A set of quoting modules for different shells (Bourne, csh, MS, etc.) is included as part of my module Net::OpenSSH, and it is a can of worms.

As part of the test suite, random strings are generated, quoted and parser by the set of supported shells found by the module in every computer where the tests are run. Over the years I have received a good number of bug reports related to that, most of the times caused by bugs or by unexpected behaviors in the shells, specially for commercial vendors maintaining their code in isolation (for instance, IIRC, AIX ksh).

So, String::ShellQuote may have its issues, and retrospectively some of them probably just caused by bad design decisions, but it should also be taken into consideration that the complexity of the problem it tries to solve is not negligible.