write-a-skill

To structure reusable agent skills, use a standardized creation process that enforces mandatory SKILL.md frontmatter, optional script directories, and optimized description formatting for reliable agent discovery.

Progressive disclosure in AI workflow development is a technique that splits detailed criteria into separate reference files and bundles executable utility scripts to optimize token usage and performance.

Create a SKILL.md file by applying built-in templates, formatting guidelines, and review checklists to ensure valid frontmatter and clear trigger terms that meet agent discovery requirements.

Add executable utility scripts to an agent skill when you need to enforce specific validation rules, such as checking pull request label compliance, following resource bundling guidance to optimize performance.

Custom agent skills have poor discoverability due to inconsistent structure and missing best practices, which makes them hard to maintain and prevents agents from effectively triggering the correct workflow.

Yes, you can build a GitHub pull request review automation by structuring the SKILL.md with trigger terms, adding a validation script for label compliance, and splitting detailed review criteria into reference files.