browser-use

Pass

Audited by Gen Agent Trust Hub on Jun 19, 2026

Risk Level: SAFECOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONCREDENTIALS_UNSAFEPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill documents the browser-use python command, which enables the execution of Python statements to interact with the browser session.
  • [REMOTE_CODE_EXECUTION]: Provides recipes for executing JavaScript via Runtime.evaluate and Python code that interacts with internal browser objects like browser._session and browser._run in references/cdp-python.md.
  • [CREDENTIALS_UNSAFE]: Documents how to extract browser cookies using the Network.getCookies command, which provides access to sensitive session data.
  • [PROMPT_INJECTION]: The skill processes untrusted data from the browser (Ingestion: browser window content). While it includes boundary rules (Boundary: Rule 3 in SKILL.md), it provides powerful execution capabilities (Capabilities: Python execution, JS evaluation in references/cdp-python.md). No explicit sanitization of web content is mentioned beyond high-level instructions to the agent (Sanitization: None).
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 19, 2026, 02:29 PM
Security Audit — agent-trust-hub — browser-use