What
Compaction is when Claude Code shrinks a long conversation into a summary so it fits the model's context window. Sometimes that summary is prepared in advance. Before reusing it, Claude Code now checks that it still covers the conversation and still fits:
- Reuse can now be refused with named reasons, including
rows_unreadableandwould_not_fit. - Loading a prepared summary back from disk no longer rejects it for
grew_too_much. It now checksmodelMatch,preservedUuidCountandrows_unreadableinstead. - The discard log now includes token sizes, and saved summaries record
preservedMessageCountandpreservedUuidCount.
The related gate tengu_sepia_moth was off for this site's account and for the anonymous baseline in earlier readings, but no reading has been taken under this release.
Why
Auto-compact should be less likely to swap in a prepared summary that is out of date or that would push the conversation over its size limit.
tengu_sepia_moth Off in both readingsThe flag server returned off for the account this site reads and for the anonymous baseline. A reading of off cannot rule out a rollout these two readings sit outside of.
This account: off · anonymous baseline: off · compiled default in v2.1.288: off
These values were read against a different version of Claude Code, so treat them as the nearest reading available instead of one taken on this release.
Read once, for one account on one subscription tier, against v2.1.288. It isn't a statement about your account. What a flag value here can and cannot tell you
It is not clear whether preparing summaries in advance runs for anyone, since its setting is only shown when a server-side switch allows it.