browse
Pass
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
- [EXTERNAL_DOWNLOADS]: Neutrally fetches machine learning models from HuggingFace for prompt injection detection and downloads browser binaries via Playwright (trusted sources).
- [COMMAND_EXECUTION]: Executes shell commands to manage the browser daemon, clean up orphaned processes, and run browser-skill scripts with environment scrubbing for security.
- [DYNAMIC_EXECUTION]: Implements JavaScript execution within the browser context for inspection and interaction. Uses Node.js workers to securely handle platform-specific credential decryption on Windows.
- [INDIRECT_PROMPT_INJECTION]: Implements a comprehensive multi-layer defense stack (datamarking, hidden element stripping, URL blocklisting, and ML-based classification) to mitigate the risk of processing malicious instructions embedded in web content.
- [DATA_EXFILTRATION]: Provides functionality to import browser cookies and session storage. Implements strict origin guards and extensive redaction of sensitive tokens and passwords in activity logs to prevent unauthorized exposure.
Audit Metadata