The (recently-enabled) live test coverage stats found a few low-hanging
fruits in the tests we do…

The live test needs a populated cluster, so let's write this down.

This can be used to test that all the existing commands work correctly. It
needs a running cluster with at least one instance to run all the tests.