embedded-captions
Pass
Audited by Gen Agent Trust Hub on Oct 13, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill downloads specialized media processing components from reputable sources. It fetches the u2net_human_seg ML model for background removal and uses uvx to execute pinned versions of the WhisperX transcription tool. Additionally, the GSAP library is referenced from the jsDelivr CDN, and the skill includes logic to verify its integrity.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted media and transcription data from source videos. While this represents an ingestion surface, the data is primarily used for visual rendering and the risk of influencing agent behavior is minimized by the scoped execution environment and standard HTML escaping during composition.
- [COMMAND_EXECUTION]: Subprocess calls are made to standard utilities like FFmpeg, ffprobe, and python3 for media analysis and asset generation. These commands are executed using secure array-based argument passing to prevent injection, and are necessary for the skill's intended functionality.
Audit Metadata