Skills

Copy page

Reusable instruction blocks you can attach to sub-agents

Overview

Skills are reusable instruction blocks that you can attach to multiple sub-agents. They govern agent behavior, reasoning, and tool usage, helping you keep shared rules, tone, and formatting consistent across the agent graph while still allowing each sub-agent to focus on its own role.

Create a skill

  1. Go to Skills in the left sidebar.
  2. Click Create skill.
  3. Fill in the required fields and save.
Create a new skill
Create a new skill

Fields

  • Name: A short identifier.
  • Description: Explains when to use this skill. Write descriptions that help the agent understand when to load the skill.
  • Content: The Markdown body, write whatever helps agents perform the task effectively.
  • Metadata (JSON): Additional properties not defined by the Agent Skills spec.
Tip
Tip

Write descriptions that include "Use when..." or "Use if..." phrases. This helps the agent decide when to load on-demand skills.

Writing effective descriptions

The description field is critical for on-demand skills. The agent uses it to decide whether to load the skill, so be specific about the scenarios where the skill applies.

Good example:

Extracts text and tables from PDF files, fills PDF forms, and merges multiple PDFs. Use when working with PDF documents or when the user mentions PDFs, forms, or document extraction.

Poor example:

Helps with PDFs.

Attach skills to sub-agents

Open an agent in the Visual Builder, select a sub-agent node, and use the Skills picker to attach one or more skills.

You can drag to reorder them to control their priority (later entries weigh more).
Skills configuration in sidepane
Skills configuration in sidepane

Always loaded and On-demand skills

Use Always loaded to include a skill in every prompt. It's off by default so skills load on demand unless you enable it.

How on-demand skills work

On-demand skills appear as an outline in the system prompt. When the agent determines a skill is relevant based on its description, it calls the built-in load_skill tool to retrieve the full content. This keeps prompts compact while still giving agents access to specialized instructions when needed.