Skip to main content
Skills are reusable capability bundles that extend your agent’s behavior. Each skill is a zip file containing a SKILL.md file with YAML frontmatter. The agent receives skill names, descriptions, and file locations in its system prompt and can read skill content at runtime using the execute_bash_commands tool.

​
Prerequisite: Enable the Bash tool

For the agent to use skills at runtime, the Sandbox tool (which provides execute_bash_commands) must be enabled:
  1. Open your agent in the Agent Builder.
  2. Go to the Tools tab.
  3. Enable Sandbox (Bash / execute_bash_commands).
  4. Save the agent.
Without this, the agent will see skill metadata in its prompt but cannot read the skill files.

​
Skill package format

Each skill is distributed as a .zip file. The zip must contain a SKILL.md file at the root. SKILL.md must:
  • Start with YAML frontmatter between --- delimiters.
  • Include a name field (used in the prompt and UI).
  • Include a description field (used in the prompt and UI).
Example SKILL.md: 
---
name: My Skill Name
description: Short description of when and how to use this skill.
---

Your skill content in Markdown. The agent will read this file when the skill is used.
The zip filename (without .zip) becomes the skill folder name on the server (e.g. my-skill.zip → folder my-skill). The server parses SKILL.md to extract name and description; invalid or missing frontmatter will cause upload to fail.

​
Using the Skills tab in the UI

  1. Open the agent: Agent Framework → Agents → [your agent] (e.g. http://localhost:3000/agent-framework/agents/{agent-id}).
  2. Select the Skills tab.

​
Saved skills

  • Lists skills already attached to the agent (stored under sandbox-data/{AgentName}_{Version}/skills/).
  • Each row shows name, description, and file_location.
  • Use the delete (trash) icon to mark for deletion. Changes apply when you click Save; the server removes the skill folder and updates the agent config.

​
Upload new skills

  • Click Upload Skill Zip and choose a .zip file that contains a root SKILL.md with valid frontmatter.
  • The file is uploaded to a temp area and parsed; name and description appear under Pending Skills.
  • You can remove a pending skill with the trash icon (deletes from temp only).
  • Click Save to commit pending skills: each pending skill is moved from temp to the agent’s skills directory and added to the agent config. After save, they appear under Saved Skills.
New agents must be created first; the upload button is disabled until the agent exists.

​
How skills are used at runtime

The agent config stores a skills array of { name, description, file_location }. When building the system prompt, each skill is injected as an available skill; the prompt instructs the agent to use the execute_bash_commands tool to read the skill file (e.g. /sandbox/skills/{name}/SKILL.md). The Sandbox tool mounts the agent’s skills directory into the container, so enabling the Sandbox (Bash) tool is required for skills to be usable at runtime.