What inspired me to build this? Literally one tweet from @karpathy yesterday. He posted about the fresh LiteLLM supply-chain attack (a hidden .pth file that exfiltrated SSH keys, cloud credentials, K8s tokens, git secrets… everything) and said supply-chain attacks are "the scariest thing imaginable in modern software." I was up at 1am, saw the thread, and thought: "This is going to keep happening and the AI/agent crowd (Cursor, Claude Code, MCP plugins, OpenClaw especially etc.) is the most exposed because we install random transitive deps at 1 a.m." So I spent next few hours building & stress testing pipguard. The problem I wanted to solve is simple but brutal: pip install should not be Russian roulette with your AWS keys. Most existing tools scan after install or only catch known malware. pipguard does it before anything touches your machine i.e. download → local scan (GuardDog + custom rules for .pth, base64 exfil, credential dumping, etc.) → safe install. Plus pipguard doctor, env-audit, and a sandboxed run command. Zero dependencies. Zero telemetry. Runs locally. Built in ~3 hours because I was genuinely scared for my own setup. Would love your feedback, especially if you're in the AI/devtools space. Try it on the next pip install and tell me what's missing (althought I already have a list of 8-10 important things to add to it 🙈). Star it if it saves you from the next attack https://github.com/atomsai/pipguard

Pipguard

Pre-install malware guard for Python packages

Pre-install malware guard for Python packages