LangGraph Agent Scaffold Generator

Generate production-ready scaffolds for AI agent projects using LangGraph and LangChain.

Workflow

Step 1: Gather Project Information

Required input from user:

• •project_name: Name of the project (kebab-case, e.g., my-agent-project)

Optional inputs:

• •agent_type: Type of agent to scaffold (default: simple)
  • •simple - Basic agent with tools
  • •react - ReAct reasoning agent
  • •rag - Retrieval-Augmented Generation agent
  • •modular - Modular agent with nodes/routes structure (customizable for any domain)
  • •orchestrator - Multi-agent coordinator
  • •evaluator - Generator + Evaluator pattern

If project name not provided, ask: "What name would you like for your agent project? (use kebab-case, e.g., my-agent-project)"

Step 2: Generate Project Structure

Create the following directory structure:

{project_name}/
├── .env.example
├── .python-version
├── pyproject.toml
├── langgraph.json
├── docker-compose.yml
├── justfile
├── README.md
├── src/
│   ├── __init__.py
│   ├── agents/
│   │   ├── __init__.py
│   │   └── {agent files based on type}
│   └── api/
│       ├── __init__.py
│       ├── main.py
│       └── db.py
└── notebooks/
    └── 01-getting-started.ipynb

Step 3: Generate Files

Use templates from references/templates.md for file contents.

For agent-specific patterns, consult references/agent-patterns.md.

Step 4: Generate README.md

Create comprehensive documentation following the template in references/readme-template.md.

File Generation Order

1. •Configuration files (.python-version, pyproject.toml, .env.example)
2. •Development tools (justfile, docker-compose.yml, langgraph.json)
3. •Source structure (src/__init__.py, src/agents/__init__.py, src/api/__init__.py)
4. •Agent files (based on selected type)
5. •API files (main.py, db.py)
6. •Notebook template
7. •README.md

Code Style Rules

Python Standards

• •Python 3.12+ required
• •Type hints mandatory on all functions
• •Docstrings required for public methods
• •PEP 8 naming: snake_case functions/variables, PascalCase classes

Import Order

# Built-in
from typing import TypedDict, Literal

# External - LangGraph
from langgraph.graph import StateGraph, START, END

# External - LangChain
from langchain.chat_models import init_chat_model

# External - Others
from pydantic import BaseModel, Field

# Internal
from agents.support.state import State

State Definition

Always inherit from MessagesState:

from langgraph.graph import MessagesState

class State(MessagesState):
    customer_name: str
    extracted_data: dict

Node Functions

Return partial state dictionaries:

def my_node(state: State) -> dict:
    return {"messages": [ai_message], "field": value}

Tools

Use decorator with description:

@tool("tool_name", description="Clear description")
def my_tool(param: str) -> str:
    """Detailed docstring."""
    return result

Prompts

Use Jinja2 templates:

template = """\
{% if name %}Hello {{ name }}{% endif %}
"""
prompt = PromptTemplate.from_template(template, template_format="jinja2")

Validation Checklist

After generation, verify:

• • All __init__.py files created
• • Type hints on all function signatures
• • Proper import organization
• • States inherit from MessagesState
• • Nodes return dictionaries
• • Tools have descriptions and docstrings
• • .env.example contains required variables
• • README.md includes setup instructions