Skills

Add specialized capabilities to your AI agent using the agentskills.io specification.

Overview

Skills give your AI agent specialized knowledge and workflows. They follow the open agentskills.io specification using SKILL.md files with YAML frontmatter.

When a user’s request matches a skill’s purpose, the agent automatically activates it to get detailed instructions.

Adding a Skill

Skills are added through the agent detail page:

1. Go to Agents and click an agent
2. On the Skills tab, click Add New Skill
3. Choose a source:
   • Custom - Write SKILL.md content directly
   • URL - Import from a GitHub raw URL

The skill is added to the project pool and assigned to the agent. Other agents can then assign the same skill from their Skills tab.

See Agents for more on resource assignment.

SKILL.md Format

---
name: pdf-processing
description: Extracts text and tables from PDF files, fills PDF forms, and merges multiple PDFs.
metadata:
  version: "1.0.0"
  author: "Astralform"
---

# PDF Processing

Instructions for the AI agent on how to handle PDF files...

## Extracting Text

1. First, check if the file is a valid PDF
2. Use the appropriate extraction method based on content type
...

Required Fields

Field	Description
name	Unique identifier (lowercase, hyphens allowed)
description	Short description shown to the agent (~100 tokens)

Optional Fields

Field	Description
runtime	Execution environment: workspace provisions a sandbox automatically
metadata.version	Semantic version (e.g., “1.0.0”)
metadata.author	Skill author name

Auto-Capsule

Skills with runtime: workspace in their frontmatter automatically provision a sandbox for their agent (auto-capsule). No manual sandbox configuration needed.

When an agent with non-system skills processes a request, an isolated Capsule sandbox (Firecracker microVM) is automatically created. The agent can then execute code, run shell commands, and manage files inside the sandbox.

See Capsule Sandboxes for details on sandbox templates and management.

How Skills Work

1. Discovery - At agent startup, skill metadata (name + description) is loaded
2. System Prompt - Available skills are listed in the agent’s context
3. Activation - Agent calls activate_skill when a request matches
4. Instructions - Full SKILL.md content is provided to the agent

User: "Extract the tables from this PDF"

Agent: [Sees pdf-processing skill matches]
       [Calls activate_skill("pdf-processing")]
       [Receives full instructions]
       [Follows the skill's workflow]

Managing Skills

Enable/Disable

Toggle skills on/off without deleting them. Disabled skills are not shown to agents.

Refresh URL Skills

For skills imported from URLs:

1. Click Refresh to re-fetch the latest content
2. Review changes in the skill editor
3. Content updates automatically

Delete

1. Click the Delete button
2. Confirm deletion

Deleting a skill removes it permanently. Consider disabling instead.

Best Practices

Do	Don’t
Write clear, actionable instructions	Include irrelevant background
Keep descriptions under 100 tokens	Write essay-length descriptions
Use step-by-step workflows	Assume agent knows procedures
Test skills with real requests	Deploy untested skills

Example Skills

Customer Support

---
name: support-agent
description: Handles customer inquiries, processes refunds, and escalates issues following company policies.
---

# Customer Support

## Greeting
Always greet the customer warmly and ask how you can help.

## Common Issues
- Password reset: Guide through self-service portal
- Billing questions: Check account status first
- Refund requests: Follow the refund policy below
...

Code Review

---
name: code-review
description: Reviews code for bugs, security issues, and style violations using project guidelines.
---

# Code Review Guidelines

## Security Checks
1. Check for SQL injection vulnerabilities
2. Verify input sanitization
3. Review authentication logic
...

Next Steps

Agents

Assign skills to specific agents

MCP Servers

Connect external tools and databases