Skill

The term 'skill' is used in a few related but slightly different ways in AI builder discourse. In the agentic context, a skill is a modular, named capability an agent can invoke: 'research this topic,' 'draft a response in my brand voice,' 'analyze this data.' The skill encapsulates all the prompting, tool use, and logic needed to accomplish that capability, so it can be reused across many tasks.

Claude Skills (in Claude's native UI) and agent skills frameworks each put a slightly different spin on the concept, but the shared idea is: packaging up recurring AI capabilities into named, callable units that can be composed together. A skill registry is a catalog of available skills that an agent or orchestrator can draw from when planning how to accomplish a task.

The practical value of the skill abstraction is modularity and reusability. Rather than rebuilding the same complex prompt chain every time you need to do a particular thing, you define it once as a skill and call it by name. It also makes it easier to update the underlying implementation without changing the interface. This is the 'functions and modules' idea from software engineering applied to AI behavior.