claude-meta

claude-meta turns your CLAUDE.md documentation into a learning system that improves with every mistake Claude makes. It combines two core ideas: meta-rules that teach Claude how to write effective rules, and a single reflection prompt that transforms errors into permanent learnings. When Claude encounters a mistake, the system triggers reflection, abstraction, and generalization—automatically updating CLAUDE.md with new guidelines that follow established quality standards. This approach lets AI handle the execution work of documentation while you focus on critical thinking and pattern spotting. Developers working with Claude Code or Cursor benefit from compounding improvements as the knowledge base grows more sophisticated and consistent over time.

Start with the CLAUDE_TEMPLATE.md provided in the repository. When Claude makes a mistake during a session, use the reflection prompt: 'Reflect on this mistake. Abstract and generalize the learning. Write it to CLAUDE.md.' Claude will automatically follow the meta-rules structure to update your documentation consistently.