| title | What Are Skills? |
|---|---|
| description | Skills are structured instructions that teach agents how to handle specific tasks. |
An agent on its own is a capable general-purpose reasoner. It can handle a lot — but it doesn't know the specifics of your workflows, your output formats, your domain expertise.
Skills fill that gap.
A skill is a set of instructions that teach an agent how to approach a specific type of task. When a skill is active, the agent follows its guidance — producing better, more consistent results for that domain than it would on its own.
Think of skills like playbooks: a hiring agent with a "candidate screening" skill knows which questions to ask, the integrations needed to validate claims, what to look for, and when to pass someone to the next round — no guesswork.
Say you often ask your agent to create Word documents. Without a skill, the agent will do its best — but it might use inconsistent formatting, forget page numbers, or structure things differently each time.
Give it the docx skill, and now it knows to:
- Use
python-docxfor document generation - Apply heading hierarchies properly
- Add page numbers and tables of contents
- Output with professional formatting every time
The agent's core intelligence stays the same. The skill just tells it how to apply that intelligence to a specific task.
A skill is a Markdown file (SKILL.md) with two parts:
- Header — A name and description that tells the agent when to use the skill
- Body — The actual instructions the agent follows
---
name: pdf-processor
description: "Use this skill whenever the user wants to work with PDF files..."
---
# PDF Processor
When the user asks to work with PDFs, follow these steps...Skills use a three-level loading system to stay efficient:
| Level | What | When |
|---|---|---|
| Metadata | Name + description (~100 words) | Always in context — used for triggering |
| SKILL.md body | Full instructions | Loaded when the skill triggers |
| Bundled resources | Scripts, references, assets | Loaded on demand |
This means skills don't bloat the agent's context until they're actually needed.
OpenSwarm has two systems that extend what agents can do. They're complementary but different:
| Skills | Integrations | |
|---|---|---|
| What they are | Markdown instruction files | External tools an agent can use |
| What they do | Teach the agent how to do something | Give the agent the ability to do something |
| Example | A skill that teaches proper Word doc formatting | A Gmail tool that lets the agent send emails |
| Storage | ~/.claude/skills/ as .md files |
Connected servers |
A skill might reference an integration in its instructions (e.g., "use the Google Sheets integration to read the data, then format it like this..."), but the skill itself is not code — it's guidance.
| Type | Description |
|---|---|
| Custom Skills | Skills you create or install yourself — stored in ~/.claude/skills/ |
| Builtin Skills | Community skills from the Anthropic Skills repository — browsable and installable from within OpenSwarm |
| AI-Generated Skills | Skills created by the Skill Builder Agent from a plain English description |
All three types work the same way once installed — the difference is just how they're authored.
Explore the skills and tools that come ready to use with OpenSwarm. Learn how to attach and configure skills on any agent. Build your own skills for any workflow or domain. Use OpenSwarm's built-in agent to generate custom skills for you.