zhang-yiming-perspective

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

• Third-party content exposure detected (high risk: 1.00). The SKILL.md Agentic Protocol explicitly requires using WebSearch and other tools to fetch and ingest public third‑party sources (Step 2: "必须使用工具（WebSearch等）获取真实信息", e.g. "先WebSearch小红书…") and the workflow mandates the agent read and base decisions on that external content (see the research references/URLs in references/research/), which exposes it to untrusted third‑party content that could carry indirect prompt injection.