[jakear]

Can they stick CSRF tokens in the right spot under this model? The typical CSRF mitigations require other origins to not be able to access the HTML of the page (as they just inject a hidden form field or similar). If the cross-origin has full access to the page’s resources they ought to be able to emulate the environment of the page as viewed in-origin quite accurately.

Worth noting this model would introduces no new holes - everything I ask for is already possible when running a native application.

[rtpg]

I get what you're saying w/r/t CSRF. While every app could be different, in practice most websites do real bog-standard CSRF tokens, and I could see a user agent be able to get things working with like 95% of websites. Though I could think of many schemes to obfuscate things dynamically if you are motivated enough! But I like the idea of a user agent that is built around making it easier for you to just get "your" data in these ways.

> introduces no new holes - everything I ask for is already possible when running a native application.

A native application involves downloading a binary and installing it on your machine. Those involve a higher degree of trust than, say, clicking on a random URL. "I will read this person's blog" vs "I will download a binary to read this preson's blog" are acts with different trust requirements. At least for most people.

I suppose in a somewhat ironic way the iOS sandbox makes me feel more comfortable downloading random native apps but it probably really shouldn't! The OS is good about isolating cookie access for exactly the sort of things you're talking about (the prompt is like "this app wants to access your data for website.com)), but I should definitely be careful