Skill icon

Skill

Gives the agent a tool that reads skills (SKILL.md folders) from GitHub repositories

Skill

AI-generated

Summary

Provide a tool that reads instruction files (SKILL.md) from specified GitHub skill repositories, enabling AI models to access and use skills for handling specific request types.

Inputs

  • Authentication (required) — Choose between Access Token or OAuth2 for GitHub API authentication.
  • Skill Repositories (required) — One or more GitHub repositories (with optional branch/folder) where the skills are located, specified as e.g. owner/repo, owner/repo@branch, owner/repo:branch/folder, or full GitHub URLs. Only SKILL.md files outside hidden folders are considered skills.
  • Skills to Include (required) — Select whether to expose all skills from the repositories or only a selected subset.
  • Skill Names or IDs — When selecting specific skills, choose from the loaded skills list found in the configured repositories.
  • owner (required) — The GitHub repository owner from which to read the skill file; must match one of the configured skill repositories.
  • repository (required) — The GitHub repository name to read from; must match one of the configured skill repositories.
  • path (required) — The file path inside the skill folder to read, typically a SKILL.md file, which can be given in a shortened form if unambiguous.
  • Options — Additional options including cache TTL in minutes (how often to check GitHub for skill changes), error instructions text to send upon read failure, maximum file size in KB to read, and the tool description as seen by the model.

Output shape

A single JSON object describing the contents of the requested skill file or folder, including keys: content (file text), error (if failed), instructions (next steps).

Returns the full content of the SKILL.md or related skill instruction file from the specified GitHub skill repository path. If the file is a folder, returns entries in that folder. Supports caching with TTL, size limits, and error instructions for read failures. Designed as an AI tool output with asynchronous parallel reading for multiple skills.

Examples

Example 1: You configure one or more GitHub repos containing SKILL.md files describing how to handle specific request types (skills). You choose to expose all or a selected subset of these skills. Then for each needed skill, you specify owner, repository, and path (usually the SKILL.md file). When the workflow runs, this operation reads those files from GitHub, returning the skill instructions as JSON for use by AI tools, honoring caching and error handling options.

Authentication via Access Token; Skill Repositories e.g. 'anthropics/skills:main/skills'; Selected Skills or All Skills; For each skill, specify owner='anthropics', repository='skills', path='faq/SKILL.md'; Configure cache TTL and max file size as desired.

Links

Discussion