Memorize

This is the curation phase of Agentic Context Engineering. After you run reflections and critiques, memorize consolidates those learnings into CLAUDE.md as concrete, actionable bullets. It's designed to fight context collapse by enforcing specificity: you get imperative rules, API gotchas, anti-patterns with evidence, and verification checklists instead of vague summaries. The skill walks you through harvesting insights from recent work, categorizing by impact, then placing them in structured sections like Architecture Decisions or Hard Rules. Think of it as building a living playbook that actually helps future sessions instead of just documenting what happened. Works best when you feed it real reflection outputs rather than raw chat logs.