ai-prototype-flow

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

• Third-party content exposure detected (high risk: 0.75). 案件初期化（gdocs URL 起点）で、Google Docs からエクスポートした Markdown（外部＝gdocs/ユーザーが指定した文書本文）がローカルに保存され、その後のフェーズ実行で LLM コンテキストに読み込まれる経路があるため（references/dispatcher_modes.md→references/ai_prototype_flow.md のエクスポート手順）。

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

• Potentially malicious external URL detected (high risk: 0.90). The skill contains runtime curl fetches of Google Docs content that the agent executes/depends on (e.g. "https://docs.google.com/feeds/download/documents/export/Export?id=&exportFormat=markdown&tab=t." and the equivalent in the included script), so remote document content is fetched at runtime and used as required input.