skill-management

Agent skills are organized for on-demand discovery by storing them as SKILL.md files in standard local directories. The agent scans these locations and loads the matching skill only when the task context aligns with its YAML frontmatter description.

Mandatory YAML frontmatter requires the name and description fields for skill recognition. This metadata ensures the agent can reliably discover the skill and match it accurately to the relevant task context during on-demand loading.

Create a new agent skill by authoring an instructional markdown body with valid YAML frontmatter and saving it to the correct local path, such as .github/skills/<kebab-name>/SKILL.md, ensuring it follows the Skill Protocol.

Agent skills are governed by the Skill Protocol to enable on-demand matching and metadata management, whereas tools and declarative instructions operate differently. Protocol governance clarifies these operational differences to ensure consistent skill activation.

Protocol versioning is managed through the compatibility field in the YAML frontmatter. This ensures that agent skills adhere to the correct Skill Protocol version, maintaining reliable discovery and governance across supported directories.

Skill management applies to both projects and personal environments. It enforces the Skill Protocol to ensure that skills are consistently discovered, activated, and authored across all supported local directories.