skill-authoring

To create an Agent Skill, follow an evaluation-first methodology that guides you from concept to implementation. You will define the skill's purpose, structure its SKILL.md file, write clear descriptions, and establish validation loops to ensure reliability and discoverability.

Discovery metadata in SKILL.md consists of the name and description fields optimized for agent discovery. Properly structured metadata ensures that agents can reliably find and invoke the correct skill during runtime operations.

You validate Agent Skills by implementing validation loops and applying quality assurance checklists provided by the framework. This process identifies anti-patterns and verifies that runtime requirements are met before the skill is deployed.

The best way to structure SKILL.md for progressive disclosure is to separate core runtime instructions from detailed context using associated reference files. This ensures agents load only necessary information, optimizing execution efficiency.

Yes, you can update existing Agent Skills by refining their internal logic and reference files while preserving the optimized name and description fields in SKILL.md. This maintains agent discovery compatibility while improving skill functionality.

Common anti-patterns in skill authoring include poorly defined discovery metadata, missing validation loops, and failing to use progressive disclosure. The framework provides specific checklists to help you avoid these design and execution pitfalls.