Your agent loops forever. It is probably tool_choice.

Forcing `tool_choice` persists across model calls, so the model is told to call a tool again after every result.

The fix merged into the OpenAI Agents SDK resets `tool_choice` to `auto` after each tool run.

A plain step counter cannot tell a six-step workflow from a six-step loop. Repeat detection on tool name plus arguments can.

Read the full post

Read the full post