The people running the agent need to be looking at the code and cleaning it, not someone else, that is the key thing that is missing in all these failures. Linux is the same, they would never accept something that has not at least the same quality as something written by a competent human