What Are Skills

If the base model is "I know Kung Fu," a skill is "I can now drive the forklift, access the warehouse, and move the boxes." Knowing is built in. Doing often requires extra capability.

That distinction is the whole point of skills. Your agent can usually reason, summarize, draft, and analyze out of the box. But reasoning alone does not grant access to tools, systems, or workflows. A skill adds that missing bridge between understanding a task and executing it.

For example, an agent may understand exactly how to run high-quality browser research, but without browser capability it cannot actually navigate pages. It may know that memory should be structured, but without the right memory tools it cannot query or maintain that structure efficiently. Skills turn "should" into "can."

Where do skills come from? Broadly, two places. There is an open marketplace where anyone can publish, which is fast and creative but uneven in quality. And there is a curated layer, where tools have been reviewed and are safer to rely on for real operations. The open layer is useful for discovery. The curated layer is usually where teams should start.

This changes how you diagnose "my agent cannot do X." Sometimes that means the task is genuinely out of scope. Often it means the agent has the reasoning but lacks the installed capability. That is a much better problem, because it is frequently solvable.

So do not treat your agent like a fixed feature list. Treat it like a capable generalist with an expandable toolkit. The baseline gets you surprisingly far. Skills are how you push past the baseline when your work demands it.