self-improvement

To log errors and learnings for continuous improvement, you record command failures, corrections, and best practices into centralized Markdown files like .learnings/LEARNINGS.md and .learnings/ERRORS.md. This captures context to aid debugging and knowledge retention across sessions.

Promoting learnings to project memory files like CLAUDE.md involves extracting broadly applicable knowledge from local logs and writing it into centralized instruction files such as CLAUDE.md, AGENTS.md, or copilot-instructions.md to guide future sessions.

Error logging in Markdown captures command failures and missing capabilities by recording the failure context in .learnings/ERRORS.md. This structured log enables debugging and ensures recurring external tool failures or knowledge gaps are tracked for later correction.

Yes, structured learning logs document outdated knowledge and better approaches by logging corrections into .learnings/LEARNINGS.md. High-value entries are then promoted to project memory or OpenClaw workflows to ensure improved approaches persist across sessions.

No, OpenClaw setup is not required to use the continuous improvement logging workflow. While it provides optional OpenClaw setup hints for working with memory and sessions, the core functionality of logging to Markdown files and promoting learnings operates independently.

You should not promote a learning entry to AGENTS.md if it lacks broad applicability. Entries addressing isolated command failures or specific debugging context should remain in local .learnings/ files rather than being promoted to centralized project memory instructions.