Update to RelayDesk

{
  "description": "Use when a user wants to connect to, inspect, control, or operate a computer, laptop, server, or VM explicitly paired to their RelayDesk account, even without mentioning RelayDesk. Covers device discovery, online status, running processes, approved OS commands, diagnostics, and a specific authorized action. Do not use for unpaired devices or unrelated web and cloud tasks.",
  "included_files": [
    {
      "relative_path": "agents/openai.yaml",
      "size_in_bytes": 428
    },
    {
      "relative_path": "assets/icon.png",
      "size_in_bytes": 3037
    }
  ],
  "name": "remote-device-operations",
  "skill_md_contents": "---\nname: remote-device-operations\ndescription: \"Use when a user wants to connect to, inspect, control, or operate a computer, laptop, server, or VM explicitly paired to their RelayDesk account, even without mentioning RelayDesk. Covers device discovery, online status, running processes, approved OS commands, diagnostics, and a specific authorized action. Do not use for unpaired devices or unrelated web and cloud tasks.\"\n---\n\n# Operate a paired computer\n\nIdentify the user-owned device from context or call list_devices. Call ping when checking reachability. If the operation is consequential and the target remains ambiguous, identify the intended machine before acting.\n\nFor system information, use focused read-only tools such as get_config, list_processes, list_sessions, list_directory, get_file_info, and read_file. Never inspect credential stores, tokens or private keys.\n\nFor an authorized command, call start_process on the selected Windows, macOS or Linux machine; retrieve real output and exit status with read_process_output. Continue existing processes when appropriate rather than restarting them unnecessarily. Use file mutation tools only for user-requested changes.\n\nRespect ownership, protected directories and blocked commands; never weaken local policy or touch unpaired computers. Report the actual device, action, results and errors, without claiming success when verification failed.\n\n## Examples\n\n- \"Can you connect to my MacBook and check if it is online?\"\n- \"Inspect running processes on my Linux server.\"\n- \"Run a safe uptime command on my laptop.\"\n- \"On the Windows VM, see whether the app is running.\"\n- \"Check the machines I have paired and resume the process I started.\"\n\nDo not invoke for ordinary web browsing, connected SaaS tasks, or files already uploaded directly into ChatGPT.\n"
}