- Increase OpenAI client timeout from 15s to 60s to handle high-latency models - Add `test-llm.js` for manual verification of LLM connectivity and response times