Comment on Ovumcy (self-hosted period tracker), 9 months later: what I did with your feedback [AIP]
terraincognita@lemmy.world 15 hours agoGood questions, thanks for actually looking at the site.
Encryption: client-side encryption is for services run by someone else, where you can’t be sure nobody reads or leaks your data. A self-hosted app has a different threat model. The data sits on a machine you control, behind a login and 2FA, and how the server side is protected is described in the docs. So for your own server it was never the goal. The sentence you quoted is about the hosted option, which exists for people who find setting up a server too hard. There someone else holds the data, so the site says it plainly. For those who want client-side encryption anyway, I’m working on a mobile app with encrypted sync. It isn’t released yet.
The vote counter: you’re right, it can be spammed. That’s the price of collecting nothing about visitors, and I chose that on purpose. So far nobody bothers, bots included. I treat it as a rough signal and still decide myself what to build.
AI and security: no code comes with guarantees, whoever wrote it. What I can offer is that all of it is public, with tests, fuzzing and static analysis in CI, and a SECURITY.md if you find something. If anything looks wrong, tell me and I’ll fix it.
savvywolf@pawb.social 11 hours ago
I can think of other cases where even if it’s self hosted, you still want this sort of encryption. What about the use case of a mother setting it up for a group of other mums in a group chat. The other mums would still want their data to be private. The hosting provider thing is valid, and I think actually extends further than you think. I may trust my own server, but I don’t know if I trust Linode or Hetzner or whoever I buy my server from to not snoop in my files. There’s also the fact that if I fuck up the configuration on my server, at least they aren’t going to get that data in the breach. (“I” in this case being an abstract person wanting to use the server. I don’t have any use for a tool like this).
That’s fair, just be aware of it. The amount of people that make decisions based on flawed customer data is a big problem for software development.
The recent SSH bug had all of this and let a security exploit slip through. The reason was that people made a sensible sounding but ultimately flawed design decision. Those are the types of errors that AIs love to make and that inexperienced users can let slip through.
terraincognita@lemmy.world 11 hours ago
These are fair, and the misconfigured server one is the strongest. Client-side encryption would limit the damage there, I can’t argue with that.
The group case is outside what I built this for. It’s meant for a household. If you run it for other people, you become their hosting provider, and the docs say that plainly: the app keeps your records from other accounts and from the network, not from whoever administers the machine.
The VPS point is true, but it applies to everything on that box, not just this app. If you don’t trust the provider with it, the answer is hardware at home.
I won’t promise it, but optional encryption for the web version is on my list to look into now.
On design mistakes: agreed, tests and fuzzing don’t catch a wrong decision. But the SSH example cuts both ways. That code was written and reviewed by very experienced people and it still got through. Nobody is immune, so what matters is whether mistakes get found and fixed in the open. That has already happened here: tonghuaroot reported a hole in the webhook address filter, it was fixed and published as an advisory (GHSA-hg2x-v5cc-m384).