context-to-video

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

• Third-party content exposure detected (medium risk: 0.30). SKILL.md は「ブログ/記事 URL」を web_fetch で本文取得し、その本文由来の要約/スクリプト生成（notes/script.jsonへの書き込み→build_video.pyでLLM/tts/字幕生成に使用）というランタイム経路を前提としており、URLの内容は外部ユーザーが投稿した自由テキストになり得るため。

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

• Potentially malicious external URL detected (high risk: 0.90). The skill's optional Avatar mode requires fetching and later executing remote SadTalker code and model checkpoints (e.g., git clone https://github.com/OpenTalker/SadTalker.git and the release asset URLs in scripts/download_models.ps1), which downloads external code/models that will be executed by the provided add_avatar.py pipeline.