Agent skill (SKILL.md)

If you paste the same long prompt into chat every week, it should probably be a skill. According to agentskills.io, Anthropic originally developed the format and released it as an open standard. The same site lists dozens of agent products that support it, including Claude Code.

How it works

SKILL.md starts with a short header holding a name and a description of what the skill does and when to use it. Markdown instructions follow. Optional folders can hold scripts and reference files.

Loading happens in stages. The name and description are always visible to the agent. The full instructions load only when a task matches the description. Extra files load only when the instructions point to them. That staging lets an agent keep many skills installed without filling its context window.

Example

A marketing manager writes a skill for her weekly paid-media report: which columns to pull, how to label week-over-week changes, the email format her director wants, and one finished report as a sample. The description says to use it when someone asks for the weekly paid-media report or "Monday numbers". Typing "Monday numbers" in any session where the skill is installed now produces the report in the same format.

Common mistake

Writing a vague description such as "helps with reports", so the skill never triggers, or rules such as "be thorough" that nobody can check. Write the description as trigger conditions and each rule as something a reviewer could mark pass or fail.