30x-web-to-video

Fail

Audited by Gen Agent Trust Hub on Aug 27, 2026

Risk Level: HIGHPRIVILEGE_ESCALATIONREMOTE_CODE_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [PRIVILEGE_ESCALATION]: The script scripts/url-to-video.ts contains logic to automatically execute sudo apt-get install to install system tools like ffmpeg and aubio-tools on Linux-based systems without requiring manual user intervention beyond a flag.
  • [REMOTE_CODE_EXECUTION]: The skill performs automatic installations of unverifiable dependencies at runtime using package managers:
  • scripts/url-to-video.ts executes npm install -g playwright.
  • scripts/analyze-audiomap.py and scripts/generate-bgm.py use pip install to install multiple Python packages including librosa, transformers, and torch.
  • [DYNAMIC_EXECUTION]: scripts/regenerate-scene.ts implements a dynamic code modification loop. It invokes a sub-agent via claude -p to rewrite the src/MainVideo.tsx file based on LLM critique. Crucially, the sub-agent is granted broad tool permissions including Bash, Edit, and Read, and uses the --dangerously-skip-permissions flag.
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests content from arbitrary user-provided URLs during the 'harvesting' phase. This untrusted data is then used as context for the sub-agent's code regeneration loop, creating a path for a malicious website to influence the generated code or the sub-agent's execution environment.
  • [COMMAND_EXECUTION]: Several scripts, including scripts/url-to-video.ts, scripts/analyze-audiomap.py, and scripts/generate-bgm.py, execute arbitrary shell commands via execFileSync and subprocess.run to manage the underlying media toolchain.
  • [EXTERNAL_DOWNLOADS]: The skill routinely downloads external resources:
  • yt-dlp is used to fetch video and audio content from YouTube and Vimeo.
  • The transformers library is instructed to download large model weights (approx. 1.5GB) from Hugging Face during the first run of the BGM generator.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Aug 27, 2026, 06:57 PM
Security Audit — agent-trust-hub — 30x-web-to-video