self-improvement

To track command failures and user corrections for continuous improvement, you log learnings and errors to a central repository under .learnings. Entries are categorized as corrections, knowledge gaps, or best practices to reduce repeated mistakes.

Maintaining a learning log across agents requires saving concise entries to .learnings/LEARNINGS.md. This central repository preserves shared error tracking and knowledge, enabling faster iteration and continuous improvement across workflows.

Promoting high-value learnings to project memory files like CLAUDE.md uses optional promotion pathways. When logged patterns in .learnings become broadly applicable, they are elevated to workspace guidance or agent memory files.

You need to save error tracking logs in a workspace when commands fail, user corrections occur, or new capabilities are identified. This ensures continuous improvement by preserving corrections and best practices for future tasks.

This continuous improvement approach works without external dependencies because it relies entirely on maintaining dedicated log files under .learnings. It organizes learnings natively within any workspace environment.