interview-article

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

• Third-party content exposure detected (high risk: 0.90). Flagged because SKILL.md explicitly requires the agent to perform "联网搜索" and mandatory external investigation (Step 1.5 and Step 2) — including visiting public GitHub repos, paicoding.com, "来自 X 的外界评价", HuggingFace/LMSYS, etc. — and to load/or cite those public documents into prompts and the evidence checklist, meaning untrusted third‑party content is read and can materially change prompt construction and behavior.