skills/garrytan/gstack/browse/Gen Agent Trust Hub

browse

Pass

Audited by Gen Agent Trust Hub on Nov 16, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill facilitates the execution of well-defined browser commands (e.g., navigating, clicking, form filling) through controlled binaries. These operations are essential to its function and are executed within a restricted and monitored environment.
  • [EXTERNAL_DOWNLOADS]: The skill can fetch web content and download files. It implements rigorous validation (validateNavigationUrl and validateOutputPath) to prevent access to sensitive local or cloud infrastructure and ensures file operations are confined to safe, authorized directories.
  • [DATA_EXPOSURE_AND_EXFILTRATION]: Robust privacy measures are in place to prevent data leakage. The filterArgs system automatically redacts sensitive information, such as passwords and authentication tokens, from command logs and telemetry. A 'dual-listener' architecture effectively isolates local control surfaces from those accessible via external tunnels.
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to handle untrusted web content safely. It employs 'datamarking' and 'hidden element stripping' to identify and remove potentially malicious instructions embedded in web pages before they are presented to the AI agent.
  • [DYNAMIC_EXECUTION]: JavaScript execution is supported for legitimate browsing tasks but is strictly governed by scope and domain checks. Outputs from these operations are consistently wrapped in 'untrusted content' markers to ensure they are handled appropriately by the agent.
  • [PRIVILEGE_ESCALATION]: No evidence was found of attempts to acquire unauthorized permissions. The skill operates within its own defined scope and follows established security protocols for all system and browser interactions.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 16, 2026, 08:27 PM