What is an "eval" in the context of validating a skill?	A short, realistic try-out or test drive to catch potential failures before sharing the skill.
Why are evals necessary when sharing a skill with a teammate?	Teammates may use the skill with different inputs, phrasing, or edge cases that the creator hasn't anticipated.
What two types of outputs does Skill-Creator generate for an eval?	One output where Claude uses the skill, and one output answering the same prompt without the skill.
What is the purpose of comparing the two outputs during an eval?	To determine if the skill version is meaningfully better than the baseline performance of Claude on its own.
When giving feedback during an eval, what makes the feedback actionable for Claude?	Being specific and descriptive (e.g., "The tone is too formal") rather than vague (e.g., "this isn't quite right").
What is the best practice when iterating on a skill after submitting feedback?	Change only one thing at a time, re-run the prompts, and review the results to see which change was effective.
What is the minimum requirement for "shipping" a skill?	That the cases you care about pass meaningfully better than the baseline, and that you have identified the cases you still need to handle.