browser-screenshot
Pass
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: SAFEDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DATA_EXFILTRATION]: The skill allows the agent to navigate to any URL provided as an argument. This could be used to access internal services (e.g., cloud metadata APIs at 169.254.169.254) or local files via
file://protocols if the browser environment is not isolated. The resulting screenshots could then be used to exfiltrate sensitive information from restricted environments. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from external URLs.
- Ingestion points: The target URL is passed directly as a command-line argument to the
scripts/screenshot.pyscript. - Boundary markers: There are no boundary markers or sanitization logic for the input URL.
- Capability inventory: The script has the capability to perform network operations (browser navigation) and write files to the disk (saving screenshots via
page.screenshot). - Sanitization: The URL and CSS selectors provided to the script are not validated or sanitized, allowing for potential manipulation of the browser automation process by an attacker-controlled page or via malicious URL schemes.
Audit Metadata