I think it would be great to separate the code base into multiple chunks of code: maybe work with layers?
In my head, there are 4 layers:

• Specific Bleach to a format+target, for instance "remove macro from OOXML office document".
• Format Bleach, responsible of cleaning all threats of a given format.
• The Bleach Factory™, responsible of cleaning a given thing no matter what it is (and so, to guess the type)
• 8th Layer - the user

This way, DocBleach would be a library, the "format bleaches" would be like addons/plugins, and calling docbleach.jar would launch rockets in every directions.