I would personally move this submodule into a separate file

I see that there are some traits with definitions for these values in aes-gcm. Is there maybe a way for these to use constants from the crate instead of hardcoding numbers here?

Should we additionaly add/ change it into a EncryptionKey::new?

I might be missing something but I don't think it matches the code

pub fn decrypt<'buf>(
  ciphertext: &'buf mut [u8],
  key: &EncryptionKey,
) -> Result<&'buf mut [u8], aes_gcm::Error> {
  if ciphertext.len() < NONCE_LEN + TAG_LEN {
    return Err(aes_gcm::Error);
  }
  let cipher = Aes256Gcm::new(key.as_ref());
  let (rest, tag) = ciphertext.split_at_mut(ciphertext.len() - TAG_LEN);
  let (nonce, buffer) = rest.split_at_mut(NONCE_LEN);
  let nonce = GenericArray::from_slice(nonce);
  let tag = GenericArray::from_slice(tag);
  cipher.decrypt_in_place_detached(nonce, b"", buffer, tag)?;
  Ok(buffer)
}

This passes the tests, but I'm not really familiar with crypto and I'm just following the types and your code comments so it might be completely wrong 😅 (also it's a bit weird to use)

good idea

I couldn't figure it out, these crypto types are weird

good idea

good catch

No, I haven't considered any "side channel attacks", whatever they are. It's a zero-sized error type, it gives no information about failure reason. It doesn't matter if we reuse it or introduce a custom one.

Yeah, it's weird to use. I'd stick with the current approach