Measuring Agent Skills: The Hidden Cost of Shipping Without Evals

Most teams ship Agent Skills without measuring whether they actually improve performance or trigger correctly when users need them. Vercel's research found Skills remain uninvoked in 56% of cases, and even Anthropic's own skills saw improvement in only 5 of 6. With Anthropic's updated skill-creator bringing rigorous evaluation frameworks, the gaps in unexamined skills become finally measurable.