Skip to content

Skills

Skills are extensions that give agents specialized capabilities sourced from GitHub repositories. They let you expand what your agents can do without modifying their core instructions, from interacting with external APIs to following project-specific coding standards.

How Skills Work

Under the hood, skills are installed using the skills CLI: 

npx skills add <github_repo_url> --skill <skill_name> --agent claude-code -y

This command fetches the skill definition from a GitHub repository and installs it into the agent's environment. Once installed, the skill becomes part of the agent's available capabilities and the agent can use it automatically when relevant.

Types of Skills

Agent Skills (Worker-Specific)

Agent skills are installed on a specific worker and are only available to that worker. Use these when a capability is relevant to just one role.

For example, you might install a Kubernetes skill only on your "DevOps Engineer" worker, or a React testing skill only on your "Frontend Developer" worker.

Global Skills (Leader)

Global skills are installed on the leader agent and are automatically shared with all agents in the team. They appear in every worker's configuration, making them available across the entire team.

Use global skills for capabilities that every agent should have, such as project conventions, documentation standards, or shared tooling integrations.

Tip: In the UI, the leader agent is labeled with "(global)" to indicate that skills installed on it are shared with all team members.

Adding Skills During Team Creation

When creating a new team, you can attach skills in Step 2 of the creation wizard:

  1. Leader skills (global): In the leader configuration section, add skills by providing the GitHub repository URL and skill name. These will be available to all agents.
  2. Worker skills: In each worker's configuration section, add skills specific to that worker's role. These will only be available to that particular worker.

Each skill entry requires:

  • Repository URL: The GitHub repository containing the skill (e.g., https://github.com/anthropics/skills)
  • Skill name: The name of the skill within that repository (e.g., create-react-app)

Finding Skills

You can discover available skills and find the exact repository URL and skill name you need at skills.sh. Browse or search the directory to find skills relevant to your agents' roles, then copy the GitHub URL and skill name directly into AgentCrew.

Adding Skills to Running Teams

You can also add skills to teams that are already running:

  1. Open the team's chat view.
  2. Click the Settings icon to open the team settings modal.
  3. Select any agent from the list:
    • Select the leader (global) to add skills shared with all agents.
    • Select a specific worker to add skills only for that worker.
  4. Add the new skill with its repository URL and skill name.
  5. Save the changes. The skill will be installed on the next agent restart.

Storage and Discovery

Skills are organized in two directories inside the agent container:

Directory Purpose
.agents/skills/ Contains the actual skill files downloaded from GitHub repositories.
.claude/skills/ Contains symlinks pointing to .agents/skills/. This is how the AI agent discovers and loads installed skills.

This two-directory structure separates skill storage from skill discovery. The .agents/skills/ directory holds the source of truth, while .claude/skills/ provides the interface that the agent expects. 

/workspace/
  .agents/
    skills/
      create-react-app/      # Actual skill files
      kubernetes-ops/         # Actual skill files
  .claude/
    skills/
      create-react-app → ../../.agents/skills/create-react-app   # Symlink
      kubernetes-ops   → ../../.agents/skills/kubernetes-ops      # Symlink
    agents/
      frontend-dev.md        # Worker agent definition
      devops-engineer.md     # Worker agent definition
    CLAUDE.md                # Leader instructions

Next Steps

  • Architecture: Learn how agent containers are structured and how skills fit into the runtime.
  • Configuration: Review all available settings for customizing your setup.