This is going to put on a slight extra burden that you need to think about commands that may produce output, and if we change a command that is initially silent to produce output, we may need to update a bunch of tests. However I think this is much safer than not letting people check things actually are silent.
OK, fixed in the attached branch.
This is going to put on a slight extra burden that you need to think about commands that may produce output, and if we change a command that is initially silent to produce output, we may need to update a bunch of tests. However I think this is much safer than not letting people check things actually are silent.