Installed more skills, but still fixing everything the AI hands back? Before adding another, check where the job is breaking.
Did the agent load the right instructions? Could its tools reach the required data? Did anyone check the result against an independent acceptance test? A failure at any of these points can produce a polished answer that you cannot use.
On October 22, Anthropic announced the donation of Agent Skills to the Agentic AI Foundation. That was last week. The useful question now is how to tell whether a skill improves your work. Here is a practical approach using refund support. This is documentation analysis and a proposed workflow, not a product test.