self-improvement

You log AI coding learnings and errors by writing them to Markdown files in a .learnings directory, separating corrections into LEARNINGS.md and command failures into ERRORS.md to capture knowledge gaps across sessions.

To promote learnings into project memory, you move broadly applicable corrections from .learnings Markdown files into instruction files such as CLAUDE.md, AGENTS.md, or .github/copilot-instructions.md.

Yes, capturing cross-session corrections requires OpenClaw workspace integration and a designated .learnings directory structure to store logs for later processing and reuse.

Yes, you can extract reusable skills from logged errors and corrections through optional OpenClaw integration, which processes accumulated learnings to generate new skills for future tasks.

Using Markdown files for logging AI coding errors limits automated querying and requires manual promotion of broadly applicable learnings into instruction files to ensure AI agents read and apply the corrections.