What Claude Skills Are and When to Reach for One

A skill is a folder containing a SKILL.md file and any resources it needs. The file carries a name, a description of when it applies, and instructions. Claude reads the description, decides whether the current work matches, and if it does, follows the instructions.

The important part is that you do not invoke it. You describe your problem in ordinary language and the right skill activates because its description matched. This is the difference between a skill and a saved prompt: a prompt waits to be pasted, a skill waits to be relevant.

That makes the description the most important line in the file. It is not documentation for humans, it is the trigger. A skill with excellent instructions and a vague description never runs, which is the single most common way a first skill fails.

Reach for a prompt when you want one specific output now and the instruction is the whole job. Copy-ready build specifications work this way: you paste, you get a scaffold, you move on. Nothing needs to persist.

Reach for a skill when the same expertise should apply repeatedly, without you remembering to invoke it. Reviewing prompts for contradictions, auditing traces, scoping an engagement: these are judgements you want available every time the situation arises, not pasted from a document you have to find.

Reach for a tool or an MCP server when the task needs to reach outside the conversation: query a database, call an API, read a file from a system Claude cannot see. Skills carry knowledge and procedure. Tools carry access. Confusing the two produces a skill that describes an API it cannot call.