Automates the review and integration of user corrections into persistent project memory and custom rules.
The Reflect skill is a core component of the rl-anything pipeline, designed to help Claude autonomously evolve by learning from its own mistakes and user feedback. It analyzes correction logs (corrections.jsonl) and provides an interactive interface for developers to approve, edit, or reject proposed memory updates. By intelligently routing feedback to the appropriate layer—such as CLAUDE.md, specific rule files, or auto-memory—it ensures that project-specific knowledge is codified, preventing repetitive errors and refining the AI's behavior over time.
主要功能
01Automated routing to global or local memory layers based on context
02Configurable confidence thresholds for automated rule application
03Promotion of frequent patterns into structured auto-memory candidates
041 GitHub stars
05Interactive review of pending corrections with semantic deduplication
06Dry-run mode for analyzing feedback without modifying project files
使用场景
01Scaling AI knowledge across large codebases through automated memory evolution
02Codifying developer feedback into persistent project-specific coding standards
03Filtering and managing false-positive AI corrections to maintain rule quality
