ai-pm

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 透過 mcp-figma 解析使用者提供的「Figma 連結」並提取頁面/Frame/文案/Label 等可讀文字，屬於在運行流程中攝取外部（使用者提交）自由文本的間接提示注入風險。

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

• Potentially malicious external URL detected (high risk: 0.90). The skill requires a Figma link as a mandatory runtime input and explicitly calls mcp-figma to parse that remote design, so fetched Figma content (e.g. https://www.figma.com/file/AbCdEf/insurance-portal?node-id=123:456) will directly control the agent's generated spec output.