screens-to-video
Pass
Audited by Gen Agent Trust Hub on Nov 20, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses Python and Node.js subprocesses to execute external command-line utilities including
ffmpeg,ffprobe,curl, and therevylCLI for image processing, media metadata extraction, and data retrieval. - Evidence: Found in
scripts/arena.py,scripts/crop_drops.py, andscripts/render.mjs. - [EXTERNAL_DOWNLOADS]: The skill downloads media assets and metadata from
api.are.naand retrieves visual look-and-feel references from specific public channels. It also communicates with the Stewdio platform (stewdio.ai) to build the final film and manage assets. - Evidence:
scripts/arena.pycontains logic to fetch channel contents and download files fromare.naURLs. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content (titles and descriptions) from the
mobile-motionare.na channel, which is then used as part of the context for generating the video's script and beats. - Ingestion points:
scripts/arena.pyretrieves block titles and media information from the external are.na API. - Boundary markers: The
SKILL.mdinstructions do not specify explicit delimiters or instructions to ignore malicious content embedded within these reference descriptions. - Capability inventory: The skill has the ability to execute shell commands, perform network operations, and interact with the Stewdio MCP server.
- Sanitization: No sanitization of the strings retrieved from the external API was found before they are incorporated into the creative brief.
- [DYNAMIC_EXECUTION]: The
scripts/render.mjsscript performs a dynamic import of the Playwright library from a path determined at runtime (either via thePLAYWRIGHT_DIRenvironment variable or a default home directory path). This is used to render a local HTML-based animation timeline. - Evidence:
scripts/render.mjsusesawait import(...)with a computed path string.
Audit Metadata