The @vanexalabs-ai/vanexa-agent package implements a daemon that pairs with a remote peer (such as a phone) over WebRTC using specific signaling relays. Commands received over the DataChannel are executed on the host via child_process.spawn/exec, allowing full remote code execution by anyone controlling the signaling relay or paired peer. The runtime dispatch logic is shipped as V8 bytecode, preventing review of pairing and authorization semantics. The package suppresses detection by ignoring categories related to malware, obfuscation, shell access, and network access. The malicious shell execution activates only when the operator explicitly runs the agent and completes pairing, bounding the attack surface to those hosts.

An attacker controlling the signaling relay or a paired remote peer can achieve full remote code execution on the host running the agent. This allows arbitrary command execution with the privileges of the agent process, potentially leading to full system compromise on affected hosts where the agent is actively running and paired.

No official patch or remediation is currently documented. Operators should avoid running the @vanexalabs-ai/vanexa-agent daemon to prevent exposure. Since the malicious behavior activates only when the agent is explicitly started and paired, not at install or require time, avoiding execution of the daemon effectively mitigates risk. Monitor for and remove this package from environments where it is not explicitly required or trusted. Patch status is not yet confirmed — check the vendor advisory or trusted sources for updates.