Security audit
Security checks for vulnerabilities and agentic risk
This skill is a disclosed OpenClaw monitoring helper with some operational authority, but its sensitive actions are purpose-aligned and require user approval.
Install this only if you want the agent to inspect OpenClaw gateway status, sessions, cron jobs, plugins, and host resource usage, and to send status or critical alerts in the active channel. Review scheduled summaries and any proposed fix command carefully before approving restarts, retries, reinstalls, or log deletion.
The trigger list includes broad natural-language phrases such as 'status', 'is everything working', and 'system health', which can cause the skill to activate in unintended contexts. Because this skill performs health checks, sends messages proactively, and can propose remediation workflows, accidental invocation may expose operational details or generate unsolicited actions/alerts.
The skill claims zero configuration and instructs 'MUST NOT' parse OpenClaw config for channel detection (L017, L052), yet the edge-case guidance later directs the agent to read timezone from OpenClaw config at gateway.timezone (L226). This is an intent-level contradiction because the documentation simultaneously rejects config parsing as an approach while depending on config parsing to implement quiet hours.
The skill is designed for proactive alerts and scheduled daily summaries in the current channel, but it does not define robust rate limiting, deduplication, or hard caps beyond general guidance not to spam. In a noisy failure condition or ambiguous trigger context, this can lead to repeated unsolicited messages, causing channel flooding and alert fatigue, which degrades availability and can mask genuinely critical events.
### MUST NOT DO (never)
- Hardcode Telegram chat IDs, environment variables, or other channel‑specific config.
- Spam the channel — routine summaries are daily, immediate alerts only for real problems.
- Build a web UI, Grafana dashboard, or any external interface.
- Assume the user uses Telegram — work with whatever channel the current conversation uses.
- Report trivial warnings as urgent — use judgment.
The file states at L052 that the skill must not parse OpenClaw config for channel detection and should rely on the current conversation context, but later says routine summaries should be sent to the 'detected channel' (L127), that alerts may go to 'the channel configured for the cron job' (L108), and earlier claims zero configuration detects the channel from OpenClaw config (L017). These instructions actively conflict about whether channel routing should come from conversation context, cron configuration, or parsed config.
The instruction "Use UK English" forces a specific language/locale variant for all responses. This is a natural-language policy concern because it removes user choice and does not document a region-specific reason for the restriction.
No suspicious patterns detected.