Finer timing breakdown for how each request to the model is built

What

Claude Code measures how long each turn takes. A turn is one round of you asking and Claude answering. These measurements are now split into smaller parts:

• Building the descriptions of the tools Claude can use (toolSchemaBuild, with tool_schema_build_detail).
• Preparing the conversation's messages (messageNormalization, with message_normalization_detail). This step now records timing marks as it runs.
• Building the request sent to the model (requestBuild, with request_build_detail).
• Time windows for the response stream and for waiting for the first reply (states).

Why

These are internal speed measurements. Nothing changes in what you see or do. They let slow spots before a request is sent be found more precisely.