If you are new here, a quick explainer.
A skill is a package of instructions your assistant loads when it is relevant. Think of it as a saved way of doing one task well. It knows when to trigger, what steps to follow, and what a good result looks like. You do not have to remember the perfect prompt, because the skill already carries it.
To use one, find it in the skills catalog and follow the install steps. Most skills are a single command. Once installed, you invoke it by describing the task or using its trigger phrases, and your assistant takes it from there.
An agent is a step up. It bundles the MCP servers it needs to reach real data, the skill that drives the work, and a runbook for setup and auth. That is how an agent can do things a plain prompt cannot, like reading your calendar or querying a system.
Start with one skill that matches something you do often. See how it feels, then add more. When you want to build your own, read build your own Claude skill for a step-by-step walkthrough. If you want the wider story of why we build these, read the knowledge harness or the operator version at edakrong.com.