browser-use

Warn

Audited by Gen Agent Trust Hub on Mar 12, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSCREDENTIALS_UNSAFE
Full Analysis
  • [COMMAND_EXECUTION]: The browser_engine.py script contains _run_cmd and _pip_install functions that utilize subprocess.check_call to execute shell commands and install Python packages. This is used for environment setup, such as installing Playwright, Patchright, and Camoufox binaries.
  • [REMOTE_CODE_EXECUTION]: The evaluate action in actions.py allows the agent to execute arbitrary JavaScript code within the browser context. While gated by an environment variable (BROWSER_USE_EVALUATE), this provides a powerful mechanism for page manipulation and data extraction.
  • [EXTERNAL_DOWNLOADS]: The skill automatically downloads browser binaries (Chromium, Firefox, and the specialized CloakBrowser) and third-party Python packages from PyPI and GitHub during initialization or when specific stealth tiers are activated.
  • [CREDENTIALS_UNSAFE]: The skill manages persistent browser identities, saving cookies, localStorage, and a credentials.json file within the user's home directory (~/.browser-use/profiles). It implements a "dual-mode" injection system to automatically fill secrets into web forms.
  • [DATA_EXFILTRATION]: Actions such as cookies_export, storage_get, and extract (HTML to Markdown conversion) provide built-in mechanisms for retrieving and exporting sensitive session data from the browser to the local filesystem or agent context.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Mar 12, 2026, 06:00 PM