A specific capability or tool that an agent can invoke, search, code execution, API calls, data analysis, etc.
An agent skill is a discrete capabilityCapabilityStrategyAn ability that enables value deliveryView reference → an agent can invoke: a single tool, or a packaged procedure that bundles several steps behind one name. It is the unit an agent definitionAgent DefinitionWorkflows & AgentsAn autonomous agent definitionView reference → reaches for when it needsNeedUserA user need, pain, desire, or constraintView reference → to do something it cannot do by reasoning alone. The 2023 to 2025 shift to LLM tool-use made the boundary explicit, and the Model Context Protocol standardised how a model discovers and calls such capabilities at runtime.
A support-triage agent holds three skills: search the knowledge base, classify a ticket by urgency, and draft a reply. The first is a thin wrapper over a vector search. The third is a packaged procedure that pulls the customer's history, applies a tone template, and returns a draft. When a ticket arrives, the agent definition picks which skills to invoke and in what order. Add a fourth skill later, escalate to a human, and every agent that references it gains the capability without rewriting the others. The skill is defined once and reused across definitions.
In the Unified Product Graph, an agent skill lives in the automation region and is reachable from its agents through Agent Definitioncapable ofAgent Skillhierarchy. When a skill is built on top of something users already see, agent_definition_capable_of_agent_skillAgent SkillextendsFeaturecross-domain ties it back to the agent_skill_extends_featureFeatureProduct SpecificationA product capability or featureView reference → it draws on, which keeps agent capability and product capability from drifting apart. Modelling skills as their own entities, rather than burying them inside each definition, means a skill is countable, shareable, and auditable: you can ask which agents can do a given thing, and which product feature a given skill depends on.feature
Type-specific fields on BaseNode
skill_triggerstringEvent or command that activates this skill
skill_descriptionstringHuman-readable description of what the skill does
invocation_countnumberNumber of times this skill has been invoked
idstringrequiredUnique identifier (UUID)
typeNodeTyperequiredDiscriminator for the entity type
titlestringrequiredDisplay name
descriptionstringOptional detailed description
statusstringLifecycle status
tagsstring[]Freeform tags for filtering
4 phases — initial: todo · template: WORK_ITEM
2 edge types connected to this entity.
agent_definition_capable_of_agent_skillagent_skill_extends_feature