Agent Skills
Agent Skills are reusable procedures packaged as folders with a SKILL.md file — loaded by an AI agent on demand when a task matches, now an open standard.
Agent Skills are reusable procedures packaged as a folder containing a SKILL.md file — YAML frontmatter that tells the agent when the skill applies, plus a Markdown body of instructions it follows once loaded.
The design solves the standing tension between capability and context: an agent that carries every procedure in its system prompt pays for all of them on every request. Skills invert that with progressive disclosure — at session start only each skill's name and description load; the full body loads when a task matches; bundled scripts and references load only when the instructions reach for them. The description doubles as the routing signal, which is why writing it well is most of the craft.
Skills began as a Claude Code feature and became a portable format: the same SKILL.md runs on claude.ai, on the Claude API via /v1/skills and code execution, and — since Anthropic published Agent Skills as an open standard in December 2025 — in dozens of third-party tools. The full guide covers the concept; the SKILL.md reference documents every field.
Frequently asked questions
- What is the difference between Agent Skills and Claude Skills?
- Same thing, two scopes. 'Claude Skills' is the everyday name for the feature across Claude Code, claude.ai, and the Claude API; 'Agent Skills' is the format's formal name — and since December 2025, an open standard (agentskills.io) that non-Claude tools like GitHub Copilot, Cursor, and Gemini CLI also implement.
- Why not just put the instructions in the system prompt?
- Cost and scale. System-prompt instructions are paid for in every request whether relevant or not. A skill costs only its name and description at rest — the body loads when a task matches, and bundled files later still. That progressive disclosure is what lets an agent carry dozens of procedures affordably.
Related
- What Are Claude Skills? The Complete GuideClaude Skills explained: what a SKILL.md is, how progressive disclosure keeps skills cheap, where they run, and how to install or write your own.
- The SKILL.md Reference: Every Frontmatter Field ExplainedA complete reference for the SKILL.md format — all frontmatter fields, naming rules, argument substitution, limits, and where skill files live.
- Writing Your First SkillA step-by-step guide to packaging a reusable procedure as a Claude Code skill that loads exactly when it's needed.
- Skills vs Agents vs CommandsHow Claude Code's two extension mechanisms — subagents and skills — differ across three invocation patterns, with a decision table for choosing the right one.
- Context WindowThe context window is the maximum text — measured in tokens — an LLM can consider at once: prompt, conversation, documents, and its own output combined.
- System PromptThe system prompt is the standing instruction layer an LLM receives before user input — defining its role, rules, tools, and tone for the whole conversation.
- Claude Skills on claude.ai and the APIHow Agent Skills work beyond Claude Code: uploading to claude.ai, the /v1/skills API with code execution, Managed Agents, and the Agent SDK.