The vanexa-agent package, when started, opens a persistent WebSocket connection to a hardcoded Cloudflare Workers relay URL. It processes incoming 'task_request' messages by executing shell commands on the host via spawn('/bin/bash', ['-c', command]) or equivalent on Windows. The only authentication is a 6-digit pairing code used as a session ID, with no JWT or message signing, allowing anyone controlling the relay or guessing the code to execute arbitrary commands. Configuration updates and sensitive data, including encrypted API keys and hostname, are sent to the relay. The package's README misleadingly advertises LAN-only communication, but the code forces WAN relay usage. This design constitutes a malicious backdoor enabling remote code execution on affected hosts.

An attacker controlling the relay server or guessing the 6-digit session code can execute arbitrary shell commands on any host running the vulnerable vanexa-agent versions. This leads to full remote code execution, potential data exfiltration, system compromise, and lateral movement within affected environments. Sensitive configuration and encrypted API keys are also exposed to the attacker. The vulnerability effectively creates a backdoor with minimal authentication and no cryptographic protections.

No official patch or fix is currently documented. Users should immediately stop running the vanexa-agent daemon and remove the package from their systems. Avoid connecting to the hardcoded relay URL and do not use the package until a vendor advisory or update is available. Monitor for any unauthorized activity on affected hosts. Patch status is not yet confirmed — check the vendor advisory for current remediation guidance.