Some repos now ship agent skills: packaged instructions and helper scripts that an AI agent can load on demand. Anthropic's convention is a SKILL.md file with YAML frontmatter, usually under skills/<name>/SKILL.md or .claude/skills/<name>/SKILL.md. A repo might contain one skill or a whole collection.
This isn't the same as the agents resource. AGENTS.md tells an agent how to work on this codebase. A skill is something the project provides for agents to use elsewhere. It's closer to "this repo publishes a CLI" or "this repo publishes a GitHub Action" than to a contributing guide.
So rather than another resource entry, this probably belongs in the tools/capabilities output: detect SKILL.md files and report that the project provides agent skills, ideally listing the skill names from the frontmatter.
Other ecosystems will likely grow their own formats (OpenAI has a similar concept brewing) so the detection should leave room for more than one skill flavour.
Some repos now ship agent skills: packaged instructions and helper scripts that an AI agent can load on demand. Anthropic's convention is a
SKILL.mdfile with YAML frontmatter, usually underskills/<name>/SKILL.mdor.claude/skills/<name>/SKILL.md. A repo might contain one skill or a whole collection.This isn't the same as the
agentsresource.AGENTS.mdtells an agent how to work on this codebase. A skill is something the project provides for agents to use elsewhere. It's closer to "this repo publishes a CLI" or "this repo publishes a GitHub Action" than to a contributing guide.So rather than another resource entry, this probably belongs in the tools/capabilities output: detect
SKILL.mdfiles and report that the project provides agent skills, ideally listing the skill names from the frontmatter.Other ecosystems will likely grow their own formats (OpenAI has a similar concept brewing) so the detection should leave room for more than one skill flavour.