advanced-video-downloader

Fail

Audited by Gen Agent Trust Hub on Feb 17, 2026

Risk Level: HIGHDATA_EXFILTRATIONCOMMAND_EXECUTIONPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [DATA_EXFILTRATION] (HIGH): The skill documentation in docs/README.md and references/supported_platforms.md explicitly instructs the use of the --cookies-from-browser flag. This flag causes yt-dlp to read sensitive authentication cookies from the user's browser profile, exposing them to the skill's environment and the AI agent.\n- [DATA_EXFILTRATION] (MEDIUM): The script scripts/transcribe_siliconflow.py transmits local audio and video files to an external API endpoint (https://api.siliconflow.cn/v1/audio/transcriptions) for processing. Users should be aware that their media is being sent to a third-party service.\n- [COMMAND_EXECUTION] (MEDIUM): The skill is built around executing complex shell commands with yt-dlp. This provides a significant attack surface for interacting with the local file system and browser data through the tool's interface.\n- [EXTERNAL_DOWNLOADS] (LOW): The documentation recommends downloading and installing external binaries like yt-dlp and ffmpeg from external sources. These are trusted tools in this domain, but remain external dependencies.\n- [PROMPT_INJECTION] (LOW): The transcription workflow creates an indirect prompt injection surface where untrusted data from the transcription API is written into Markdown files without sanitization. \n
  • Ingestion points: API transcription result in scripts/transcribe_siliconflow.py. \n
  • Boundary markers: Absent in the generated Markdown. \n
  • Capability inventory: Subprocess execution via yt-dlp. \n
  • Sanitization: None; external API text is directly interpolated into the output.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Feb 17, 2026, 05:41 PM