AI agents are clicking buttons with your credentials. WebMCP lets pages declare tools for agents to call, but it trusts the page to be honest. That's the same assumption that gave us phishing.
I wrote about why agentic AI needs a real consent layer, not just better sandboxes.
Most teams treat skills, MDC rules and system prompts as write-once artifacts, refined by vibes. The post looks at two practical approaches to actually measuring whether they work: deterministic rubric testing and paired comparisons borrowed from RLHF.
It also covers token cost as a forcing function for justifying what context stays.
Curious whether others have built eval pipelines for their prompt context.
Firefox can autofill password fields if you choose to remember them yeah. There is a platform bug filed for making these container specific or choosing not to auto populate when they were saved in a container.
The test pilot experiment is really to help us to figure out how people are using containers.
Yeah Test Pilot just is there to help us gather more data on how people use Containers so we can shape it's future. Lots of people are asking for features and we would like to concentrate on the ones that users would like.
Containers are indeed internal to the browser using origin attributes which are a platform feature. We have opened this up to extension authors also in WebExtensions so they can play with ideas also.
I wrote about why agentic AI needs a real consent layer, not just better sandboxes.