NYC

Chrome Bridge Automation

Warn

Audited by Gen Agent Trust Hub on Feb 17, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • REMOTE_CODE_EXECUTION (MEDIUM): The skill uses 'npx @midscene/web@1' to download and run code from the npm registry. As @midscene is not a trusted organization, this constitutes unverifiable remote code execution.
  • DATA_EXFILTRATION (MEDIUM): The skill connects to the user's real browser, accessing cookies and login states. While intended, this provides a pathway for data exposure if the agent is directed to sensitive URLs.
  • COMMAND_EXECUTION (LOW): Relies on Bash to execute CLI commands for automation.
  • PROMPT_INJECTION (LOW): Indirect prompt injection surface exists as the skill processes untrusted web content. * Ingestion points: External URLs and page screenshots. * Boundary markers: Absent. * Capability inventory: Element interaction (act) and shell execution. * Sanitization: None.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Feb 17, 2026, 05:21 PM