Agent Skill

An Agent Skill is a high-level, reusable capability bundle that AI agents adopt — specialized prompts/instructions plus the tools needed to apply them. Skills sit one level above tools: where a tool is a single callable, a skill packages domain expertise (the "how" and "when") together with the low-level tools it relies on. The same skill can be adopted by many agents, so modeling it as its own entity avoids duplicating that expertise on every agent.

Skills follow a "git as source of truth" pattern (e.g. the agentskills.io standard): the skill is defined in a repository and cataloged in DataHub for discovery, governance, and reuse.

Identity

Agent Skills are identified by a single field:

• id: A unique identifier for the skill, typically a human-readable slug such as customer-service-skill or a generated UUID.

An example URN is urn:li:agentSkill:customer-service-skill.

Important Capabilities

Skill Info

The agentSkillInfo aspect holds the skill's definition:

• name: Display name, searchable with autocomplete.
• description: What the skill does and when to use it.
• instructions: The markdown body of the skill's SKILL.md file (excluding YAML frontmatter) — the execution guidance loaded when the skill is activated.
• sourceRepository: Where the skill is defined (see below).
• requiredTools: The API entities the skill needs to operate (SkillRequiresTool).
• created / lastModified: Audit stamps.

Source Repository

The sourceRepository field (a SkillSourceRepository) points to the git location that owns the skill definition. It carries either a repositoryUrn (preferred, when the Repository is already cataloged in DataHub) or an external url, plus a path to the definition file within the repository (e.g. customer-service/SKILL.md).

Relationship to Agents

Agents adopt skills via the AgentHasSkill relationship declared on the agent's aiAgentDependencies aspect — so an Agent Skill profile surfaces the agents that use it, and a skill in turn exposes the tools it requires. This makes the agent → skill → tool chain traversable from either end.

Governance and Versioning

Agent Skills support the standard governance aspects — ownership, tags, glossary terms, domains, structured properties, and institutional memory — plus native versioning (versionProperties).