The engineering problem

Virtualization depends on precise contracts between machine state, memory layout, and exit handling. Wraith makes those contracts explicit with a small custom guest before attempting a general-purpose operating system.

Two implementation paths

The Linux path opens KVM, allocates guest memory, loads ELF32 segments, and configures a protected-mode virtual CPU. A freestanding C kernel and assembly entry point provide a controlled guest workload. The separate UEFI path works closer to AMD SVM: VMCB state, nested paging, intercept dispatch, and a guest-to-host call interface.

Why the design matters

Keeping the guest small makes register and memory transitions easier to reason about. Separating the KVM and firmware paths exposes which responsibilities the operating system handles and which a bare-metal monitor must implement itself.

Evidence to inspect

The source is organized around wraith.c, guest_kernel/, and WraithPkg/. The Linux host and guest compile with the project build. Runtime behavior on AMD hardware, UEFI boot behavior, and isolation against an adversarial guest need dedicated validation.

Current scope

Research prototype. Running an unmodified Linux guest and production-grade isolation are not established capabilities. The Linux host and guest build passed during the November 2026 audit; hardware execution and the UEFI path were not revalidated.