self-improvement

You can track errors and learnings across sessions by automating structured logging to Markdown files. The system records entries in LEARNINGS.md, ERRORS.md, and FEATURE_REQUESTS.md under a .learnings directory to support continuous improvement.

Logging API failures and tool errors in Markdown is best handled by writing structured entries to ERRORS.md. This automated approach captures tool failures consistently, ensuring errors are documented for future reference and correction.

To create a learning log, initialize a .learnings directory and use templates to log insights to LEARNINGS.md. This captures corrections, prompts, and outdated knowledge automatically for continuous self-improvement.

Yes, you can promote high-value learnings into project memory. The system moves important entries from standard logs into CLAUDE.md and AGENTS.md, integrating them directly into agent resources for future sessions.

You do not strictly need OpenClaw hooks to automate logging after tool failures, but integrating with them allows the system to automatically prompt for logging immediately after prompts execute or tools fail.