What a skill is
Tool Use and Function Calling covers how a digital agent invokes a well-defined function with a name, a description, and a set of parameters, and gets a result back. A skill is the physical robotics equivalent of that same pattern. It's a discrete, well-defined capability the agent can invoke by name, like track a face, adjust a light, pick up an object, or navigate to a marked location. Under the hood a skill is still code, usually a small piece of software that knows how to translate a high-level intent into the specific sequence of motor commands or control signals needed to carry it out, but from the agent's point of view it looks exactly like calling a function: give it a goal and some parameters, and it either succeeds, fails, or reports that it's still in progress.
Packaging physical capability this way keeps the agent's reasoning simple. The agent never needs to know the exact torque curve for a particular motor or the calibration offset of a specific camera, it just needs to know that a skill called adjust a light exists, what parameters it accepts, and what it's supposed to accomplish. All of the messy, hardware-specific detail is buried inside the skill's implementation, not in the agent's decision-making.