extracting-brand-kits
Pass
Audited by Gen Agent Trust Hub on Nov 11, 2026
Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: The skill executes
ffmpegcommands locally to extract frames, perform scene detection, and generate contact sheets from reference video files. These operations are core to the skill's purpose of visual analysis. - [DYNAMIC_EXECUTION]: The agent authors custom HTML and CSS to reproduce visual motifs (cards). The instructions include a specific security constraint to exclude
<script>tags and@keyframes, ensuring the generated content remains inert and safe for rendering. - [INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect injection as it processes external, untrusted video and image content.
- Ingestion points: User-provided reference videos and images processed in
SKILL.md. - Boundary markers: The agent is instructed to use manual visual curation and avoid prose summarization, which limits the influence of embedded text in the reference material.
- Capability inventory: Local command execution (
ffmpeg), file uploads to storage via presigned URLs, and API interactions with the CueFrame platform. - Sanitization: Explicit restrictions against generating executable HTML tags.
- [EXTERNAL_DOWNLOADS]: The instructions direct the agent to download reference videos from user-provided URLs when necessary to perform the extraction process.
Audit Metadata