vps-server-management
Warn
Audited by Gen Agent Trust Hub on Jul 7, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to perform high-privilege operations on remote VPS servers as the 'root' user. It explicitly recommends running a local agent on the server using the '--yolo' flag (e.g., 'codex --yolo'). This flag typically disables human-in-the-loop confirmations for shell commands, creating a high risk of unintended or destructive system changes.
- [DATA_EXFILTRATION]: The instructions direct the agent to access and modify sensitive server-side configuration files (e.g., '~/.hermes/config.yaml') and read internal infrastructure documentation ('library/infrastructure.md'). Accessing these files as root exposes credentials and system architecture details.
- [PRIVILEGE_ESCALATION]: The skill promotes the use of the 'root' account for all management tasks instead of using a non-privileged user with sudo, violating the security principle of least privilege.
- [INDIRECT_PROMPT_INJECTION]: The skill involves reading and inspecting system logs (via 'journalctl') which may contain data from external, untrusted sources (e.g., Discord/gateway traffic).
- Ingestion points: System logs accessed via 'journalctl', external documentation in 'library/'.
- Boundary markers: None provided to separate log data from agent instructions.
- Capability inventory: Full root SSH access, shell command execution, service management, and file modification.
- Sanitization: No instructions for escaping or validating the content of logs before processing.
Audit Metadata