The general advice when it comes to cryptography is dont roll your own crypto. The Chainlink VRF best practices page says that in order to get multiple values from a single random value, one should hash a counter along with the random value. This project does not follow that guidance, and instead recursively hashes previous hashes. Logically, if you have a source of entropy, then truncate, and hash again, over and over, youre going to lose entropy if theres a weakness in the hashing function, and every hashing function will eventually be found to have a weakness. Unless there is a very good reason not to, the code should follow the best practice Chainlink outlines. Furthermore, the current scheme wastes gas updating the random seed in every function call, as is outlined in my separate gas report.
There is 1 instance of this issue:
https://github.com/code-423n4/2022-09-artgobblers/blob/d2087c5a8a6a4f1b9784520e7fe75afa3a9cbdbe/src/ArtGobblers.sol#L664-L679
