Helper requests to a denied model now use your main model instead

Besides your main conversation, Claude Code makes small helper requests to a model, for jobs such as generating titles. If the model a helper request asks for is blocked by the deniedModels policy, the request now goes to the model running your main conversation, and a debug line records the switch. Previously the requested model was always used.

If the helper asked for extended thinking to be off and the substitute model cannot turn it off, thinking is left at the model's default and the response length limit is raised by 2048 tokens.