Render an agent event stream incrementally and handle completion states

Goal

Show progress while preserving event order and finish correctly on completion, failure, or cancellation.

Steps

1. Identify event categories such as text deltas, tool calls, handoffs, and run completion. [openai-streaming] [openai-running]
2. Start with result=Runner.run_streamed(agent, input) and consume async for event in result.stream_events(). [openai-streaming] [openai-running]
3. Branch on event.type for raw_response_event versus run_item_stream_event; only text deltas are user-facing text. [openai-streaming] [openai-running]
4. Close the connection and set a final state on stream end, timeout, or user cancellation. [openai-streaming] [openai-running]
5. Test a mock stream with text, tool events, and failure; ensure the UI does not expose hidden arguments. [openai-streaming] [openai-running]

How to verify

Event tests should cover normal end and mid-stream failure; tool events must not be appended to user-visible answers.

Troubleshooting

If output is duplicated, inspect event ordering and ensure deltas are not appended twice after reconnect.