vdms-dataprep-user

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

• Third-party content exposure detected (medium risk: 0.30). SKILL.md describes runtime ingestion via first-party user-controlled HTTP requests to localhost endpoints (e.g., POST /v1/dataprep/videos/upload and POST /v1/dataprep/summary), so outsider-authored free text fields like video_summary can be submitted directly into the required workflow without selecting a specific item.

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

• Potentially malicious external URL detected (high risk: 1.00). The skill instructs runtime fetching from https://raw.githubusercontent.com/open-edge-platform/edge-ai-libraries/main/microservices/visual-data-preparation-for-retrieval/vdms (e.g., curl $RAW/setup.sh) and then sources setup.sh, meaning remote code from raw.githubusercontent.com is downloaded and executed at runtime.