response-analysis

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 1.00). The prompt explicitly requires copying verbatim quotes from the source ("Every score needs 3 direct quotes" and "Never paraphrase quotes") into the agent's output, so if the source contains API keys, passwords, or other secrets the LLM would be forced to reproduce them, creating a high exfiltration risk.

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (medium risk: 0.30). SKILL.md describes an LLM workflow that scores “user research, customer interviews, and feedback triage” by taking “one answer, one ticket, one paragraph,” “multi-turn interview, sales call, or conversation,” or “multiple responses to analyse together” as input and quoting that provided text verbatim for scoring.