There's no particular reason "agent harness" can't have practically the same definition, substituting test-specific concepts for agent-specific ones.