[simonw]

That's why I'm careful not to include allow-same-origin in the sandbox attribute - without that the iframe content is treated as a separate origin from the parent.

And I serve the content in srcdoc= to ensure there's no URL a user can visit which would directly execute the content outside of that iframe sandbox.

[cxr]

> That's why I'm careful not to include allow-same-origin in the sandbox attribute

It doesn't matter. I just said there is no combination of CSP or the iframe sandbox attribute that can be relied upon here.

[simonw]

If that's true then my project is fatally flawed and I need to stop distributing it.

I'm not convinced it's true - I've been thinking about this for months, and building experimental prototypes to help me get to the combination that I think makes sense.

Can you describe an exploit that the combination I'm using of iframe sandbox= srcdoc= with an injected meta CSP tag doesn't handle?

Would moving the untrusted content to be served from a separate domain entirely close the hole?

(In case it's not clear the iframe sandbox= is the bit that's doing most of the work here - the CSP stuff is there mainly to protect against malicious apps that deliberately exfiltrate stolen private data.)

[cxr]

> Would moving the untrusted content to be served from a separate domain entirely close the hole?

Yes.