anti-deception
Pass
Audited by Gen Agent Trust Hub on Jul 2, 2026
Risk Level: SAFE
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the AI to call the 'anti-deception' tool from the 'ejentum' MCP server. This is a legitimate use of the MCP extension framework for enhancing AI behavior.
- [DATA_EXFILTRATION]: A description of the user's framing (e.g., 'user pressure to validate') is sent to the external tool. This transmits conversation context to the vendor's service to generate an appropriate response scaffold.
- [PROMPT_INJECTION]: The skill presents an indirect prompt injection surface by instructing the AI to follow instructions returned by an external tool. However, this is used for safety reinforcement: Ingestion points include the tool output; Boundary markers include instructions to absorb internally and suppress labels; Capability inventory shows no local code execution or file access; Sanitization is not explicitly defined.
Audit Metadata