memory-flush

To curate memory candidates, review pending entries in .claude/memory, verify provenance, replace stable keys, promote keepers into canonical files, and reset the pending queue. This prevents duplicate or stale session artifacts from accumulating.

Cleaning up pending memory and backlog items requires a closure sweep to surface prose-based closure signals, manage aged entries with user confirmation, and write verified keepers into canonical memory files before resetting the queue.

You need a memory curation workflow when auto-extracted candidates accumulate noise or duplicates, typically at session-start or ad-hoc intervals, to promote durable discoveries and decisions into canonical notes and clear the pending queue.

Yes, memory curation supports both session-start cleanup and ad-hoc workflows. It reviews pending questions, backlog items, and canonical memory entries, applying provenance verification and stable-key replacement before resetting the pending queue.

Closure and decay handling automatically sweeps closed entries, surfaces prose-based closure signals, and identifies stale or aged items. It manages these artifacts by requesting user confirmation before removing or updating canonical memory files.

Repository memory files accumulate duplicates and noise when auto-extracted candidates are not curated. Running a curation workflow verifies provenance, replaces stable keys, promotes valid entries into canonical files, and resets the pending queue to prevent buildup.