Deep Agent Scaffold Generator

Architecture

# Deep Agent scaffold — LangChain `deepagents` (research preview; verify the API against your version).
# pip install deepagents

from deepagents import create_deep_agent
from langchain.agents.middleware.todo import TodoListMiddleware  # planning / write_todos

# ---- Tools (plain functions; the docstring + type hints are the tool's schema) ----
def internet_search(query: str) -> str:
    """Search the web for a query.

    Args:
        query:
    """
    # TODO: implement
    return "result"

# ---- Subagents (delegated, context-isolated specialists) ----
research_agent = {
    "name": "research_agent",
    "description": "Researches questions in depth.",
    "system_prompt": "You are a thorough researcher. Cite sources.",
    "tools": [internet_search]
}

agent = create_deep_agent(
    model="anthropic:claude-sonnet-4-6",
    tools=[internet_search],
    system_prompt="You are a capable deep agent. Break the task into a plan, use your tools and subagents, and keep notes in files as you go.",
    subagents=[research_agent],
    middleware=[TodoListMiddleware()]
)

# A virtual filesystem is built in; long-horizon runs offload context to files automatically.
result = agent.invoke({"messages": "Describe the task for your deep agent here."})
print(result["messages"][-1].content)