dgx-spark

CRITICAL E005: Suspicious download URL detected in skill instructions.

• Suspicious download URL detected (high risk: 0.80). Several URLs point to locally-hosted apt/LVFS mirrors configured to accept untrusted firmware/packages and to third-party "curl | sh" installer endpoints (get.hermes.dev, get.openclaw.ai) which are not official vendor domains — both patterns are commonly used to distribute malware or unauthorized binaries.

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

• Third-party content exposure detected (medium risk: 0.30). Skill.md の探索手順に従い、カテゴリ README/ページ（例: references/community/known-issues.md や troubleshooting.md）を Read して詳細確認する構成であり、これらは外部（NVIDIA Developer Forums 等）由来のユーザ投稿テキストを含み得ます（直接的な外部入力経路は不明でも、少なくともコミュニティ由来文書の自由記述がLLMに読まれる）。

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

• Potentially malicious external URL detected (high risk: 0.90). Multiple playbooks contain commands that fetch and immediately execute remote scripts or clone repositories that are then built/run (e.g., "curl -fsSL https://get.hermes.dev | sh" and "wget -O sync-pulp.py https://gitlab.com/fwupd/lvfs-website/raw/master/contrib/sync-pulp.py" ), so the skill includes runtime external dependencies that execute remote code.