What problem does it solve? Repeated AI coding mistakes and recurring fix patterns often go unaddressed because nobody systematically reviews correction signals or git history. This Skill digests pending correction signals and scans git logs to surface recurring failure patterns, then writes concrete rule-change proposals for user review instead of silently modifying anything. ## Core Features & Use Cases - Signal Digestion: Drains correction signals from .codex/evolution/signals.jsonl and converts them into actionable proposals written to proposals.md. - Git Pattern Mining: Scans git log and git diff to detect repeatedly occurring fix patterns that deserve a standing rule. - Bidirectional Rule Maintenance: Proposes both adding rules for confirmed failure modes and retiring rules that are internalized, never triggered, or duplicated. - Forced Abstraction: Strips product-specific names and terms from signals so rules capture product-agnostic general lessons, while project-specific instances go to user memory. - Use Case: After several sessions where the AI repeatedly made the same type of mistake, run this Skill to consolidate those corrections into a short list of rule proposals that the main Agent reviews with you one by one at session start. ## Quick Start Ask the AI to run the evolution-engine skill to digest pending correction signals and scan git history for recurring fix patterns, then review the proposals it writes.
