Skip to content

Security: wavyrai/MacSpark

SECURITY.md

Security policy

Scope

MacSpark installs no service by itself, but unsafe documentation can still cause credential disclosure, remote lockout, excessive privilege, or network exposure. Security reports may cover instructions, examples, agent skills, and repository automation.

Reporting

Use GitHub's private security advisory form. Do not open a public issue for a credential leak, authentication bypass, command injection, unsafe privilege escalation, or procedure that exposes a service.

Describe the affected file, impact, exact version context, and a sanitized reproduction. Never include real tokens, hostnames, tailnet names, usernames, IP addresses, device identifiers, or model credentials.

This is a volunteer publication and cannot promise a response or embargo window.

Threat model

The guide assumes:

  • the Mac, DGX, and their operator accounts are trusted;
  • the DGX is reachable only through a controlled LAN, Tailscale, or SSH tunnel;
  • Tailscale policy or equivalent firewall rules enforce least privilege;
  • Docker users are effectively privileged administrators;
  • models, repositories, installers, images, and weights are third-party code and data;
  • plain HTTP endpoints receive transport confidentiality only when carried through an encrypted SSH or Tailscale path. A trusted-LAN firewall alone does not encrypt prompts, responses, or bearer tokens.

Do not expose model APIs, gateways, SMB, VNC, dashboards, or Docker to the public internet. Prefer application authentication in addition to network isolation, and rotate any credential that appears in shell history, logs, screenshots, or shared URLs.

There aren't any published security advisories