vps-server-management
Fail
Audited by Gen Agent Trust Hub on Aug 15, 2026
Risk Level: HIGHCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONCREDENTIALS_UNSAFE
Full Analysis
- [COMMAND_EXECUTION]: Instructions explicitly advocate for running agents on remote VPS hosts using flags that bypass human confirmation (e.g.,
codex --yolo). This creates a high-risk autonomous execution environment where the AI can perform destructive operations on production infrastructure without user review. - [CREDENTIALS_UNSAFE]: The skill directs the agent to use the
rootaccount as the default SSH user for all operations. This violates the principle of least privilege, as any error or injection results in full system compromise rather than being restricted to a user-level shell. - [REMOTE_CODE_EXECUTION]: The skill provides commands (
hermes update) that trigger remote dependency fetching and software rebuilding directly on the production server, which can be exploited if the source repository or package registry is compromised. - [DATA_EXFILTRATION]: The skill identifies sensitive file targets, specifically pointing the agent to
~/.hermes/config.yamland.envfiles, which are standard locations for API keys and database credentials, and mapping internal infrastructure details like specific IP placeholders and OS versions.
Recommendations
- AI detected serious security threats
Audit Metadata