Openclaw Self Improving Agent 3.0.1

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 coherent and focused on logging learnings. Before enabling: (1) review the hook files (hooks/openclaw/handler.*) and scripts to confirm you’re comfortable with their behavior, (2) only copy/enable hooks if you want session-start reminders injected into every session, (3) check scripts' permissions (chmod +x) and run them in a test workspace first, and (4) do not expose secrets—this skill does not require credentials but will read platform-provided variables like CLAUDE_TOOL_OUTPUT when used as a PostToolUse hook. Enable it deliberately and verify its output in a safe project before using it in sensitive environments.

Type: OpenClaw Skill Name: openclaw-self-improving-agent-3-0-1 Version: 1.0.0 The skill bundle implements a 'self-improvement' loop that allows the agent to modify its own behavioral guidelines and workflow instructions (e.g., SOUL.md, AGENTS.md, CLAUDE.md) and create new skill scaffolds via extract-skill.sh. While these capabilities are aligned with the stated purpose and the scripts include robust security checks (such as path-traversal prevention in extract-skill.sh), the ability for an agent to persistently alter its own core instructions and generate new file structures represents a high-risk capability and a significant attack surface for prompt injection. The scripts activator.sh and error-detector.sh are used to monitor tool outputs and inject reminders into the agent's context, further facilitating this self-modification behavior.