hwc-media-content
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill contains patterns that process untrusted data and render it directly into the DOM, creating a vulnerability surface for indirect prompt injection or cross-site scripting (XSS).
- Ingestion points: Untrusted data is ingested from file input metadata (file names and sizes) in
references/2024-09-17-stimulus-image-upload-previews.mdand from third-party library event objects (marker descriptions and innerText) inreferences/2024-07-02-stimulus-wavesurfer-add-markers.md. - Boundary markers: The implementation does not utilize delimiters or specific instructions to isolate untrusted content during prompt interpolation or UI rendering.
- Capability inventory: Stimulus controllers use powerful DOM manipulation methods, including
innerHTMLandinsertAdjacentHTML, across multiple files to dynamically build the interface. - Sanitization: Content is interpolated into HTML strings via manual replacement or template literals without providing robust escaping for variables like
file.nameorregion.content.innerText.
Audit Metadata