In #30 I hardcoded the list of packages in the `prepare-to-publish` script. That should be improved by instead reading from directories.
