Navid Pour

Write a custom scorer for the eval.

1

> Thought: The eval should be more detailed for extracts. There is a big different between completely missing 10 out of 20 commits on an eval, vs. one commit starting...

``` [stagehand:action] continuing to next sub action [stagehand:action] taking action: sign up with email "{random 12 digit number}@gmail.com", password "TEstTEst.1234". Use whatever else you want for all other fields. You...

Make extract scroll down the page more often

1

Sometimes, extract finishes prematurely before it has fetched all of the content on the page

support iframes + better debugging dev ex in dom processing

4

# why We want to be able to extract elements from iframes and use them for extraction / taking action. This PR unlocks iframe support overall, but iframe access is...

fix completion verification + perform action on first xpath that can be found on the page

1

# why - Currently we try to find the element from the first xpath fetched. Sometimes we can't use that xpath for the element. Now we check against all xpaths....

Support shadow doms

1

Repro: 1. Go to "https://amplitude.com" 2. Try to talk with their support chat bot on the bottom right of the page. It's in a shadow dom. Right now, we just...

add o1 to model

1

# why - # what changed add o1 to list of supported models # test plan tested using example