######## foo.pl ############
#!/usr/bin/perl
use strict;
use warnings;

MyApp->import( 'run' );
run() unless caller();

package MyApp;
BEGIN {
    use Exporter 'import';
    our @EXPORT_OK = qw(run mysub);
}

sub run {
    my $result = mysub();
    print "$result\n";
}

sub mysub {
    return "mysub called";
}

########## foo.t #############
#!/usr/bin/perl
use strict;
use warnings;
use Test::Simple tests => 1;

require 'foo.pl';
MyApp->import(qw(mysub));

ok ( mysub() =~ 'mysub' , '&mysub successfully imported' );
[download]

• foo.pl does what I mean,
• prove foo.t does what I mean,
• I can still copy only foo.pl on whatever server I want and it would do it's job,
• I've reduced untestable code from 100% to just the tiny main:: package (which is copy-pasteable) and the run() subroutine (whish can be kept concise enough),
• all new magic ends after the BEGIN block in the package (new subs should be added into @EXPORT_OK for the benefit of the test script); the BEGIN block was needed to have @MyApp::EXPORT_OK defined before the import() call, otherwise I would have needed to move the main package to the end of the file.

All this being said, I think it's clear enough to show to people less Perl-inclined and to help preach the benefits of testing, even for sysadmin scripts. Any improvement ideas?