varg-unboxing-video
Pass
Audited by Gen Agent Trust Hub on Nov 1, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes external image files via URLs which constitutes a potential indirect prompt injection attack surface. * Ingestion points: The skill accepts image URLs (BOX_URL, PRODUCT_URL) as inputs for processing in SKILL.md. * Boundary markers: Uses explicit roles (@image1, @image2) in the generation prompt to distinguish data references from instructions. * Capability inventory: The skill utilizes curl for network requests, ffmpeg and ffprobe for media processing, and standard file Read, Write, and Edit tools. * Sanitization: No explicit input sanitization or validation logic is defined within the instructions.
- [EXTERNAL_DOWNLOADS]: Fetches data from the official vendor domain (varg.ai) and references marketing research from well-known industry sources such as runneth.com and influee.co.
- [COMMAND_EXECUTION]: Employs curl for API communication and ffmpeg for media assembly, which are standard for the skill's intended purpose and correctly defined within the allowed-tools configuration.
Audit Metadata