agent-browser
Pass
Audited by Gen Agent Trust Hub on Mar 20, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSPROMPT_INJECTIONCOMMAND_EXECUTIONCREDENTIALS_UNSAFE
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill references and provides instructions to install the
agent-browserCLI tool from a repository managed by a well-known organization. - [PROMPT_INJECTION]: The skill manages potential indirect prompt injection from untrusted web data. 1. Ingestion points: External websites accessed via the
opencommand. 2. Boundary markers: OptionalAGENT_BROWSER_CONTENT_BOUNDARIESenvironment variable. 3. Capability inventory: Bash access, file writing, and network navigation. 4. Sanitization: SupportsAGENT_BROWSER_ALLOWED_DOMAINSfor navigation restriction andAGENT_BROWSER_ACTION_POLICYto limit high-risk interactions like JavaScript evaluation. - [COMMAND_EXECUTION]: Browser interactions are performed via a specialized CLI. Security is maintained through an action policy that can deny sensitive actions like
eval,download, ornetworkrequests. - [CREDENTIALS_UNSAFE]: The documentation provides safe patterns for authentication, such as using environment variables and saving session state from files, explicitly warning against hardcoding secrets in scripts.
Audit Metadata