qa-agent-testing
Pass
Audited by Gen Agent Trust Hub on Oct 17, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill is a comprehensive utility for agent QA and security testing, designed to help developers build robust and safe systems.\n- [PROMPT_INJECTION]: The skill includes adversarial test cases (e.g., 'Ignore previous instructions', 'DAN' mode) in its documentation, specifically in
references/prompt-injection-testing.md. These are intended for scanning other agents' defenses and do not influence the host agent's own behavior.\n- [COMMAND_EXECUTION]: The unit test scriptscripts/test_score_suite.pyusessubprocess.runto verify the output of the localscore_suite.pyutility. This is standard testing practice for a CLI tool and does not pose a security risk in this context.\n- [EXTERNAL_DOWNLOADS]: Numerous links to external tools and research papers are provided indata/sources.jsonand documentation. All identified domains point to trusted organizations such as NIST, OWASP, Anthropic, OpenAI, Microsoft, and NVIDIA.\n- [INDIRECT_PROMPT_INJECTION]: The skill identifies and provide guidance on testing for indirect prompt injection via tool outputs or RAG context. It emphasizes strong mitigation strategies like sandboxing and objective oracles, as detailed inreferences/tool-sandboxing.md.
Audit Metadata