metamorphic-testing
Pass
Audited by Gen Agent Trust Hub on Oct 14, 2026
Risk Level: SAFENO_CODEINDIRECT_PROMPT_INJECTION
Full Analysis
- [NO_CODE]: The skill is composed of markdown-based strategy and instructions. It does not include scripts, executables, or configurations that perform active operations.\n- [INDIRECT_PROMPT_INJECTION]: The skill outlines a testing process that utilizes untrusted user inputs, establishing a potential attack surface. However, it explicitly warns about this risk and provides mitigation strategies.\n
- Ingestion points: Untrusted user text is processed as seed inputs for metamorphic relations (Security & Guardrails section).\n
- Boundary markers: Not explicitly defined in the methodology, though isolation from instruction context is recommended as a mitigation.\n
- Capability inventory: The strategy involves using the LLM as a judge for equivalence checking (Step 3).\n
- Sanitization: Recommended treating transformed inputs as untrusted data, though specific escaping implementations are left to the user.\n- [SAFE]: No malicious patterns such as prompt injection, data exfiltration, or obfuscation were found. The skill aligns with legitimate testing best practices.
Audit Metadata