What problem does it solve? Persistent file-based memory (MEMORY.md plus per-fact markdown files) accumulates cruft, duplicates, stale index hooks, and untracked action items over time, and naive cleanup risks deleting valuable lessons and decisions. This Skill provides a disciplined playbook for curating that memory corpus while preserving every fact that still matters. ## Core Features & Use Cases - Structured housekeeping: Groups the memory index by kind, trims bloated index hooks, fixes stale hooks that drifted from their files, and verifies 1:1 index-to-file integrity. - Zero-loss deduplication: Removes only truly dead content after verifying it lives elsewhere, with verbatim-reorder scripts and git-diff proofs that nothing was lost. - Type-routing of facts: Routes each fact to its correct home — domain skill, canon docs, memory, or playbook — and migrates open action items into the 00_07 tracker with back-pointers. - Curation trap catalog: Ships a traps.md companion documenting 19 measured pitfalls (scout undercounting, bilingual dedup blindness, false homelessness) to consult before any consolidation verdict. - Use Case: When the memory index has grown bloated and you suspect duplicated rules across memory files, skills, and docs, invoke this Skill to run the gate audit, consolidate scattered classes into single homes, and report honestly what was cleaned versus deliberately kept. ## Quick Start Use the memory-maintenance skill to audit my memory corpus, remove duplicates, fix stale index hooks, and make sure every open action item is tracked in 00_07.
