paddleocr

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

• Third-party content exposure detected (low risk: 0.10). The required workflow OCRs a user-supplied PDF or page images, which are outsider-authored papers or documents the agent selects but cannot be forced to receive, and then reads the resulting Markdown, so this is low exposure with no URL source.

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

• Potentially malicious external URL detected (medium risk: 0.30). The setup instruction npx skills add uxfion/skills --skill vllm-serve -g fetches executable skill material, including the serve.py that this skill runs, from an individual's GitHub repository, which is unofficial but not shown to be suspicious, so the tier is medium (the PaddlePaddle --find-links index is an official vendor source and is strong, so it does not raise the tier).