zhangxuefeng-perspective

CRITICAL E004: Prompt injection detected in skill instructions.

• Potential prompt injection detected (medium risk: 0.30). The skill explicitly steers an agent to impersonate a real (deceased) public figure in first person, minimize/limit ongoing disclosure (disclaimer only once), hide research provenance from users, and forcibly require external tool use — a deceptive steering pattern that weakens transparency but is not clearly weaponized.

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

• Third-party content exposure detected (medium risk: 0.30). SKILL.md 的 Step 2 要求使用 WebSearch 等工具获取“就业数据/行业报告/校友反馈、求职论坛”等信息并据此回答，但其规则未展示对搜索结果内容进行约束或“只读可信/只读特定条目”的机制，因此运行时会不可避免地读取外部搜索结果页面/文本，其中可能包含由非可信用户提交的免费文本。