Malicious code in @vanexalabs-ai/vanexa-agent (npm)

The @vanexalabs-ai/vanexa-agent npm package contains intentionally malicious code that establishes a WebRTC daemon paired with remote peers (such as phones) using signaling relays at https://vanexa-agent-relay.workers.dev and https://vanexa-agent-relay.tubefood.workers.dev. The daemon listens for commands over WebRTC DataChannel and executes them via child_process.spawn/exec on the host machine, providing full remote code execution to anyone controlling the signaling relay or a paired peer.\n\nThe package deliberately obfuscates its runtime dispatch logic by shipping it only as V8 bytecode (dist/bundle.jsc loaded via bytenode), with the fallback .cjs file minified. The developers frame this obfuscation as 'commercial IP protection,' but it prevents security review of the pairing and authorization semantics. A socket.json file at the tarball root explicitly suppresses security scanning across all relevant categories: aiMalware, gptMalware, obfuscatedCode, usesEval, nativeCode, shellAccess, filesystemAccess, networkAccess, bypassesEnvironment, and installScripts.\n\nThe malicious behavior is triggered only when the agent is explicitly executed by an operator (not during npm install or require), which bounds the blast radius to systems where the package is actively run. However, any host running the agent becomes vulnerable to remote code execution from the relay operators or paired peers.