vps-blue-green-prepare
Pass
Audited by Gen Agent Trust Hub on Oct 24, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill operates in a preparation-only mode. It explicitly instructs the agent: 'Produce project-specific deployment artifacts but do not connect to or deploy on the VPS' and 'Do not execute remote commands or deploy as part of preparation.'
- [DATA_EXPOSURE_AND_EXFILTRATION]: The instructions include strong safeguards against credential leakage: 'Never embed credentials', 'do not invent... credentials', and 'Are secrets absent from code, script arguments, docs, and logs?'
- [COMMAND_EXECUTION]: While the skill generates deployment scripts (e.g., shell, systemd), it mandates strict error handling, validated arguments, safe quoting, and non-destructive operations. It requires a manual review step: 'Do not claim that the generated artifacts are production-ready until their assumptions, syntax, and available tests have been checked.'
- [PRIVILEGE_ESCALATION]: The skill instructs the agent to preserve existing topology and avoids unnecessary privilege escalations. It encourages using the smallest appropriate set of files and avoiding changes to unrelated services.
- [INDIRECT_PROMPT_INJECTION]: The skill identifies a potential attack surface by inspecting repository files. However, it mitigates risk by requiring the agent to ask focused questions for missing facts rather than making assumptions, and it implements a checklist to verify the safety of generated configurations.
Audit Metadata