Self-Improving Agent (ORBIT)

Captures learnings, errors, and corrections to enable continuous improvement. Use when: (1) A command or operation fails unexpectedly, (2) User corrects Clau...

This skill appears to be what it claims: lightweight reminders, local logging, and helpers for turning learnings into files or new skills. Before enabling it or installing hooks: 1) Inspect the hook handler and scripts (they are small and readable) to confirm you’re comfortable with them writing files under your workspace and printing reminders to sessions. 2) Only enable the OpenClaw/Claude hooks if you trust the skill source; enabling hooks lets the skill inject context into every session (intended behavior for a reminder tool but a prompt‑injection capability nonetheless). 3) Verify the installation source: SKILL.md references a GitHub repo and the embedded _meta.json ownerId differs slightly from the registry owner slug — this may be a fork or renamed package; confirm the repository and publisher before cloning. 4) Run scripts in dry-run mode (extract-skill supports --dry-run) or from a controlled workspace first. 5) If you use the extract-skill helper, note it will create files under the relative ./skills path you provide — ensure you run it from the intended directory. If any of these checks raise doubt (mismatched repository, unexpected remote endpoints, or if you don’t want session-level injections), don’t enable the hooks and use the skill manually by keeping .learnings/ files in your project.

Type: OpenClaw Skill Name: self-improving-agent-orbit Version: 1.0.11 The 'self-improving-agent-orbit' skill bundle is designed to facilitate a continuous learning loop for AI agents by logging errors, user corrections, and best practices to markdown files (e.g., in a `.learnings/` directory). The bundle includes utility scripts like `extract-skill.sh` for scaffolding new skills and OpenClaw hooks (`handler.js`/`handler.ts`) that inject reminders into the agent's context. While the skill involves writing to the local filesystem and reading tool output for error detection, these actions are strictly aligned with its stated purpose of self-improvement and lack any indicators of malicious intent, data exfiltration, or unauthorized persistence.