> It's the easy option, with a better safety score than human review.
Considering I went through a phase before I started using Claude Code where I'd eventually reach a point where I'd mindlessly approve ridiculously long command chains, I believe this.
As the article writes: "many users are clicking through reflexively rather than reviewing each command".
It's far better (for my working style, anyway) to have some external isolation/sandboxing, but leave the actual command approval to LLM.
(This isn't vibes: they're red-teaming Auto mode heavily)