self-improvement

You can log AI prompt errors and corrections by capturing them into structured markdown files under a .learnings directory. This creates a persistent trail of learnings, errors, and feature requests to help improve responses over time.

Tracking failed commands and missing features involves logging them into centralized markdown files like ERRORS.md and FEATURE_REQUESTS.md. This approach ensures that operational failures and knowledge gaps are recorded for cross-session reuse.

Yes, logged learnings can be promoted directly to workspace memory files such as CLAUDE.md, AGENTS.md, and .github/copilot-instructions.md. This integrates continuous improvement entries into your project-wide AI instructions.

Logging learnings with OpenClaw uses hook-driven reminders to surface past errors and corrections across different sessions. This workflow ensures that previously captured knowledge is automatically applied to new AI executions.

You should trigger a learning log entry when a command fails, a user provides a correction, a requested feature is missing, an external tool fails, or when existing knowledge is outdated. This captures actionable context for future sessions.