deepagents-skills (Python)

Overview

Skills are reusable agent capabilities that provide specialized workflows and domain knowledge. They use progressive disclosure: agents only load skill content when it's relevant to the task.

How it works:

1. Match: Agent sees skill descriptions in system prompt
2. Read: If relevant, agent reads full SKILL.md using read_file
3. Execute: Agent follows instructions and accesses supporting files

Skills vs Memory

Skills	Memory (AGENTS.md)
On-demand loading	Always loaded at startup
Task-specific instructions	General preferences
Large documentation	Compact context
SKILL.md in directories	Single AGENTS.md file

Creating Skills

Skill Directory Structure

skills/
└── langgraph-docs/
    ├── SKILL.md        # Required: main skill file
    ├── examples.py     # Optional: supporting files
    └── templates/      # Optional: templates

SKILL.md Format

---
name: langgraph-docs
description: Answer questions about LangGraph using official documentation
---

# LangGraph Documentation Skill

## Overview
This skill provides access to LangGraph documentation for answering questions.

## When to Use
Use this skill when the user asks about LangGraph features, APIs, or usage.

## Instructions
1. Search documentation for relevant topics
2. Provide code examples from examples.py
3. Link to official docs

## Supporting Files
- examples.py: Common usage patterns
- templates/graph.py: Graph template

Using Skills in Agents

With FilesystemBackend

from deepagents import create_deep_agent
from deepagents.backends import FilesystemBackend
from langgraph.checkpoint.memory import MemorySaver

agent = create_deep_agent(
    backend=FilesystemBackend(root_dir=".", virtual_mode=True),
    skills=["./skills/"],  # Path to skills directory
    checkpointer=MemorySaver()
)

result = agent.invoke({
    "messages": [{
        "role": "user",
        "content": "What is LangGraph? Use the langgraph-docs skill if available."
    }]
})

With StoreBackend

from urllib.request import urlopen
from deepagents import create_deep_agent
from deepagents.backends import StoreBackend
from deepagents.backends.utils import create_file_data
from langgraph.store.memory import InMemoryStore

store = InMemoryStore()

# Load skill from URL
skill_url = "https://raw.githubusercontent.com/langchain-ai/deepagents/main/libs/cli/examples/skills/langgraph-docs/SKILL.md"
with urlopen(skill_url) as response:
    skill_content = response.read().decode('utf-8')

# Put skill in store
store.put(
    namespace=("filesystem",),
    key="/skills/langgraph-docs/SKILL.md",
    value=create_file_data(skill_content)
)

agent = create_deep_agent(
    backend=lambda rt: StoreBackend(rt),
    store=store,
    skills=["/skills/"]
)

result = agent.invoke({
    "messages": [{
        "role": "user",
        "content": "What is LangGraph?"
    }]
})

With StateBackend (In-State Files)

from deepagents import create_deep_agent
from deepagents.backends.utils import create_file_data
from langgraph.checkpoint.memory import MemorySaver

# Prepare skill content
skill_content = """---
name: python-testing
description: Best practices for Python testing with pytest
---
# Python Testing Skill
..."""

skills_files = {
    "/skills/python-testing/SKILL.md": create_file_data(skill_content)
}

agent = create_deep_agent(
    skills=["/skills/"],
    checkpointer=MemorySaver()
)

result = agent.invoke({
    "messages": [{
        "role": "user",
        "content": "How should I write tests?"
    }],
    "files": skills_files  # Seed state with skill files
})

Decision Table: When to Create Skills

Create a Skill When	Use Memory Instead
Instructions are task-specific	Context always relevant
Content is large (>500 lines)	Content is compact
Only needed occasionally	Needed every session
Multiple supporting files	Single preference
Domain-specific expertise	General preferences

Example Skills

Example 1: API Documentation Skill

---
name: fastapi-docs
description: FastAPI best practices and patterns
---

# FastAPI Documentation Skill

## Endpoints
Always use async handlers:
\`\`\`python
@app.get("/users/{user_id}")
async def get_user(user_id: int, db: AsyncSession = Depends(get_db)):
    result = await db.execute(select(User).where(User.id == user_id))
    return result.scalar_one_or_none()
\`\`\`

## Validation
Use Pydantic models for request/response validation.

## Supporting Files
See endpoints.py for complete examples.

Example 2: Testing Skill

---
name: pytest-patterns
description: Pytest testing patterns and fixtures
---

# Pytest Patterns Skill

## Fixtures
\`\`\`python
@pytest.fixture
async def db_session():
    async with AsyncSession(engine) as session:
        yield session
        await session.rollback()
\`\`\`

## Mocking
Use pytest-mock for external dependencies.

Boundaries

What Agents CAN Do

✅ Load skills on-demand when relevant
✅ Read SKILL.md and supporting files
✅ Follow skill instructions
✅ Update skills (if permitted)
✅ Create new skills

What Agents CANNOT Do

❌ Load all skills at startup (only descriptions)
❌ Change the SKILL.md frontmatter format
❌ Access skills outside configured directories
❌ Share skills across agents without proper backend

Gotchas

1. Skills Require Backend Setup

# ❌ Skills won't load without backend
agent = create_deep_agent(
    skills=["./skills/"]
)

# ✅ Provide backend
agent = create_deep_agent(
    backend=FilesystemBackend(root_dir=".", virtual_mode=True),
    skills=["./skills/"]
)

2. Frontmatter is Required

# ❌ Missing frontmatter
# My Skill
This is my skill...

# ✅ Include frontmatter
---
name: my-skill
description: What this skill does
---
# My Skill

3. Skill Descriptions Drive Discovery

# ❌ Vague description
description: Helpful skill

# ✅ Specific description
description: Python testing best practices with pytest fixtures and mocking

4. Custom Subagents Don't Inherit Skills

# Main agent skills NOT inherited by custom subagents
agent = create_deep_agent(
    skills=["/main-skills/"],
    subagents=[{"name": "helper", ...}]  # No skills
)

# ✅ Provide skills to subagent explicitly
agent = create_deep_agent(
    skills=["/main-skills/"],
    subagents=[{
        "name": "helper",
        "skills": ["/helper-skills/"],  # Explicit
        ...
    }]
)

Full Documentation

• Skills Guide
• Agent Skills Protocol
• Progressive Disclosure