last30days

Fail

Audited by Gen Agent Trust Hub on Apr 26, 2026

Risk Level: HIGHCREDENTIALS_UNSAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
  • [CREDENTIALS_UNSAFE]: The script scripts/lib/chrome_cookies.py attempts to retrieve the Chrome Safe Storage master key from the macOS Keychain via the security command. Additionally, scripts/lib/safari_cookies.py and scripts/lib/cookie_extract.py automate the extraction of sensitive authentication cookies from local browser databases for various social media platforms.
  • [PROMPT_INJECTION]: The SKILL.md file contains instructions (referred to as 'LAWs') that explicitly command the agent to override or supersede default tool behaviors and constraints, specifically targeting the citation requirements of the WebSearch tool. It also uses self-referential checks to force the agent to follow its internal formatting rules over global system guidelines.
  • [EXTERNAL_DOWNLOADS]: The scripts/lib/setup_wizard.py script includes logic to automatically download and install the yt-dlp utility using the brew package manager at runtime if the dependency is missing.
  • [COMMAND_EXECUTION]: Multiple scripts execute arbitrary system commands using the subprocess module, including sensitive utilities like security, openssl, gh, and node, often with user-derived inputs or to facilitate the extraction of encrypted host data.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Apr 26, 2026, 01:53 PM