Analyzes and integrates user corrections into project-specific rules to evolve Claude's behavior autonomously.
The Reflect skill is a specialized component of the rl-anything pipeline designed to bridge the gap between user feedback and project documentation. It scans interaction logs for user-provided corrections and provides an interactive interface to review, approve, or refine these changes. By routing insights to the appropriate memory layers—such as CLAUDE.md, custom rules, or an auto-memory directory—it ensures that Claude learns project-specific nuances and avoids repeating past mistakes. This skill is essential for teams looking to maintain a dynamic, self-evolving knowledge base directly within their codebase.
主要功能
01Automatic promotion of frequent corrections to long-term project rules
02Interactive review and routing of user corrections to memory layers
031 GitHub stars
04Duplicate detection using semantic validation and SHA-256 hashing
05Dry-run mode for safe analysis of proposed documentation updates
06Confidence-based filtering with automatic application of high-reliability rules
使用场景
01Maintaining evolving project coding standards without manual documentation
02Automating the 'learning' phase of an AI development workflow
03Filtering out false-positive corrections to keep project rules clean and relevant
