Skip to content

Skills

Skills are reusable knowledge and workflows that extend what the agent can do. Think of them as plug-in expertise — instead of explaining your deploy process every session, install a skill once and the agent knows it forever.

A skill is a directory with a SKILL.md file that describes:

  • When to activate — the trigger or context that makes this skill relevant.
  • Instructions — step-by-step procedures the agent follows.
  • Knowledge — reference docs, patterns, or constraints the agent should know.

Skills can cover anything: deployment checklists, coding standards, framework patterns, testing procedures, API integration guides, or domain-specific knowledge.

The Skills panel in the workspace shows all installed skills and lets you manage them without leaving the editor:

  • Toggle on/off — enable or disable a skill per session.
  • Install new skills — paste a repo URL, npm package name, or local path.
  • View details — see the skill’s name, description, and status.

You can install skills from multiple sources:

SourceExample
GitHub repogithub.com/your-org/deploy-skill
npm package@your-org/skill-tailwind-v4
Local path.agents/skills/my-custom-skill
URLAny URL pointing to a skill directory
skills.shBrowse and install community skills

Once installed, the skill appears in the Skills panel and is available to the agent immediately.

The simplest way to install a skill is to paste a skills.sh URL in the chat. The demo below shows a user pasting an SEO Audit skill URL — the agent runs the load_skill tool to install it, then confirms. The skill appears in the Skills panel:

The agent can also load skills on demand during a session. When it encounters a task that matches a skill’s activation criteria, it loads the skill automatically and follows its instructions. You’ll see a LoadSkill tool call in the chat when this happens.

Without skills, every session starts from zero — you have to re-explain your conventions, processes, and preferences. Skills give you:

  • Consistency — the agent follows the same process every time.
  • Reusability — write a procedure once, use it across sessions and teams.
  • Extensibility — add new capabilities without changing the agent’s core.
  • Shareability — distribute skills as packages or repos across your organization.