What problem does it solve? Corrections given during a conversation are often lost after the session ends, causing the same mistakes to repeat. This Skill turns user corrections and recurring lessons into durable updates across agent memory, instruction files, skills, AGENTS.md, hooks, docs, evals, and tests. ## Core Features & Use Cases - Correction capture: Identifies the exact correction or new requirement from the active conversation and decides whether it generalizes into a reusable rule. - Persistent artifact selection: Chooses the smallest authoritative surface, from agent memory and .agents/MEMORY.md to instruction files, custom agents, skills, AGENTS.md, and hooks. - Regression guarding: Adds or adjusts evals and tests when the lesson is objective enough to verify automatically. - Use Case: After a user corrects you for running Python outside the repo's .venv, update the relevant instruction file and validation guidance so future runs use python -m pytest -q correctly. ## Quick Start Ask the agent to capture the correction you just gave and update the right instructions, memory file, or tests so the mistake does not happen again.
