What problem does it solve?
Converts informal user feedback, remembered preferences, or requests to create or improve skills into actionable edits and new skill artifacts so Claude's behavior improves persistently and consistently across contexts.
Core Features & Use Cases
- Duplicate detection & consolidation: Finds repeated instructions across CLAUDE.md and skills and recommends or performs consolidation to the optimal location (user-level, project-level, or plugin-level).
- Hook-first policy for marketplaces: Detects plugin marketplace projects and prioritizes hook reason or plugin-level changes instead of user-level edits.
- Skill extraction & TDD workflow: Extracts large CLAUDE.md sections into new skills using a RED-GREEN-REFACTOR approach and updates allowlists and allowed-tools as needed.
- Safe edits & commits: Applies language-aware edits, verifies staleness via git, summarizes changes, and offers to commit user-level configuration updates.
- Use case: When a user says "remember this" or "do not add emojis to commits", run the scan, determine best location, apply the edit, and report what changed.
Quick Start
Tell Claude: /self-improvement please incorporate this feedback into CLAUDE.md and create or update the corresponding skill.