url = $item->findvalue('url');
                    $items{$sku}{$name} = $url;
                }
            }
        }
    }
}
my @columns = sort keys %{{map {$_=>1} map {keys %$_} values %items}};

my $csv = Text::CSV->new({binary=>1, auto_diag=>2, eol=>$/,
    always_quote=>1 });
$csv->print(select, ["sku", @columns]);
for my $sku (sort keys %items) {
    $csv->print(select, [ $sku, map { $items{$sku}{$_} } @columns ]);
}
[download]

Note that this does assume that SKUs and the item names per SKU are unique (Update: in other words, that later <item>s don't overwrite earlier ones in the %items hash). Also, note that unless you need to separate out the <downloads>, <group>, or <group_items> elements, you don't need all those nested loops, this produces the same output:

for my $product ($dom->findnodes('/products/product')) {
    my $sku = $product->findvalue('sku');
    for my $item ( $product->findnodes(
            'downloads/group/group_items/item') ) {
        my $name = $item->findvalue('name');
        my $url = $item->findvalue('url');
        $items{$sku}{$name} = $url;
    }
}
[download]