Self-Improvement Agent

Captures learnings, errors, and corrections for continuous improvement. Use when: (1) A command fails unexpectedly, (2) User corrects you ('No, that's wrong....

This skill appears coherent and implements a local 'log learnings' workflow. Key things to consider before installing or enabling hooks: - No credentials are requested, and there are no remote downloads; the code is local and readable. Review the scripts (activator.sh, error-detector.sh, extract-skill.sh) and handler.js/handler.ts yourself before enabling. - Enabling the provided hook(s) (project or user-level) will cause the scripts/handler to run automatically during agent bootstrap or after tool use; those scripts run with the same permissions as your agent and can read/write files in the workspace and skills directories. If you prefer limited scope, enable the hook only at the project level (not user/global) or keep PostToolUse hooks disabled. - The skill suggests using inter-session APIs (sessions_history, sessions_send). If you enable those behaviors, sessions_history can expose other session transcripts — enable only if you are comfortable sharing session content between sessions. - If you install, set script permissions intentionally (chmod +x) and prefer absolute paths in hook configs to avoid surprises. If you want minimal exposure, use the activator reminder only (no PostToolUse error detector) and avoid running extract-skill.sh unless you explicitly want to create new skill files. Overall: coherent and reasonable for its stated purpose, but treat hook/script activation as an explicit opt-in action and review them before enabling globally.

Type: OpenClaw Skill Name: self-improvement-agent Version: 1.0.0 This skill bundle implements a structured 'Self-Improvement' framework for OpenClaw agents, allowing them to log errors, corrections, and feature requests to a `.learnings/` directory. It utilizes OpenClaw hooks (`handler.js/ts`) to inject reminders into the agent's context and provides utility scripts (`extract-skill.sh`) for scaffolding new skills from captured learnings. The code and instructions are transparent, lack any indicators of data exfiltration or unauthorized execution, and align entirely with the stated goal of improving agent performance through iterative feedback.