The Xbox agent:

• listens on every Xbox network interface on TCP port 4380;
• has no user authentication;
• has no transport encryption;
• exposes memory and MMIO reads;
• can be explicitly armed for raw reads, dumps, writes, title control, and arbitrary x86 code execution;
• runs with the privileges available to an Xbox title.

The ARM I_ACCEPT_THE_RISK phrase only reduces accidental execution. Anyone who can reach the port can send it. It is not a password and provides no protection from a malicious or compromised device on the same network.

• Use only on a private, trusted, non-internet-routed LAN.
• Do not configure port forwarding, a public tunnel, UPnP exposure, or cloud relay.
• Prefer a dedicated switch or isolated VLAN with only the operator workstation and Xbox.
• Back up irreplaceable console storage before use.
• Stop the agent and return to the dashboard as soon as the measurement session is complete.
• Review all addresses and payload source before running them.
• Do not use raw EXEC with code from an untrusted source.
• Do not assume a read is harmless. Hardware registers can have side effects.
• Do not publish raw dumps or unsanitized results.

The R, RB, and W handlers enforce address allowlists. W refuses flash ranges. All raw access and title-control commands require session arming. These checks do not protect against:

• a bug in the agent or Xbox kernel;
• side effects from allowed MMIO;
• a malicious or incorrect EXEC payload;
• another device reaching port 4380;
• storage writes performed by a launched XBE or uploaded payload;
• hardware differences on untested Xbox revisions.

Open a GitHub security advisory or contact the repository owner privately. Do not include console keys, credentials, private dumps, or personal hardware identifiers in a report.

This project is licensed under the MIT License and provided “AS IS”, WITHOUT WARRANTY OF ANY KIND. See LICENSE.