self-improvement

To log AI agent command failures for continuous improvement, capture the failed command, error message, and execution context into structured markdown files. This records unexpected behaviors and exceptions to prevent recurring mistakes during development workflows.

The best way to capture user corrections and knowledge gaps in AI workflows is logging the corrected output and missing knowledge context into structured markdown files. This records discovered best practices to build a knowledge base for future agent improvements.

To promote logged learnings to permanent project memory like CLAUDE.md, elevate captured corrections and best practices from error logs into the project's core configuration files. This ensures AI agents persistently reference operational improvements across sessions.

To extract recurring learnings into reusable skills, analyze aggregated error logs and user corrections to identify repeated patterns. Mechanisms within the self-improvement workflow package these recurring resolutions into new, standalone skills for future deployment.

Yes, you can track feature requests that don't exist yet by documenting user-requested capabilities and missing functionalities within structured markdown logs. This captures operational knowledge gaps alongside standard error logging and continuous improvement data.