Agent Skills

A Skill packages reusable, task-specific instructions and resources that can be shared across Visual and Code Agents. Use Skills to give Agents specialized knowledge or workflows without repeating that content in every Agent.

Skills use progressive discovery, with the Agent initially only receiving the skill name, description, and optional metadata. The Agent then decides autonomously when and whether to load the skill, and when or whether to load additional resources, keeping the context focused on the task at hand.

Using Skills

Add Skills to a Simple Visual Agent or an Agentic Loop block when they need reusable guidance for a task. Code Agents can use Skills through the Skill Loader API.

You can publish an Agent Skill to a Catalog Collection to make it easier for other users to discover and reuse.

Each Skill contains the following fields:

  • Name (Skill ID): Used by the agent to identify the skill. (Must use lowercase alphanumeric characters and single hyphens only. Must start and end with an alphanumeric character. Max 64 characters.)

  • Description: Describes the skill’s purpose and usage. The agent uses this to determine whether the Skill should be loaded. (Max 1024 characters.)

  • Metadata (optional): Key/value pairs used by the agent as additional information

  • Instructions: Loaded only when the Agent decides the Skill is relevant. Provides detailed guidance for the agent to follow, such as workflows, rules, assumptions, and response format. Supports Markdown.

  • Resources (optional): Additional text and image files that can be referenced in the instructions. Only loaded when relevant.

See the Agent Skills format for more information about writing SKILL.md files.