vox-explainer

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

• Third-party content exposure detected (medium risk: 0.30). In “source mode” the workflow explicitly reads every user-supplied file and fetches every supplied URL, then uses the extracted free text (via “storyline mine” and receipts) to draft the script and overlays, exposing the agent to outsider-authored prompt-injection content embedded in those documents/links.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill spawns npx to fetch and run the HyperFrames package at runtime (e.g., "npx --yes hyperframes@0.8.14 preview" and the documented "npx skills add heygen-com/hyperframes"/"npx hyperframes init"), which downloads and executes remote code from the npm registry as a required step for the seam preview/verification flow.