windows-host-browser

Warn

Audited by Gen Agent Trust Hub on Oct 12, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONDYNAMIC_EXECUTIONDATA_EXFILTRATIONPRIVILEGE_ESCALATIONPERSISTENCEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses SSH to execute remote commands on both a WSL environment and a Windows host, including the use of schtasks, pkill, and complex PowerShell scripts through the -EncodedCommand transport.\n- [DYNAMIC_EXECUTION]: The cdp.py script contains an eval command that permits the execution of arbitrary JavaScript within the Chrome extension's service worker context via the Chrome DevTools Protocol Runtime.evaluate method.\n- [DATA_EXFILTRATION]: The skill is explicitly designed to access sensitive data behind active logins (such as Vercel, Google Cloud, and Heroku) and can exfiltrate network egress information through external lookups to ip-api.com.\n- [PRIVILEGE_ESCALATION]: connect.sh attempts to modify network interface MTU settings using ip link set, and the skill routinely manipulates Windows Scheduled Tasks via Set-ScheduledTask.\n- [PERSISTENCE]: The skill establishes persistence on the Windows host by creating a persistent desktop shortcut (Agent Chrome.lnk) configured to launch background PowerShell processes and manages long-lived scheduled tasks.\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the browser, such as tab titles and URLs, which serves as a potential attack surface for indirect prompt injection if the agent processes this data without sanitization.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Oct 12, 2026, 02:40 PM