Not sure what it would look like, but it'd be sweet to catch failures like #185 before they happen in user repositories.