Agent Skills
Package know-how as Agent Skills: the SKILL.md format, progressive disclosure, writing a description that triggers, attaching Anthropic and custom skills to API requests, and auditing a skill before you trust it.
Last updated
After this section you can
- Write a SKILL.md whose description fires on the right requests, and test it
- Explain what progressive disclosure keeps out of context, and what it costs when a skill fires
- Attach Anthropic and custom skills to an API request, and audit a skill before trusting it
Agent Skills: Expertise That Loads on Demand
A skill is a folder of instructions, scripts and reference files. Claude sees one line about it until a request needs it, then reads the rest from disk. That is how an agent carries dozens of procedures without paying for them on every call.
A skill moves know-how out of the system prompt and onto a filesystem. Only its name and description sit in context. When a request matches the description, Claude reads SKILL.md, then opens further files or runs scripts only if the task needs them. The description is the trigger, so it decides whether the skill is ever used.