5000 / conversation / 然而
Pass
Audited by Gen Agent Trust Hub on Mar 18, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill instructs the agent to focus on sensitive geopolitical questions (Russia-Ukraine war, North Korea) as primary evidence for extraction.
- [PROMPT_INJECTION]: Metadata tags include '作为一个ai语言模型' (As an AI language model) and direct questions like 'Would you support Russia or Ukraine', which are patterns used to influence agent personas or safety behaviors.
- [PROMPT_INJECTION]: The skill exhibits an Indirect Prompt Injection surface: (1) Ingestion points: Untrusted user questions provided in the prompt; (2) Boundary markers: None present to distinguish data from instructions; (3) Capability inventory: No code or file-system access; (4) Sanitization: None present.
Audit Metadata