What problem does it solve? After running /insights, you get a list of suggestions for new skills, CLAUDE.md additions, and hook configurations, but turning each one into a real artifact is manual and error-prone. This Skill reads the latest /insights output, walks you through each candidate one at a time, and commits only the ones you approve. ## Core Features & Use Cases - Candidate triage: Parses the suggestions buckets (claude_md_additions, features_to_try, usage_patterns) and maps each to the right destination: a new SKILL.md, a settings.json hook entry, or a CLAUDE.md section. - Per-candidate approval loop: Every candidate is presented individually with options to codify, preview the full draft, skip, or stop, so nothing is written to disk without explicit consent. - Delegated authoring: Skill creation is handed to skill-creator and hook configuration to update-config, with staging in /tmp and collision checks before any commit. - Use Case: After a week of Claude Code usage, run /insights, then invoke this Skill to review five suggested improvements, approve two as new personal skills, add one CLAUDE.md rule, and skip the rest. ## Quick Start Ask the assistant to act on the latest insights output and codify the approved suggestions.
