The format of those files is not overly complex. Do we have clean enough metadata to generate them programmatically? I assume the book title and book number are low-hanging fruit, but I'm not sure about other information, e.g. illustrator(s), editor(s), subtitle...