Agent Skills Overview

Agent Skills is an open format designed to extend the capabilities of AI agents by packaging specialized knowledge and workflows into modular units called skills. md file, which holds metadata such as the name and description, along with instructions that guide an agent in performing a specific task. Additional resources like scripts, reference materials, templates, and other files can also be included within the skill folder, allowing for a comprehensive and portable solution to augment agent functionality.

The format addresses the challenge of providing AI agents with the necessary context and procedural knowledge to perform complex or domain-specific tasks reliably. By encapsulating expertise—such as legal review processes, data analysis workflows, or presentation formatting—into reusable and version-controlled folders, Agent Skills enables agents to access and execute repeatable, auditable procedures tailored to company, team, or user-specific contexts. Skills are loaded by agents through a progressive disclosure process: initially, only basic metadata is loaded to determine relevance; when a task matches a skill, the full instructions are read into context; and during execution, the agent may utilize bundled code or reference additional resources as needed. This approach allows agents to manage a large set of skills efficiently, minimizing context overhead until detailed instructions are required.

Agent Skills can be used across a variety of AI tools and agentic clients that support the format, promoting cross-product reuse of skills. The format is open and was originally developed by Anthropic, released as a standard that welcomes contributions from the broader community. The development process is collaborative and ongoing, with opportunities for engagement through GitHub and Discord. Agent Skills is positioned as an open standard for enhancing AI agents with modular, context-rich expertise and workflows.