Fair enough. Although even then I would consider using a function or a class or lambda first, I suppose as long as you completely control the code being evaled it's not fundamentally different than any other string or code on the server. It would still fill me with a gnawing sense of dread though.
(1) Prints some info about the test that's about to execute
(2) Evals the command
(3) Checks/reports the result
Given the choice I'd rather use lambda, but not all languages make that very convenient/readable. In such languages, a carefully used eval might be preferable to ad-hoc interpretation.