There's another element to the solution here: runtime behavioral analysis. No matter how completely the maintainer's credentials are compromised, no matter how well the malware is concealed, it still has to act like malware (in the case of LiteLLM, credential harvesting, in this case a remote-access Trojan). It's possible to detect the behavior, rather than relying on supply-chain integrity.
We built a free tool that runs local behavior analysis on your machine, it's caught every supply-chain attack in the last couple weeks: https://www.producthunt.com/products/axios-litellm-detector