deep-agent-skills-system

Deep Agent Skills System

Create SKILL.md files that your Deep Agent loads contextually — only when relevant. This reduces startup tokens while keeping the agent knowledgeable.

When to Use

• When the agent needs domain knowledge loaded on-demand
• When creating reusable instructions for the agent
• When reducing context window usage
• When organizing agent capabilities into modules

Available Operations

1. Create SKILL.md files with frontmatter
2. Organize skills in directories
3. Load skills into the agent
4. Create skills with supporting files

Multi-Step Workflow

Step 1: Create Skill Directory

mkdir -p skills/deploy/
mkdir -p skills/code-review/
mkdir -p skills/database-migration/

Step 2: Write SKILL.md

# skills/deploy/SKILL.md

---
name: deploy
description: Deploy application to production
---

# Deploy Skill

## Steps
1. Run tests: `pytest --tb=short`
2. Build Docker image: `docker build -t app:latest .`
3. Push to registry: `docker push registry/app:latest`
4. Update K8s: `kubectl set image deployment/app app=registry/app:latest`
5. Verify: `kubectl rollout status deployment/app`

## Rollback
If deploy fails: `kubectl rollout undo deployment/app`

Step 3: Add Supporting Files

skills/deploy/
├── SKILL.md              # Main instructions
├── references/
│   └── k8s-commands.md   # Reference doc loaded when needed
└── scripts/
    └── healthcheck.sh    # Script the agent can execute

Step 4: Load Skills into Agent

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

agent = create_deep_agent(
    model="anthropic:claude-sonnet-4-6",
    backend=FilesystemBackend(root_dir=".", virtual_mode=True),
    skills=["./skills/"],  # Directory containing skill folders
    checkpointer=MemorySaver()  # Required for skills
)

Step 5: Load Skills via Store (No Filesystem)

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()

# Inject skill content into store
skill_content = open("skills/deploy/SKILL.md").read()
store.put(
    namespace=("filesystem",),
    key="/skills/deploy/SKILL.md",
    value=create_file_data(skill_content)
)

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