self-improvement

To log command failures and corrections for continuous learning, you capture errors, insights, and user corrections directly into markdown files. This centralized logging enables easy review, sharing, and reuse of knowledge across different projects and sessions.

The best way to share learnings across sessions is to promote valuable corrections and insights into project memory or agent-facing guidelines like CLAUDE.md and AGENTS.md. This ensures persistent knowledge sharing throughout your workflow.

To track errors and feature requests, you initialize a .learnings directory at your workspace root containing LEARNINGS.md, ERRORS.md, and FEATURE_REQUESTS.md files. You then log entries using a defined markdown format.

Yes, you can promote logged insights into agent guidance files like CLAUDE.md, AGENTS.md, or copilot-instructions. This elevates valuable corrections and best practices into permanent agent-facing guidelines for future operations.

OpenClaw supports automated hooks and prompts that trigger logging workflows when a requested capability is missing or an operation fails. This automates the capture of knowledge gaps for continuous improvement.