Keep GitHub Copilot Agent Skills Small and Focused

GitHub Copilot Agent Skills work best when kept small and focused on a single behaviour. Overloading a skill with too many instructions creates unpredictable output and competing priorities. The post recommends designing skills around one clear job, using specific descriptions so the agent loads the right skill, keeping SKILL.md concise while offloading deeper material to references/ directories, and using concrete examples over long rule lists. Scripts should only be added when repeatable validation is needed. The goal is consistent, predictable output that teams can rely on — not a mega-prompt trying to cover every scenario.