They are typically not the semantics that's desired, i.e. they only provide the command output but does not assert that the command status is success. We should make sure we're not silently letting failed commands through.

Do we even want to allow command_output methods? It's likely less footgunny if we only provided CommandWrapper::{run, run_fail, run_fail_assert_exit_code} methods which perform asserts on the command exit status.