Skills Tool icon

Skills Tool

Gives an AI agent access to a library of SKILL.md instruction files. The agent can list available skills, read their full content, and fetch sub-resource files.

Skills Tool

AI-generated

Summary

Exposes a library of SKILL.md instruction files to an AI agent as a single tool, with sources from Google Drive, OneDrive, a local folder, or one manually typed skill. The agent can read a skill, list its resource files, or fetch a text resource.

Inputs

  • sourceMode (required) — Where skills come from: Google Drive, OneDrive, Local Folder (default), or Manual.
  • googleDriveFolderId (required) — Shown for Google Drive. Top-level folder whose sub-folders are skills (each containing SKILL.md). Choose from a list or paste an ID. Needs a Google Drive OAuth2 credential.
  • oneDriveFolderId (required) — Shown for OneDrive. Top-level folder whose sub-folders are skills. Choose from a list or paste an item ID. Needs a Microsoft OneDrive OAuth2 credential.
  • localFolderPath (required) — Shown for Local Folder. Absolute path inside the n8n container (default /data/skills).
  • manualSkillName (required) — Shown for Manual. Name the agent uses to identify the skill.
  • manualSkillDescription — Shown for Manual. One-line description the agent reads when choosing a skill.
  • manualSkillContent (required) — Shown for Manual. Full SKILL.md text, including frontmatter and instructions.
  • allowedSkills — Shown for non-Manual sources. Limits which skills are listed in the tool description. Empty means all skills.

Output shape

Connects to an AI agent's tool input as a LangChain tool. The agent calls it with action (get_skill, list_skill_resources, or get_skill_resource), plus name and, for resources, resourcePath. get_skill returns {name, content}. list_skill_resources returns {name, resources}. get_skill_resource returns {name, resourcePath, content}.

For Drive/OneDrive/local sources, the skill list in the tool description comes from a cache populated by a separate Skills Cache Builder node. If the cache is empty, the tool throws an error telling you to run that builder first. Manual mode needs no cache. Allowed Skills only filters the advertised list, and execute does not enforce it. Resources must be text files (md, txt, sql, json, yaml, py, js, ts, sh, csv, xml, html, css). Binary files are rejected, and Manual mode does not support resources. Resource lists come from the frontmatter resources field, or else a shallow folder listing. The local folder adapter blocks path traversal. Cached content expires after 60 minutes.

Examples

Example 1: Give an agent a team playbook library stored in a mounted folder

Source=Local Folder, Skills Folder Path=/data/skills, Allowed Skills empty. Run the Skills Cache Builder first, then connect this node to the agent's tool input.

Example 2: Expose only selected skills from Google Drive

Source=Google Drive, pick the skills folder, set Allowed Skill Names to the chosen skills, and use a Google Drive OAuth2 credential.

Example 3: Quickly test one skill without external storage

Source=Manual, Skill Name=sql-query-builder, add a description, and paste SKILL.md text into Skill Content.

Links

Discussion