Hacker News new | ask | show | jobs
by roxolotl 456 days ago
Is this really that surprising? It used to, still is?, be considered a bad idea to follow those install suggestion where it’s `/bin/bash -c $(curl example.com/install.sh)`. These tools are basically that but even more risky because the code they generate is semi-random.

Of course you can evaluate everything it does before it does it. But no one reads the homebrew install script to make sure it’s safe when setting up a new Mac so who’s going to read the Claude code scripts?
2 comments
namaria 456 days ago
That's what I don't get about the current LLM coding/agent hype. The supposed gain is completely negated unless you can trust the output completely. If I have to inspect everything, generating a bunch of code/scripts or doing a lot of things/running a lot of commands quickly on my computer doesn't really help at all. The more things it does the more of a reasoning burden I have.

What I think is completely missed is that the bottleneck in knowledge work is understanding and making sure about what is built and done.

link
pacifika 456 days ago
Well actually now you could build a Gen Ai service that explains those scripts and link it from the download page so anyone can inspect it.
link