bestblogs-transcribe-youtube

Warn

Audited by Gen Agent Trust Hub on Mar 12, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONCREDENTIALS_UNSAFEPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The scripts/transcribe.ts file uses child_process.execSync to run osascript, which allows for the execution of arbitrary AppleScript commands to control macOS applications like Google Chrome.\n- [CREDENTIALS_UNSAFE]: The script programmatically extracts the SNlM0e session token from a user's active Gemini tab in Chrome. This token is a sensitive credential used to authenticate requests to Google's backend services.\n- [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it processes content from YouTube URLs without sanitization or boundary markers. Adversarial content within a video's metadata or transcript could be used to influence the Gemini model's output.\n
  • Ingestion points: YouTube URL parameter in scripts/transcribe.ts.\n
  • Boundary markers: None present in the constructed API request when targeting the Gem ID.\n
  • Capability inventory: Shell command execution via AppleScript and local file writing via Node.js fs module.\n
  • Sanitization: No semantic validation or filtering of content retrieved from the external YouTube source is performed.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Mar 12, 2026, 04:31 AM