What Makes a Good GitHub Copilot Agent Skill?

This article explores what makes a GitHub Copilot skill genuinely useful versus disappointing, based on experience building skills for Azure API Management, infrastructure as code, and diagram generation. It emphasizes that a skill is not just knowledge but a combination of trigger, boundary, and operating model. Key insights include the critical role of the short description in YAML frontmatter for agent relevance, the need for clear boundaries and task-specific design, keeping the skill body lean with on-demand detail, and testing triggers rather than just content. The article provides practical guidance for developers creating Copilot extensions, focusing on real-world workflow alignment and resource bundling.