How does the browser extract locators ?

[typohead]

I used the browser. It is very good at analysing the page structure and finding out all UI itmes on the page. Kudos to the team.
I specifically want to explore this part of the browser agent layer.I want to know how exactly is the browser doing that.
Where can I start ?

Thanks in advance.

[Nikhil (shadowfax92)]

Thanks — glad the page understanding feels solid, that part's had a lot of work.

The short version of how it works: the agent talks to the page over the Chrome DevTools Protocol (CDP) and builds a combined "visual + semantic DOM" — it pulls the DOM and the accessibility tree, figures out which elements are actually interactive, and labels them so the model can refer to them. It's not classic Selenium-style XPath/CSS locators; it's more an indexed snapshot of interactive nodes derived from the accessibility + DOM layers.

Where to start in the code: it's all open source under packages/browseros-agent. The CDP plumbing lives in packages/browseros-agent/packages/cdp-protocol (look at the dom and accessibility domains), and the agent layer that turns that into the labeled element set is in the same browseros-agent tree. Clone it and start there. Closing this, but if you get into it and want a pointer to a specific file, reopen and I'll point you at it.