What
The prompt cache lets the model reuse parts of a conversation it has already processed instead of reading them again. Claude Code marks where cacheable parts end.
For a fork, a request that branches off an existing conversation, Claude Code can now place an extra marker just before the reply where the branch diverges. When it does, it skips the older way of pinning the fork point.
Why
This changes which part of a forked request can be reused from the cache. Whether it saves anything in practice depends on which requests use it.
Something disagreesSomething we can check disagrees with this entry, or the writer said they could not settle it.
The writer flagged doubt
It is not clear which requests supply the point of divergence, so how often this applies is unknown.