self-improvement

You can log agent learnings and errors automatically by using hook-based scripts to capture non-obvious issues and corrections into centralized markdown files, building a persistent shared memory across future sessions.

Capturing recurring workflow patterns as shared memory is best achieved by logging learnings, user feedback, and errors into centralized files like LEARNINGS.md and ERRORS.md, then promoting useful entries into broader agent context files.

To log feature requests and corrections during OpenClaw agent workflows, use the provided helper scripts to detect and document recurring problems and user feedback automatically into dedicated markdown tracking files.

Yes, logged learnings can be promoted into agent context files like CLAUDE.md, AGENTS.md, and copilot-instructions, expanding useful corrections and non-obvious issues into broader reusable context for future agent workflows.

The self-improvement logging workflow requires a SKILL.md file with YAML frontmatter containing a name and description, alongside consistent markdown logging into centralized files within the .learnings directory structure.

Using markdown files for persistent agent memory requires consistent logging workflows and helper scripts to detect entries; without active hook-based reminders, non-obvious issues and recurring patterns might go uncaptured across sessions.