self-improvement

To capture agent errors and corrections for continuous improvement, log failures and user corrections into structured markdown files within a .learnings/ directory. This organizes knowledge gaps and best practices for future review and reuse.

Workspace memory promotion is the process of elevating broadly applicable learnings from local error logs into persistent workspace files like AGENTS.md, SOUL.md, and TOOLS.md to improve agent consistency across sessions.

Yes, you can automate error logging when commands fail by using provided activator and error-detector scripts. These automation hooks flag command failures and prompt evaluation after tasks to ensure mistakes are logged.

The best way to track feature requests and knowledge gaps in markdown is to use a deterministic logging format with YAML frontmatter. This routes entries into dedicated files like FEATURE_REQUESTS.md and ERRORS.md for structured review.

Your agent repeats mistakes because local learnings are not automatically applied to future sessions. You must review the logs and promote broadly applicable corrections into workspace memory files like CLAUDE.md to ensure persistent behavioral changes.