소개
The Reflect skill implements 'Loop 1.5' (per-session micro-learning) to solve the problem of recurring LLM mistakes. It scans conversation history for learning signals like user corrections and approvals, classifies them using VERIX-aligned confidence levels, and persists them directly into human-readable SKILL.md files. By bridging the gap between execution and meta-optimization, it ensures Claude evolves with every interaction, transforming ephemeral session insights into permanent knowledge that survives across new conversations and projects.
