youtube-downloader

Fail

Audited by Gen Agent Trust Hub on Nov 9, 2026

Risk Level: HIGHPRIVILEGE_ESCALATIONCOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [PRIVILEGE_ESCALATION]: The skill attempts to acquire administrative privileges by executing sudo apt update && sudo apt install -y yt-dlp during the installation check in SKILL.md.
  • [COMMAND_EXECUTION]: The script constructs shell commands using the output of yt-dlp --print "%(title)s". While it attempts to sanitize the title using tr, it only targets specific characters (/, :, ?, ") and fails to filter shell metacharacters such as backticks (`), dollar signs ($), or semicolons (;). An attacker could provide a video with a malicious title to execute arbitrary shell commands.
  • [DYNAMIC_EXECUTION]: The skill uses python3 -c and heredocs to process transcript data. It interpolates shell variables like $VIDEO_TITLE and $VTT_FILE directly into these Python scripts. If a video title contains Python escape characters or triple quotes, it could lead to arbitrary code execution within the Python environment.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes external data in the form of YouTube transcripts which could contain instructions designed to influence agent behavior.
  • Ingestion points: Raw subtitle content downloaded via yt-dlp in SKILL.md.
  • Boundary markers: None. The transcript content is processed and written directly to markdown files without delimiters or warnings.
  • Capability inventory: File system writes, shell command execution, and privilege escalation via sudo.
  • Sanitization: Minimal text cleanup is performed (removing HTML-like tags and filler words), but no sanitization exists to detect or strip embedded instructions.
  • [EXTERNAL_DOWNLOADS]: The skill automates the installation of the yt-dlp package using system package managers (brew, apt) and pip. While yt-dlp is a well-known tool, performing system-wide installations automatically represents a significant security footprint.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Nov 9, 2026, 12:32 PM