Engineering principles
The binding principles from the implementation plan, with the honest status of each — principles are
practiced, not just stated.
Safety by isolation
V0.1 executes only in QEMU with project-generated disposable images — never pass-through host disks. The blast radius of a kernel bug is a throwaway VM.
IN DEVELOPMENT Intelligence is not authority
AI agents must not execute privileged kernel operations directly. Agent intent flows through policy, permission evaluation, user approval and deterministic services. On the V0.11 development branch, shipped in v1.0.0, the first agent does exactly that: it files proposals the kernel checks, only the console approves, and the kernel executes, verifies and rolls back.
IN DEVELOPMENT Small trusted core
Kernel responsibilities are explicit; clear subsystem boundaries and narrow interfaces are preferred over hidden global state. Pure logic lives outside the kernel where it can be host-tested.
IN DEVELOPMENT Deterministic observability
Every boot stage is observable through serial output even if the framebuffer or keyboard path fails. One stage, one machine-parseable marker.
IN DEVELOPMENT Failure is data
Panic output, QEMU exit codes, timeouts, exception vectors, failing assertions and serial tails are captured as structured evidence wherever possible.
IN DEVELOPMENT Reproducibility
A fresh authorized developer or agent should be able to reconstruct the environment from versioned documentation and scripts alone — toolchain pinned, commands documented, builds repeatable.
IN DEVELOPMENT