I'd love to see some user controls on extensions. A great many ad blockers have a feature built in that allows you to disable them on particular websites to work around cases where the blocker has broken the site inadvertently. I want to see this in reverse: I'd like to be able to very easily say to my browser, "OK, this youtube extension can run on youtube.com, and also on *.tumbler.com to make some change to embedded youtube players on people's blogs, but nowhere else." This isn't even particularly advanced to implement: the extension can simply be disabled by default, and either prompt to activate through some browser-controlled trigger, or wait until the user turns the extension on. I think that would stand a fair, usable middle ground between the more common all-or-nothing approach to that particular permission.
The problems with Greasemonkey come down to the same answer you'll find any time you're discussing extensions: the Firefox team is seriously underinvested in the ecosystem. Greasemonkey today is about as rough to use as it was years ago and almost has the smell of an unmaintained/abandoned project.
The point is: this isn't exactly an unsolved problem so much as it is nobody taking up the effort or interest to apply a solution.
zeta0134's approach would be to put the user in charge of building this white- or blacklist. Following the way plugins are currently implemented, this would allow you to say "I always trust this extension", "I don't trust this by default, ask me every time" and "I trust this extension for example.com".
This would definitely need some security UX testing, but it seems like a reasonable approach to me.
Yes.
> It would still be a predefined set of URLs (or URL patterns) with a binary option for the user (install and accept everything, or don't).
No.
> zeta0134's approach would be to put the user in charge of building this white- or blacklist.
As I said before, this is what Greasemonkey gives you.
That model is probably not that good a fit for browser extensions. There might be a couple of permissions where this would work, but the biggest issue is definitely the "Read and change all your data on the websites you visit." permission, and that's not really something you could work around with using on-demand prompts. The idea of combining code reviews with an explanation of the required permissions provided by the developer and vetted by the reviewers sounds interesting, but it's not clear to me whether a model like that would work in practice.