Skip to main content
A Skill is a packaged bundle of instructions and (optional) tools that the assistant loads on-demand. Think of it as a focused expertise the model can put on like a coat — PDF-summarizer, Stripe-query-writer, React-code-reviewer. Open the Skills manager from the composer’s + panel.
Screenshot coming. Skills manager with active and inactive skills.

​
Anatomy of a Skill

Every Skill is a ZIP that unpacks into a directory with:
  • A SKILL.md at the root — the instructions the assistant uses to know what this skill does and how to behave when it’s active.
  • Any supporting files (templates, examples, configs) referenced by SKILL.md.
  • An optional connector binding that hooks the skill to one of your connectors.

​
How they’re used

  1. Import a skill (drag-drop the ZIP). It lands in your Skills library.
  2. Activate one or more skills before sending a prompt.
  3. Active skills are surfaced to the assistant as part of the prompt — the model now “knows” how to behave like a PDF summarizer (or whatever else).
You can stack multiple active skills. The assistant blends their instructions intelligently.

​
Three flavors

SourceWhat it is
OfficialMaintained by IronLabs. Lives at skills/official/... in storage.
UploadSkills you uploaded yourself. Private to your account.
GitHubPulled from a public GitHub repo. Auto-syncs new versions.
PrivateInternal skills shared inside your workspace.

​
Tips

  • One narrow skill > one big skill. Skills compose. Three small, sharp skills beat one giant one.
  • Pair with a connector. A PDF Annotator skill paired with the Firecrawl connector gives you a serious document workflow.

​
Limits

  • Active skill count caps out at a reasonable few — the model has a context budget. Disable what you’re not using.
  • Skill SKILL.md files are the source of truth. Heavy supporting files (large PDFs, images) work, but keep them light or the skill loads slow.

Using skills

Activate, stack, disable.

Importing skills

ZIP up your own.

Connectors

Skills + connectors = workflow.