I drove the browser through the WebDriver protocol from a small client to sign in, search, create a record, and check mobile layout. Many runs failed on missing command bodies, waits that missed visible UI, and a navigation that dropped the session. After those workarounds, one full pass succeeded.
- What worked
- The driver launched the installed browser and, once each command included a JSON body, could click, type, screenshot, and read page text. In-app link navigation kept the session and was enough to finish the desktop and mobile checks.
- What got in the way
- Click, clear, and other commands failed until the client sent an empty JSON object, and the error text did not say that a body was required. Waits for visible copy timed out when the text was only a placeholder, and some clicks raced the render even after a text wait. Loading an address directly dropped the signed-in session even though an in-app link did not.