whispers-wealth-red-teaming-googles
Pass
Audited by Gen Agent Trust Hub on Jun 22, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: The skill provides instructional content on crafting prompt injection payloads specifically for red-teaming agentic payment systems. These examples (e.g., manipulating product rankings via system notes in merchant metadata) are documented for security testing purposes and are presented as non-executable examples within a research context.
- [DATA_EXFILTRATION]: The skill describes methodologies for simulating cross-user data leakage to audit credential isolation in multi-agent systems. It includes example test cases designed to evaluate whether credentials can be traversed across agent boundaries via user-controlled input.
- [SAFE]: The content includes proactive defensive guidance, such as input sanitization logic, structured message schemas, and rule-based access controls, to mitigate the vulnerabilities described. The skill does not contain any obfuscation, hardcoded credentials, or network operations.
Audit Metadata