wenqu-image

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

• Third-party content exposure detected (medium risk: 0.30). references/gen-workflow.md 在“第二步：列出所有画图点”与“第三步：生成图片”会直接扫描调用方文章中的占位标记（如 > 🖼️ 待配图：[一句话描述]）并把其正文内容转换成 --prompt 供 LLM/图片生成使用，因此外部用户可通过向文章/队列写入该占位标记文本来影响运行时注入。

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.80). The skill fetches style reference images at runtime from GitHub raw URLs (e.g. https://raw.githubusercontent.com/gogoingai/wenqu-skills/master/wenqu-image-assets/styles/mono-marker/mono-marker-02-branch-decision.png via scripts/fetch-ref.sh), and those downloaded images are passed as --ref to the image-generation commands and therefore directly influence model outputs—this is an external runtime dependency used to control generation.