Skip to content
logiccloud DELEGATE

Skills and plugins

A skill is a folder with a SKILL.md (the open Agent Skills layout): instructions an agent loads when they fit the task, plus the files beside them, such as references, scripts and templates. A plugin is a Claude Code plugin: skills, commands, agents, hooks and MCP configuration in one folder with a .claude-plugin/plugin.json.

An agent lists the skills and plugins it loads. Every workspace started from the agent gets them at its next start.

Import from GitHub

Section titled “Import from GitHub”
  1. Open Skills and choose From GitHub.
  2. Enter the repository as owner/name or as its GitHub URL. A URL to a folder (…/tree/main/skills) searches only that folder. Add a branch or tag if you do not want the default branch.
  3. For a private repository, add a personal access token with read access to its contents. It is stored encrypted and never shown again.
  4. Choose Browse. The list shows every skill folder and plugin of the repository, its size, and whether it is already imported.
  5. Select what you need and choose Import.

Skills come with their whole folder. Binary files (images, archives, office documents) are left out, and the list says how many. A skill whose key is already taken in the organization is not overwritten; the result says so.

The repository becomes a source on the Sources tab.

Sync

Section titled “Sync”

Sync on a source reads the repository again:

  • A skill whose folder changed gets a new version, which becomes active. Agents that load “latest active” use it at their next start.
  • A plugin whose folder changed gets the new files.
  • Folders that did not change are left alone.

A skill with an unsaved draft is not synced; discard or activate the draft first. Add more browses the repository again with the stored token. Removing a source keeps its skills and plugins, but no longer syncs them.

What each engine gets

Section titled “What each engine gets”
Skills (with their files)Plugins
Claude Codeyesthe whole plugin
Codexyesits skills and commands
OpenCodeyesits skills, commands and agents

Scripts keep their executable mode. Everything an agent runs still goes through the egress proxy: a script cannot reach a host the agent is not granted.