self-improvement

To capture learnings across OpenClaw sessions, you log errors and corrections by adding structured entries to files under the `.learnings/` directory. This requires a frontmatter with name and description to structure outputs.

Recording command failures and user corrections involves capturing these events as structured outputs in `.learnings/LEARNINGS.md`. This turns lessons into repeatable improvements and applies to external tool failures and outdated knowledge.

Yes, you can route noteworthy learnings to broad-priority channels like CLAUDE.md, AGENTS.md, or copilot-instructions.md. This optional promotion path applies when patterns are widely applicable across your workspace.

No, the continuous improvement workflow has no external dependencies. It provides a workspace-friendly structure that integrates with optional hooks for automation, requiring only a frontmatter with name and description.

You should avoid this learning capture structure if your tasks do not involve command failures, user corrections, missing capabilities, or outdated knowledge. It is specifically designed to turn these recurring lessons into repeatable improvements.