The mangomind-agent npm package (versions 0.1.0 through 0.2.1, specifically listed) includes a backdoor that connects to wss://relay.mangomindbd.com. The package listens for messages from this relay and executes commands from the relay operator via child_process.execSync, allowing arbitrary command execution. It also proxies requests to a local AI coding agent, granting the relay operator filesystem and shell access. The only gate is a token handshake controlled by the relay operator, meaning full remote control is possible for anyone controlling the relay server. This is a malicious package rather than a typical vulnerability.

Any host running the affected versions of mangomind-agent is subject to full remote code execution and arbitrary command execution by the operator of the remote relay server. This includes the ability to execute shell commands and interact with a local AI coding agent with filesystem access. The compromise is complete for the user context running the agent, posing a severe security risk including potential data theft, system manipulation, or further lateral movement.

No official patch or remediation is currently documented. Users should immediately remove the mangomind-agent package from their environments and avoid installing or running it. Since the malicious behavior is triggered by communication with a hardcoded remote relay, blocking network access to relay.mangomindbd.com may reduce risk but does not remove the malicious code. Monitor for presence of this package and replace it with trusted alternatives. Patch status is not yet confirmed — check the vendor advisory or trusted sources for updates.