Agent Skills Joins AAIF: What Developers Can Reuse

Anthropic has donated Agent Skills to AAIF. What the donation includes, how skills work, and how they fit alongside AGENTS.md and MCP.

On October 22, Anthropic announced the donation of Agent Skills to the Agentic AI Foundation. The donation includes the name and logo, GitHub organization, and agentskills.io domain. Developers can keep using and contributing to the standard without joining AAIF.

I covered some of this in my earlier post on agents for community managers and my Skills Over MCP talk. Both explored how instructions and tools can work together across agent applications.

Agent Skills gives reusable procedures a shared format. A team can package its review process, data-analysis steps, or document templates in a folder that compatible agents can read. The move to AAIF puts the standard's development in a neutral home under the Linux Foundation.

What changes with the donation

Agent Skills was already an open standard before this announcement. Anthropic published it in December 2025, following its initial development for Claude Code. The donation moves stewardship of the standard to AAIF.

Engineers from Anthropic, Google, OpenAI, and other companies will maintain it as individuals. David Soria Parra and Mahesh Murag are the lead maintainers. Developers can contribute without becoming foundation members.

For teams adopting skills, shared stewardship provides a place to discuss specification changes across implementations. Client developers and skill authors can raise compatibility problems against the same public standard.

What a skill contains

A skill is a directory with a SKILL.md file. That file contains YAML metadata and Markdown instructions; the directory can also contain scripts, reference documents, and assets. The required metadata fields are name and description. The specification defines the format and its constraints.

The optional directories have different jobs:

my-skill/
├── SKILL.md         # Metadata and instructions
├── scripts/         # Executable code
├── references/
│   └── guide.md     # Supporting documentation
└── assets/          # Templates and other resources

The instructions describe the procedure, expected inputs and outputs, and relevant exceptions. Supporting files let authors include the material an agent needs without putting everything into one document.

The specification also defines optional metadata. A compatibility field can describe required software or network access, while license records the terms under which the skill is shared. Those details help users assess whether a skill fits their environment before running it.

How agents load skills

The Agent Skills overview describes progressive disclosure in three stages:

1. During discovery, the agent gets the names and descriptions of available skills.
2. When a skill is relevant, the agent reads its full SKILL.md instructions.
3. During execution, the agent loads supporting files or runs bundled code as needed.

This lets an agent keep a catalog of skills available while loading detailed instructions for the task at hand. The description helps it choose which skill to use; the body explains how to do the work.

Authors still need to decide what knowledge belongs in a skill. The authoring guidance recommends drawing from real tasks, corrections, and project artifacts. A team's conventions and recurring mistakes give an agent context it would otherwise lack.

Where AGENTS.md and MCP fit

Agent Skills complements two other AAIF projects that developers may already use:

AGENTS.md gives coding agents a predictable place to find project guidance. A skill can explain a particular procedure, and MCP can provide access to the external capabilities it needs. An agent application coordinates those pieces.

Reusing a skill across applications still depends on the surrounding environment. A skill that requires a particular tool or script needs that dependency in each client. Permissions and approval controls also belong to the application and services running the workflow.

The client implementation guide describes different discovery and activation mechanisms. The format defines what goes inside a skill directory; clients decide where to discover it and how to load it. Support for the format gives authors a common starting point, while behavior still needs checking in each target client.

Anthropic has proposed versioned releases, a formal proposal process, and security guidance as next steps. Developers can read the specification and join the discussion in the Agent Skills repository.