Monitor

Create monitors for anything. User defines what to check, skill handles scheduling and alerts.

Before installing, review and control these points: - Inspect monitor definitions (~/monitor/monitors.json) and avoid storing secrets (API keys, auth headers) in plain JSON. Use environment variables or secure stores where possible. - Be cautious with 'custom' checks: they run arbitrary commands you supply. Do not add untrusted commands that could access or transmit sensitive data. - Only provide webhook URLs or channels you trust. Alert payloads may include status, timestamps, and any logged output; a misconfigured webhook can leak information. - Limit granted access (SSH, Docker) to only the hosts/services you intend to monitor; the skill asks for explicit permission but will act on whatever is listed in 'requires'. - Set filesystem permissions on ~/monitor so only the intended user can read logs and configs (e.g., chmod 700 ~/monitor). - If you need stronger isolation, run this agent under a restricted user account or in an environment with limited network access, and review scheduled intervals to avoid resource or network abuse. Overall: the skill appears coherent for its stated purpose, but take care around custom commands, webhook targets, and any sensitive data that could be captured in logs or alert payloads.

Type: OpenClaw Skill Name: monitor Version: 1.0.2 The skill is classified as suspicious due to a critical vulnerability allowing arbitrary command execution (RCE) via prompt injection against the AI agent. Both `SKILL.md` and `templates.md` explicitly instruct the agent to execute user-provided commands for 'custom' monitor types (e.g., `"command": "user-provided-command"`). This allows a malicious user to define a monitor that executes arbitrary shell commands on the system where the agent runs. Additionally, `alerts.md` shows `curl` commands for sending alerts, which could be vulnerable to shell injection if user-controlled inputs like `$WEBHOOK_URL` or `$MESSAGE` are not properly sanitized before execution.