aaaa

This skill appears to implement a reasonable 'log learnings and remind the agent' workflow, but review these before installing: - Inconsistency: the registry lists this as 'instruction-only' yet the package contains executable scripts and hook handlers. Verify the package contents and source repository you install from. - Persistent context risk: the skill encourages promoting learnings into workspace files (SOUL.md, AGENTS.md, CLAUDE.md, TOOLS.md) that OpenClaw injects into every session. Only promote content you fully trust and sanitize — otherwise you can unintentionally introduce persistent prompt-injection or behavioral changes. - Hooks are opt-in: the hook handlers only run if you copy/enable them. If you don't want global reminders or injected bootstrap files, don't enable the hook. - Inspect scripts before running: activator.sh and error-detector.sh only print reminders and parse CLAUDE_TOOL_OUTPUT, and extract-skill.sh scaffolds files under ./skills but can write to disk. Check file permissions and the scripts' exact behavior; run extract-skill.sh with --dry-run first. - Remote install caution: SKILL.md recommends git-cloning a GitHub repo. If you follow that, verify the repository and its commit history before cloning/executing scripts. If you want to proceed: only enable hooks at project scope (not global), avoid promoting untrusted learnings into injected workspace files, and run the scripts in a controlled environment first. If you want more confidence, provide the canonical upstream repo URL and confirm owner metadata — the packaged _meta.json ownerId differs from the registry ownerId, which is another inconsistency worth resolving.