feat: Native framework integration for LangChain, CrewAI, AutoGen, LlamaIndex

[ryandao]

Summary

• Auto-detection: FrameworkDetector inspects sys.modules and importlib to detect which agent frameworks (LangChain, CrewAI, AutoGen, LlamaIndex) are installed and active at runtime
• Adapter pattern: Each framework gets a dedicated adapter that monkey-patches core execution methods to emit lifecycle events — no @agent decorator needed
• Simple API: Users call agentq_sdk.auto_integrate() (one-liner) or use AgentQ() as a context manager to activate framework integration
• Clean unpatch: All adapters store direct class references for reliable restoration of original methods

Architecture

agentq_sdk/
├── __init__.py          # Public API surface
├── core.py              # AgentQ class + auto_integrate()
├── detection.py         # Framework detection engine
├── registry.py          # Singleton adapter registry
└── adapters/
    ├── base.py          # BaseAdapter ABC + event system
    ├── langchain.py     # AgentExecutor.invoke + LCEL RunnableSequence
    ├── crewai.py        # Crew.kickoff + Agent.execute_task
    ├── autogen.py       # ConversableAgent.initiate_chat + generate_reply
    └── llamaindex.py    # ReActAgent/AgentRunner chat + query

Test plan

• 52 unit tests covering detection, registry, core module, and all 4 adapters
• Mock-based testing (frameworks not required in test env)
• Patch/unpatch roundtrip verified for each adapter
• Error event emission tested (exceptions propagated, events emitted)
• Adapter interface consistency checked via parametrized tests

🤖 Generated with Claude Code