evaluation-operations
Pass
Audited by Gen Agent Trust Hub on Oct 22, 2026
Risk Level: SAFENO_CODE
Full Analysis
- [NO_CODE]: The skill consists entirely of Markdown documentation and instructional references. There are no scripts, executables, or code blocks that would trigger traditional attack vectors.
- [EXTERNAL_DOWNLOADS]: The skill references established open-source tools (promptfoo, Microsoft PyRIT, NVIDIA garak) as authoritative sources for its methodology. These references are documented neutrally and point to official repositories.
- [PROMPT_INJECTION]: The instructions emphasize strict adherence to user-granted authority modes (read-only, resume, or monitoring) and project-defined records, effectively instructing the agent to respect external boundaries rather than bypassing them.
- [CREDENTIALS_UNSAFE]: The documentation explicitly instructs the agent to keep credentials in the configured secret mechanism and never in reports or worker packets.
Audit Metadata