SkillFence

Runtime security monitor for OpenClaw skills. Watches what your installed skills actually DO — network calls, file access, credential reads, process activity. Not a scanner. A watchdog.

SkillFence appears coherent with its purpose as a local runtime monitor. Before installing: 1) Review the monitor.js file yourself (it’s small and included) to confirm behavior you’re comfortable with. 2) Confirm you trust the source/ GitHub repo the README points to (the registry owner ID, homepage, and GitHub repo user differ — verify origin). 3) Run it as a normal user (not root) so logs and scans run with limited privileges. 4) Note it will create files in your HOME (audit log, session state, license); back up sensitive files if you need to. 5) If you plan to enable any Pro/dashboard integration, confirm what data (if any) is sent externally — the bundled code appears local-only, but documentation references an external dashboard. Finally, when using features that analyze arbitrary commands/messages, avoid passing untrusted input that might be interpreted as shell commands; the monitor inspects strings, but always be cautious.

Type: OpenClaw Skill Name: skillfence Version: 1.0.6 The OpenClaw SkillFence skill is a runtime security monitor designed to detect malicious behavior in other skills. Its code (`monitor.js`) and documentation (`SKILL.md`, `README.md`) consistently describe a tool that performs system introspection (process monitoring, network connection analysis, file access timestamp checks) and code scanning of other skills. Crucially, the skill explicitly claims and the code verifies that it is 'read-only' (only modifying its own state/log files, not other system files or credentials' contents) and 'never makes outbound network requests'. The use of `execSync` for system commands (like `ps aux`, `lsof`, `ss`) and `fs.readFileSync` for scanning other skill files are necessary for its stated security purpose. Prompt injection instructions in `SKILL.md` guide the AI agent to perform security monitoring and transparently report findings, not to subvert its function or hide actions. There is no evidence of intentional harmful behavior such as data exfiltration, persistence, or unauthorized remote control.