LangChain publishes guide to evaluating agent skills

The guide covers best practices for testing AI agent skills, including setting up a clean environment and running tasks with and without skills. It emphasizes progressive disclosure and using LangSmith for observability and evals.