What problem does it solve? Over time, an AI assistant's accumulated memory can fill with stale facts, contradicted claims, and unreliable categories that silently degrade future sessions. This Skill closes the feedback loop by measuring whether stored memory is actually earning its place, using signals already present in the memory files. ## Core Features & Use Cases - Citation Analysis: Classifies each memory topic file as active, stale, or never cited based on Last-cited stamps within a configurable day window (default 90). - Contradiction Scan: Surfaces facts that developers explicitly corrected, listing each with its topic file and originating session. - Rollback-Informed Confidence: Computes rollback rates per fact category and writes audit hints that lower promotion confidence for categories exceeding a 10% rollback rate. - Use Case: Run a quarterly audit to archive uncited topic files, flag contradicted facts for review, and automatically tune how aggressively future learnings get promoted into long-term memory. ## Quick Start Ask the assistant to run a memory quality audit over the last 90 days and propose archive, flag, and confidence-hint actions.
