deeplake

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

• Third-party content exposure detected (low risk: 0.10). The required workflow ingests HuggingFace datasets and parses third-party files (videos, text, PDFs, images) and user formats via the SDK, which processes untrusted content.

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

• Potentially malicious external URL detected (high risk: 0.70). The installation instruction in SKILL.md executes a remote shell script via curl -fsSL https://deeplake.ai/install.sh | bash, which constitutes an unverifiable runtime external dependency from an unofficial or external publisher domain.

MEDIUM W013: Attempt to modify system services in skill instructions.

• Attempt to modify system services in skill instructions detected (high risk: 1.00). The skill prompt instructs the agent to install optional packages such as ffmpeg using apt-get with sudo privileges (sudo apt-get install ffmpeg), which compromises the machine's state by asking for root privileges and modifying system packages.