Keep GitHub Copilot Agent Skills Small and Focused

This article discusses the common mistake of overloading GitHub Copilot Agent Skills with too many instructions, leading to unpredictable AI output. It advocates for creating small, focused skills that capture reusable engineering judgement for specific tasks, such as Terraform module authoring versus provider upgrades. The piece explains how skills are loaded, the importance of progressive disclosure, and when to use scripts instead of cramming guidance into one file. It provides practical tips on splitting skills, using examples over rules, and designing for predictable behavior, making it a useful guide for developers optimizing AI-assisted coding workflows.