What
When the prompt cache breaks, Claude Code sends a usage report about it. A cache break means the model has to reprocess parts of the conversation it would otherwise have reused. That report now also records:
- whether the request came from a subagent, a helper agent that Claude Code starts to work on part of a task
- whether the session was resumed, meaning an earlier conversation was reopened
- whether that resumed session was reopened in a new process, meaning Claude Code had been closed and started again
- an outline of the messages involved
The way Claude Code works out what broke the cache has also been reworked for sessions whose earlier state was restored from disk.
Why
This only changes what Claude Code reports about itself. It does not change how Claude Code behaves or what you see.