What problem does it solve? AI assistants repeat the same mistakes across sessions because nothing persists between conversations. This Skill builds a cumulative learning memory: every time a user flags a repeated error, a project convention is discovered, or a CLI footgun is hit, it gets appended to a persistent markdown store that auto-loads at every session start. ## Core Features & Use Cases - Mistake Logging: Appends dated entries with symptom, root cause, prevention, and source to mistakes-learned.md the moment a user says "you did this again". - Four Categorized Memory Files: Separates learnings into mistakes-learned.md, project-patterns.md, korean-context.md (Korean API and UX quirks), and tool-quirks.md under ~/.claude/instincts/. - Auto-Load Bootstrap: Injects an Instincts block into ~/.claude/CLAUDE.md so all four files are referenced at every session start, with an idempotent install check. - Helper Script: scripts/append.sh appends templated entries interactively or via flags, auto-populating dates and enforcing all fields. - Use Case: A user says "λ μ΄κ±° μ λ²μλ κ·Έλ¬μμ" about grep -c exiting 1 on zero matches; the Skill immediately appends a structured entry to tool-quirks.md so the mistake never recurs. ## Quick Start Tell the assistant to record this repeated mistake in your instincts memory so it never happens again.
