Agent Skills

Agent Skills is a simple, open format for giving agents new capabilities and domain expertise. Originally created by Anthropic, Agent Skills has now been adopted as a standard by Microsoft, OpenAI, and other major AI platforms. It packages specialized instructions and workflows into reusable skills that agents and agent platforms can consume. The project focuses on repeatability, interoperability, and enabling agents to perform multi-step, auditable tasks across different skills-compatible products.

• Domain expertise — Package specialized knowledge (for example legal review processes or data analysis pipelines) into reusable, shareable skills.
• New capabilities — Extend agents with concrete capabilities such as presentation creation, MCP server building, or dataset analysis by attaching skills that encode procedures and interfaces.
• Repeatable workflows — Define multi-step tasks as consistent workflows that agents can execute reliably and audibly for auditing and reproducibility.
• Interoperability — Reuse the same skill across different skills-compatible agent products to avoid vendor lock-in and speed integration.
• Open specification — The format is documented with an overview, a formal specification, and integration guidance for implementers and platform owners.

To get started, read the Overview, review the Specification, and follow the Integrate skills guidance to author or adopt skills for your agent platform. Developers typically implement skill handlers or adapters in their agents to load and execute skills according to the specification.