browser-use

Warn

Audited by Gen Agent Trust Hub on Jun 28, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONDATA_EXFILTRATIONCREDENTIALS_UNSAFEEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill provides the browser-use python command, which allows the execution of arbitrary Python statements on the host system (as seen in references/cdp-python.md).
  • [COMMAND_EXECUTION]: The browser-use eval command allows the execution of arbitrary JavaScript code within the browser context, which can be used to manipulate web pages or exfiltrate data from them.
  • [DATA_EXFILTRATION]: The skill includes tools to retrieve, export, and import browser cookies (browser-use cookies get, browser-use cookies export), creating a risk for session hijacking and credential theft.
  • [DATA_EXFILTRATION]: The browser-use tunnel command allows users to expose local ports via Cloudflare tunnels, which could unintentionally expose internal services to the public internet.
  • [CREDENTIALS_UNSAFE]: The skill facilitates access to existing Chrome profiles and persistent login sessions via the --profile and connect commands, potentially exposing sensitive personal or corporate data to the agent.
  • [PROMPT_INJECTION]: The skill is highly susceptible to indirect prompt injection because it ingests untrusted data from the web (via state, get html, get text) and possesses powerful capabilities to execute scripts and manage credentials.
  • Ingestion points: browser-use state, browser-use get html, and browser-use get text in SKILL.md.
  • Boundary markers: No explicit markers or warnings are defined in the instructions to prevent the agent from following instructions found on web pages.
  • Capability inventory: Subprocess execution via browser-use python, JavaScript execution via browser-use eval, cookie access via browser-use cookies, and network exposure via browser-use tunnel.
  • Sanitization: There is no evidence of data sanitization or validation of content retrieved from the browser before it enters the agent's context.
  • [EXTERNAL_DOWNLOADS]: The instructions suggest that users install an external skill (x402) from a remote GitHub repository to handle cryptocurrency payments for cloud services.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Jun 28, 2026, 01:28 PM
Security Audit — agent-trust-hub — browser-use