reverse-teardown

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 0.90). The skill forces verbatim insertion and reproduction of "用户给的真实信息" (replace placeholders with user-supplied content and output text exactly), so if a user pastes secrets (API keys, tokens, passwords) the model will include them verbatim, enabling secret exfiltration.

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

• Third-party content exposure detected (medium risk: 0.30). SKILL.md defines a “reverse-teardown” workflow that ingests user-provided free text by having the user paste a product page/webpage/solution/process/dashboard or other materials into the prompt (e.g., “【粘贴产品页面、网页…】”), so outsider-authored content can be directly consumed by the LLM at runtime.