Agent Skill Harbor

This page is a guide from the administrators of your Agent Skill Harbor instance to developers in your organization.

For an overview of Agent Skill Harbor itself, see the repository README.

You can override the content of this page by placing guide/index.md (guide/index_ja.md) in the repository. You can also add pages by placing files like guide/01-foo.md (guide/01-foo_ja.md).

What Is Agent Skill Harbor?

Agent Skill Harbor is a skill sharing and discovery service for teams, and a skill governance tool for organizations.

It catalogs Agent Skills (SKILL.md) across all repositories in a GitHub Organization and publishes a browsable internal skill catalog. This helps developers discover useful skills already used inside the organization and learn from existing skill authoring practices.

How It Works

Skills are collected by periodically crawling each repository, so developers do not need to go through a separate registration or submission process after creating a skill.

Agent Skill Harbor can also analyze the provenance of skills installed from external repositories. To enable this, developers need to use agent-skill-porter when downloading skills from public repositories. The CLI can add skills with a simple command like this:

npm install -g agent-skill-porter
sk add https://github.com/anthropics/skills

The provenance mechanism is simple. agent-skill-porter adds a special _from property to the frontmatter of the target SKILL.md. It records a value such as anthropics/skills@b0cbd3d, using the format <owner>/<repository>@<sha>. Combined with the name property, this allows the download source to be identified uniquely.

Notes

Adding custom properties to SKILL.md is acceptable. The Agent Skill format is standardized, but extra frontmatter properties are not prohibited. In practice, many teams already extend SKILL.md with organization-specific metadata, and unknown properties are generally ignored by agents.

The additional token cost is also small. These properties usually consume around 10 to 20 tokens, so even if a project loads dozens of skills, the impact remains minimal.