ClawGuard

System-level watchdog for OpenClaw gateway restarts and sub-agent task PIDs. Monitors registered PIDs and optional log/directory freshness. Auto-reverts conf...

Before installing: 1) Review and confirm you trust the author — this will install a persistent user service and a CLI under your home directory. 2) Understand it will copy/restore your ~/.openclaw/openclaw.json from backups and will run systemctl/launchctl to restart the gateway — that's the intended behavior but is powerful, so ensure backups are what you expect. 3) Notifications include snippets of status/journal output and will be sent through your configured OpenClaw channels — verify that those channels are appropriate to receive potential debug output. 4) The installer requires python3 in PATH; it embeds your current PATH into the service environment. If you have concerns about the exact python binary used, inspect/modify install.sh before running. 5) If you need tighter control, consider installing but not enabling ExecStartPre in the gateway service, or run the daemon manually first to observe behavior. If you want more assurance, review the three script files locally (they are small and self-contained) before running the installer.

Type: OpenClaw Skill Name: openclaw-claw-guard Version: 1.1.0 The OpenClaw ClawGuard skill is classified as suspicious due to its powerful system-level capabilities and reliance on external components, which introduce potential vulnerabilities. The `scripts/install.sh` script installs a persistent user-level service (systemd/launchd) that runs the `scripts/claw-guard.py` daemon. This daemon performs critical actions such as modifying the `~/.openclaw/openclaw.json` configuration file, restarting the OpenClaw gateway service, and sending notifications via the `openclaw` CLI. The `notify` function in `scripts/claw-guard.py` executes `openclaw message send` with user-controlled `target` and `message` arguments. While `subprocess.run` with a list of arguments mitigates direct shell injection, it relies on the `openclaw` CLI being robust against argument injection. Furthermore, sending gateway status, error logs, or journal entries to a user-defined target (a core feature) poses an information disclosure risk if an untrusted target is specified. These capabilities, while aligned with the stated watchdog purpose, represent significant power that could be exploited if the agent or its inputs are compromised, leading to potential system compromise or data leakage, thus warranting a 'suspicious' classification rather than 'benign' or 'malicious'.