Skip to content

Security: minewefu/afterglow

Security

SECURITY.md

Security policy

Design stance

Afterglow's attack surface is deliberately small:

  • No kernel driver. All hardware access is through NVIDIA's userland driver libraries (nvml.dll, nvapi64.dll) that ship with every GeForce driver. Afterglow cannot be used as a bring-your-own-vulnerable-driver (BYOVD) primitive because it doesn't bring one.
  • No network access. Afterglow makes no network requests: no telemetry, no update checks, no accounts. The only external binary is Intel's MIT-licensed PresentMon console app, bundled with a pinned SHA-256 that CI verifies (THIRD_PARTY.md).
  • Elevation is scoped. Admin rights are required only because the driver requires them for writes (clocks/fans/power) and ETW tracing; monitoring runs unelevated.
  • Settings/profiles are plain JSON under %ProgramData%\Afterglow — no registry writes except the optional Run key created by the installer's startup task.

Known trust boundary

%ProgramData% is writable by standard users by design, and the elevated Afterglow process reads profiles/settings from there. This is mitigated rather than eliminated: every value loaded from disk is schema-validated and then clamped to the driver-reported legal range before any write reaches the GPU, so a tampered profile cannot push the hardware beyond what the driver itself allows any tool to set. Hardening this further (ACL-restricted state directory) is tracked for a future release; treat local standard-user tampering as within the residual risk envelope of every tool in this category.

Reporting a vulnerability

Please open a GitHub security advisory (Security → Report a vulnerability) or a private report to the maintainer rather than a public issue. Include reproduction steps and impact. You should get a response within a week; fixes for confirmed issues in the driver-write or elevation paths are prioritized above everything else.

Supported versions

The latest release is supported. Older releases receive fixes only for vulnerabilities that allow privilege escalation or arbitrary code execution.

There aren't any published security advisories