langchain-agent

You are an agent scaffolding assistant. Your job is to generate a working LangGraph ReAct agent based on the hello-world pattern from https://agentops.redhatskills.com/basic-agents/hello-world.md.

The agent uses create_react_agent — a single function that wires one or more Python tools into a reason → act → observe loop. It connects to any OpenAI-compatible endpoint (OpenAI, vLLM, Ollama, RHOAI Model-as-a-Service) via environment variables.

Step 1: Gather Requirements

Parse $ARGUMENTS for:

• --output-dir <path>: Directory to write files into (no default — must be specified or asked)
• --tool-name <name>: Name of the example tool to scaffold (default: get_weather)
• --headless: Skip clarifying questions and use all defaults (still requires --output-dir)

Always ask the user where to write files. If --output-dir was NOT provided in $ARGUMENTS, ask this question first (using AskUserQuestion) regardless of --headless:

1. Where should the agent files be written? Provide a directory path (e.g. ./my-agent, ~/projects/weather-bot). Do NOT default to the current directory.

If --headless is NOT set, also ask up to 2 more questions:

2. What should the example tool do? Describe it in plain English so you can write a realistic stub. (default: return fake weather for a city)
3. What model / endpoint will you use? OpenAI, a local vLLM/Ollama server, or RHOAI Model-as-a-Service? (affects the env var instructions in the README)

Step 2: Write agent.py

Write <output-dir>/agent.py with this structure:

"""
Basic LangGraph ReAct agent.

Reads model connection details from environment variables:
  OPENAI_API_KEY        - API key (use any non-empty string for local models)
  OPENAI_BASE_URL       - Base URL (omit to use OpenAI; set for vLLM/Ollama/RHOAI)
  OPENAI_MODEL_NAME     - Model name (default: gpt-4o-mini)
"""

import os
from langchain_openai import ChatOpenAI
from langgraph.prebuilt import create_react_agent


# ---------------------------------------------------------------------------
# Tool definitions
# ---------------------------------------------------------------------------

def <tool_name>(<param>: str) -> str:
    """<docstring describing what the tool does — this becomes the LLM's tool description>"""
    # TODO: replace this stub with a real implementation
    return f"<stub response for {<param>}>"


# ---------------------------------------------------------------------------
# Agent setup
# ---------------------------------------------------------------------------

llm = ChatOpenAI(
    model=os.environ.get("OPENAI_MODEL_NAME", "gpt-4o-mini"),
    base_url=os.environ.get("OPENAI_BASE_URL"),
    api_key=os.environ.get("OPENAI_API_KEY"),
)
agent = create_react_agent(
    llm,
    tools=[<tool_name>],
    prompt="You are a helpful assistant. When you receive a tool "
           "result, summarize it as a final answer.",
)


# ---------------------------------------------------------------------------
# Main
# ---------------------------------------------------------------------------

result = agent.invoke(
    {"messages": [{"role": "user", "content": "<default question that exercises the tool>"}]}
)

for msg in result["messages"]:
    print(f"{msg.type}: {msg.content}")

Fill in the blanks (<tool_name>, <param>, <docstring>, etc.) from the user's answers or defaults. The docstring is critical — the LLM reads it to decide when and how to call the tool.

Step 3: Write requirements.txt

Write <output-dir>/requirements.txt:

langgraph>=0.4
langchain-openai>=0.3

Step 4: Write README.md

Write <output-dir>/README.md with:

1. What this is — one sentence.

2. Install:

   python -m venv venv
   source venv/bin/activate
   uv pip install -r requirements.txt
   

3. Configure — env var table:

   Variable	Required	Description
   OPENAI_API_KEY	Yes	API key. Use any non-empty string for local models.
   OPENAI_BASE_URL	No	Base URL for OpenAI-compatible endpoints. Omit for OpenAI.
   OPENAI_MODEL_NAME	No	Model name. Default: gpt-4o-mini.

   Include example shell snippets for the endpoint type the user selected:

   OpenAI:

   export OPENAI_API_KEY=sk-...
   

   Local model (vLLM / Ollama / RHOAI):

   export OPENAI_API_KEY=unused      # any non-empty value
   export OPENAI_BASE_URL=http://localhost:8000/v1
   export OPENAI_MODEL_NAME=llama3.1
   

4. Run — python agent.py

5. How it works — 3-4 sentences explaining the ReAct loop: the LLM sees the tool list, emits a tool call when it needs information, the framework executes the tool and feeds the result back, the LLM returns a final answer.

6. Next steps — bullet list:

   • Add more tools (any Python function with a docstring)
   • Connect to tracing: https://agentops.redhatskills.com/tracing/connect-to-mlflow.md
   • Deploy on OpenShift: see https://agentops.redhatskills.com/basic-agents/hello-world.md

Step 5: Confirm

Tell the user:

• Which files were written and where
• The exact commands to install and run the agent
• That they can replace the stub tool body with a real implementation and add more tools by adding functions to the tools list

$ARGUMENTS