Restart Guard

Deterministic OpenClaw gateway restart with down/up state-machine verification, origin-session proactive ACK, and backward-compatible config.

This skill will automatically perform gateway restarts and then try to proactively notify the originating session and external channels. Before installing or enabling it: - Be aware it is designed to auto-run on simple natural-language triggers (e.g. "restart now") and may restart a gateway without extra confirmation. Consider requiring explicit human confirmation in your deployment if you don't want one-shot voice/text triggers to restart production. - Review and limit the GATEWAY_AUTH_TOKEN and any notification tokens (Telegram/Discord/Slack/webhook) it will use. The code will also read ~/.openclaw/.env to resolve tokens — inspect that file for other secrets before allowing the skill access. - Audit the included scripts (restart.py, guardian.py, notify.py, write_context.py) in your environment and test in a staging system first. Pay attention to the notification webhook URLs and content templates to avoid accidental secret leakage. - If you want stricter safety, modify the workflow to require an explicit confirmation step before executing the auto_restart flow, or limit agent permissions so automatic invocation cannot be triggered by casual chat.

Type: OpenClaw Skill Name: restart-guard Version: 2.2.0 The OpenClaw AgentSkills skill bundle is classified as 'suspicious' due to its inherent high-risk capabilities, such as restarting the OpenClaw gateway, performing system-level operations (e.g., `lsof`, `kill`), and sending notifications to external services (Telegram, Discord, Slack, generic webhooks). While these actions are aligned with the skill's stated purpose of deterministic gateway restart and reporting, they represent significant control over the system and external communication channels. The code demonstrates strong security awareness, with explicit fixes for past vulnerabilities (webhook template injection, host/port validation, shell metacharacter rejection in commands) detailed in `CHANGELOG.md` and verified by `tests/test_security_fixes.py`. Sensitive data is redacted from logs and diagnostics. No evidence of intentional malicious behavior (e.g., credential theft, unauthorized exfiltration to unknown endpoints, persistence mechanisms) was found. The `README.md` explicitly notes that the skill may be marked 'suspicious' due to its capabilities, not malicious intent, which aligns with this classification.