skill-authoring

To structure a Claude Skill, create a mandatory SKILL.md file with YAML frontmatter and organize optional scripts, references, and assets directories to ensure discoverability and efficient agent loading.

Progressive disclosure in LLM agent development is a content organization pattern that layers skill information, managing token budgets by loading only necessary reference files when specific agent decisions are triggered.

Writing YAML frontmatter for Claude Skills requires defining mandatory metadata fields that establish naming conventions and descriptions, ensuring the skill is correctly parsed and easily discovered by the agent environment.

Claude Skill loading fails during validation due to common anti-patterns like malformed YAML frontmatter or incorrect file structures, which can be resolved by checking SKILL.md formatting and adjusting content organization.

Bundled resources and scripts are optional for developing a Claude Skill, but adding these directories allows you to organize supplementary reference files and executable code for more complex agent workflows.

The best way to organize reference files for a developer guide skill is using progressive disclosure patterns and decision trees within the references directory, optimizing content retrieval while respecting token budget limits.