youtube-transcribe-skill

Pass

Audited by Gen Agent Trust Hub on Sep 21, 2026

Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses yt-dlp to interact with YouTube. It executes shell commands to retrieve video titles and download subtitles, specifically using the --cookies-from-browser flag to access the user's local browser data.
  • [DATA_EXPOSURE]: The skill explicitly instructs the agent to access sensitive user data by reading browser cookies (defaulting to Chrome) via the yt-dlp --cookies-from-browser=chrome command. While used for the stated purpose of bypassing sign-in restrictions, this grants the process access to the user's active session data for the YouTube domain.
  • [INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests untrusted data from external YouTube pages.
  • Ingestion points: Browser automation snapshots (take_snapshot) and DOM extraction (evaluate_script) read the video description and transcript panel content into the agent's context.
  • Boundary markers: None. The instructions do not include delimiters or warnings to ignore commands embedded within the video content.
  • Capability inventory: The skill has the ability to execute shell commands (yt-dlp) and write files to the local system.
  • Sanitization: None. The extracted text is saved directly to a file and reported to the user without filtering.
  • [DYNAMIC_EXECUTION]: The skill utilizes browser automation to execute a JavaScript snippet via evaluate_script for DOM scraping. While the script itself is provided in the instructions, this mechanism executes code within the context of the opened web page.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 21, 2026, 03:11 PM