devils-advocate

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 1.00). 本 skill 明确要求把用户提供材料的“逐字摘录 / 逐字引用”写入产出（JSON 字段 excerpt 与报告），因此若材料包含 API 密钥、token 或密码，模型必须在输出中逐字处理并泄露这些秘密。

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

• Third-party content exposure detected (medium risk: 0.30). The runtime workflow ingests LLM-readable text only from the user-provided “evidence materials” (Step 0 input confirmation + Step 2 “逐假设检索反证” over provided file/URL lists), so outsider-authored free text exposure exists but only within attacker-delivered materials rather than arbitrary monitoring feeds.