Skip to content

Skill

A Skill is a behavior package the agent activates only when the user’s request is relevant to its description. It’s the Claude “skills” model: instead of putting everything in the base prompt, you encapsulate specific capabilities that turn on when needed.

It connects to the skills port of an AI Agent and you can connect several. When a skill activates, it does two things: injects its instructions into the prompt and enables its tools.

FieldTypeRequiredDescription
skill_nametextYesSkill name (e.g. billing).
descriptiontextareaYesWhen to use it. Compared against the request to decide whether to activate.
keywordstextNoKeywords (CSV) that trigger it (e.g. invoice, receipt, vat).
prompt_fragmenttextareaNoInstructions it injects when activated.
enables_toolstextNoTools (CSV of names) this skill enables.
attach_totextYes*Agent alias.

The agent activates the skill if the user’s request matches any of its keywords or words in its description. On activation:

  1. It adds the prompt_fragment to the system prompt.
  2. It enables the tools in its enables_tools.

A support agent with a skill that only applies to tax topics and enables the invoice-issuing tool.

  1. Have a Tool issue_invoice connected to the tools port.
  2. Skill:
    • skill_name: billing
    • description: “Issuing invoices and receipts, VAT queries and tax data.”
    • keywords: invoice, receipt, vat, tax
    • prompt_fragment: “When issuing an invoice, first confirm the customer’s tax details.”
    • enables_tools: issue_invoice
  3. Connect it to the skills port (pink).

Result:

  • General question → the skill does not activate and issue_invoice is not offered to the LLM.
  • The user mentions “invoice” or “VAT” → the skill activates: it injects its instruction and enables issue_invoice.