video-sound
Pass
Audited by Gen Agent Trust Hub on Nov 14, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill utilizes
ffmpegandffprobeto perform audio analysis (loudness, silence detection, duration) and processing (mixing, fit, and muxing). Scripts such assound.py,audition.py, andjudge_takes.pyexecute these tools using thesubprocessmodule with argument lists. - [EXTERNAL_DOWNLOADS]: The
fetch.shscript and theaudition.pytool perform network requests to download audio assets. These operations target vendor-owned infrastructure ats3.varg.aiandapi.varg.aito retrieve generated content such as voice lines and music beds. - [DYNAMIC_EXECUTION]: FFmpeg filter graphs are constructed at runtime in
sound.pyandjudge_takes.pyto handle tasks such as loudness normalization, audio crossfading, and visual waveform generation. - [INDIRECT_PROMPT_INJECTION]: The skill implements a mechanism for ranking audio takes using an LLM judge. By interpolating user-supplied text and transcripts of audio recordings into prompts, it creates an attack surface where instructions embedded in the input data could potentially influence the judge's verdict.
Audit Metadata