I think by that logic dangerously-skip-permissions and openclaw should've never been a thing. I agree that people use them too liberally, but I think at some point you have to find a balance between systemic safety risks and individual freedom.

One can reduce every tool to a toy and justify it with some hand-wavy security slop, but removing capabilities destroys use cases.

The ability to control your tools is good. You should be able to run anything on your devices. Therefore, those who propose the toyification of tools should carry the burden of justifying the change.

The same infantilization of users currently happens with Signal, where high-level decision makers are asked by strangers to share their deepest secrets. Since these strangers introduce themselves very nicely, users start blurting out their secrets. ... now everyone is pretending this is a Signal problem. It is not. The world is not a kindergarten and people have agency.

A good compromise is to set a safe mode as the default and include an option that lets users confirm they know what they are doing. Obsidian already does this. Given that, I do not understand why anyone would demand to make the entire tool worse.

I wonder: What level of user effort would make you comfortable with users exiting safe modes? Would you want users to be able to run software with full permissions at all?