I do not say this to detract from the value of the project or its very interesting nature, by the way. Just an orthogonal observation.
EDIT: Found the disclosure in the repo: >I've used Claude CLI to help with some parts of this project - mostly making the web UI pretty, as I'm NOT a frontend developer. I also used it to generate the docs, but I review them manually - no automatically-generated AI code goes into the project without review from myself.
I guess that's fair.
Disclosure is something the author volunteered, its his freedom to do with HIS creation as he wishes.
Get off your soapbox.
When using such a server, its of critical importance its secure. If someone can enter it, they can change your images, knock over a machine and get it to boot a rogue image etc.
Id be interested what thread models are taken into account. If there is any fuzzing.
Perhaps a clear list of all the third party packages it pulls in and assessment of those packages.
It sounds like a lot but actually AI can help set up a lot of tooling around this stuff to make it more managable to do a lot of thorough testing / vetting of things.
I do think its also interesting project, and ofc it might be somehting that matures over time in this regard. (i am super biassed about security also as its my domain and i've litterally seen colleagues root servers which hosted images for entire infras of companies. thats a scary vector. if you can tamper with 1 PXE boot you can overwrite firmware.
(this is not saying anything about secure boot ofc, my experiences with PXE predate that being actively deployed)
What you are describing can happen with any tool, iVentoy literally injects code into your images without you knowing as it's closed source. NetbootXYZ have crazy CI Pipelines and ansible to make it work.
I tried to make it as single binary, open and single shot as possible.
Explanation still required, therefore statement must concisely outline product. Feature retains attention.
Simple outtro. Attention retained.
It is not a bug, it is a feature.