ai-ethics-security
Pass
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: SAFENO_CODEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill is entirely defensive, providing a VP AI Ethics & Security role focused on implementing safety guardrails, governance, and compliance (EU AI Act, NIST AI RMF). No malicious instructions, obfuscation, or unauthorized access patterns were found.
- [INDIRECT_PROMPT_INJECTION]: The skill architecture is designed to ingest and process untrusted user inputs through a security pipeline.
- Ingestion points: User prompts entering the proposed "Prompt Injection Guard" and "PII Detection" layers as described in Phase 2 of the security framework (SKILL.md).
- Boundary markers: The skill explicitly recommends instruction hierarchy (PS-01) and specific delimiters (SYSTEM, HUMAN, ASSISTANT) to isolate untrusted content (references/ai-security-controls-library.md).
- Capability inventory: The skill does not implement active capabilities or tool access; it serves as a knowledge framework for building secure systems.
- Sanitization: The skill's core purpose is sanitization, employing PII masking pipelines (DS-01) and injection detection classifiers (PS-02).
- [EXTERNAL_DOWNLOADS]: The implementation library references standard open-source security tools and libraries including Microsoft's Presidio (presidio-analyzer, presidio-anonymizer) and Sigstore's Cosign for model integrity verification. These are well-known security utilities.
Audit Metadata