Do you systematically evaluate agent skills, either before or while using them? If so, what tools or methods do you use? Or do you mostly rely on reading the SKILL.md and trying things out?