I'm not a spice merchant, and most exploits rarely involve a single step. This screenshot is just a system design red flag.
You're free to examine the repo yourself and find your own spice, my 5 min look tells me that piefed needs to expend a significant amount of effort on infosec to maintain user trust in the longer term.
As others have pointed out, it does still require (with some caveats about the infra setup) the user to be an admin. But if someone manages to get in to the interface, or another person is granted admin access who shouldn't have been, it makes it more risky than it needs to be. It also for me is a design choice that indicates other parts of the system should be carefully examined for how they're handling and sanitizing input.
Any webserver you browse is possibly capable of ACE depending on the implementation. When it starts to hold user data is when that starts to be a big concern. The more points of entry, the more that needs to be secured.
I don't have any experience with piefed admin, or any opinion on piefed itself, just too many years of web admin experience. And as soon as I see intentionally made doors that allow code input, I start to worry about how much experience the devs who made it have with web admin.
Well, just copy and pasted rather than written. I would have hoped that infra read-level permission, infra write-level permission and admin interface permissions were all separate to begin with, even if the person who spun up the instance obviously has all three.
You do need a level of trust in an admin, of course, but wide open text boxes for putting in code are a questionable system design choice, in my opinion. It adds an extra point of possible entry that then relies on the security of the overall admin interface instead of limiting it to what should require highest level infra admin permissions to access. And if it is something that would be limited to someone who has those, then what is the actual utility of having a textarea for it in the first place?
I get that many people are concerned about is scoring systems, but it seems a lot more worrying to me that it allows arbitrary code execution.
Out of curiosity, what sort of customizations are you doing with it? I'm just a bit surprised that docker rebuild or a non-trivial fork would be needed, so I'm assuming they're pretty big changes.