detect-framework

name	detect-framework
description	Detect Python agent frameworks from code imports and map them to agent plane executor types. Load when the user has existing agent code to integrate.

Framework Detection

When the user has existing Python code they want to integrate into agent plane, detect the framework from import statements and recommend the appropriate executor type.

Detection Procedure

1. Ask the user for the path to their agent code (or look for Python files in the current directory if filesystem access is enabled).

2. Scan Python files for import patterns. Check in this priority order:

Import pattern	Framework	Executor type
import anthropic or from anthropic + agent patterns (e.g. Agent, tool, system prompt setup)	Claude SDK	claude_sdk
import openai or from openai + agents patterns (e.g. Agent, Runner, function_tool)	OpenAI Agents SDK	agents_sdk
from langgraph or import langgraph	LangGraph	Native executor (planned — use remote for now)
from deepagents or import deepagents	DeepAgents	Native executor (planned — use remote for now)
from langchain or import langchain	LangChain	remote (native planned)
from crewai or import crewai	CrewAI	remote
from autogen or import autogen	AutoGen	remote
None of the above	Unknown	remote

3. Report what you found and recommend the executor type.

What to generate for each executor type

llm (default — no existing code)

Generate a standard agent directory:

executor:
  type: llm  # or omit entirely (llm is the default)

claude_sdk

The user's Claude SDK code runs directly. Generate config that points to their entry module:

executor:
  type: claude_sdk

agents_sdk

The user's OpenAI Agents SDK code runs directly:

executor:
  type: agents_sdk

remote

For unsupported frameworks, the agent runs behind an HTTP endpoint. Generate a FastAPI wrapper app that:

1. Imports the user's agent
2. Exposes it via the remote executor protocol
3. Lives inside the user's agent directory

executor:
  type: remote
  endpoint: http://localhost:5001

Asking about unsupported frameworks

If the user's framework is not natively supported, let them know:

• Explain that a remote executor wrapper will be generated
• Offer to show them a pre-filled GitHub issue URL requesting first-class support for their framework
• The issue URL format: https://github.com/dbczumar/agent-plane/issues/new?title=...&body=...