OpenClaw Guardian by MyClaw.ai

Deploy and manage a Guardian watchdog for OpenClaw Gateway. Auto-monitor every 30s, self-repair via doctor --fix, git-based workspace rollback, daily snapsho...

This skill generally does what it says, but review and test before deploying to production: - Backup your workspace first. The guardian runs git reset --hard and will discard uncommitted changes; it also auto-commits all changes daily. Ensure the repo contains only content you are willing to have committed and potentially rolled back. - Confirm availability of required binaries: git, pgrep, pkill, curl, nohup, and especially the OpenClaw CLI (openclaw). The registry metadata omitted pkill and openclaw; ensure they exist. - Review what openclaw doctor --fix and openclaw gateway do in your environment — these commands can perform wide-reaching repairs/restarts. - Discord webhook: the webhook is optional, but the registry marks it as the primary credential; verify the webhook URL destination and treat it as a secret. Notifications contain only brief status messages, but do not supply other workspace contents by default. - Run the script in a sandbox or staging environment first to confirm behavior and to tune intervals/cooldowns/repair attempts. - If you need stricter safety, consider modifying the script to: (a) require manual approval before git reset --hard, (b) restrict auto-commit behavior or exclude certain files, and (c) log (and rotate) backups outside the workspace. Given the above, the skill appears coherent with its purpose but requires operational caution.

Type: OpenClaw Skill Name: myclaw-guardian Version: 1.0.2 The skill is designed as a legitimate watchdog for OpenClaw Gateway. However, the `scripts/guardian.sh` script directly executes the `$OPENCLAW_CMD` environment variable (e.g., `$OPENCLAW_CMD doctor --fix` and `$OPENCLAW_CMD gateway`). If an attacker can control the `OPENCLAW_CMD` environment variable, this allows for arbitrary command injection and potential Remote Code Execution (RCE). While this is a critical vulnerability, it is a flaw that *allows* an attack rather than evidence of intentional malicious design within the skill itself, thus classifying it as suspicious.