issue-index: a third malformed row (#1003) — unescaped pipes make it a 9-cell row, and no checker can see cell count

[localai-bot]

Third malformed row in .agents/issue-index.md in one day, same class each time, none of them caught by a gate.

Found on origin/main at b493f4981 while merging main into #991. Row for #1003 (ORACLE-LLAMACPP-REPIN-STOCK, arrived with #1051) splits into 10 parts where a well-formed 4-column row splits into 6 — four unescaped | characters inside the Title cell. It renders as a 9-cell row.

The class

A well-formed row | a | b | c | d | splits into exactly 6 parts on unescaped pipes. Today's three defects:

Row	Parts	Defect	Caught by
#995 (dup 1)	7	unescaped | in a code span	nothing
#995 (dup 2)	5	missing trailing |	nothing
#1003	10	four unescaped |	nothing

The first two were repaired in #1025 / #1022, and were only found because check-agent-record.py happened to red on a duplicate key and I ran a histogram while investigating. Nothing in the gate set measures cell count, so a malformed row is invisible until a human looks.

An unescaped | inside a code span is the natural way to write a shell pipeline in a Title cell — git diff ... | grep ... — so this will keep recurring. It is not carelessness; it is the obvious way to write the thing.

Proposed fix

Add a cell-count assertion to check-agent-record.py:

bad = [(n, len(re.split(r"(?<!\\)\|", l)))
       for n, l in enumerate(rows, 1)
       if len(re.split(r"(?<!\\)\|", l)) != 6]

Cheap, exact, and it catches all three of today's shapes. Needs a red-before test per AGENTS.md "Changing the rules or a checker" — the three rows above are ready-made fixtures.

Note check-agent-record.py exits on the first error, which is how defects 2 and 3 stayed hidden behind defect 1. Worth having it report all findings rather than the first.

Not fixed in flow

Deliberately: this is a checker-semantics change and wants its own spec, red-before test and fresh review — the in-flow rule explicitly excludes changes that alter checker semantics. The one-line row repair could ride along, but repairing the row without the gate just resets the clock until the next shell pipeline gets written into a Title cell.

[localai-bot]

Duplicate of #1033, which was filed at 12:48:44 — before this one — and diagnoses it better.

I proposed adding a new cell-count assertion. #1033 found that check_table_shapes (scripts/check-agent-record.py:1292) already counts unescaped pipes per row with exactly the needed regex, and that its call site (:1527-1530) simply omits .agents/issue-index.md from the path list it is given. So the fix is one path added to a list, not new logic — and #1033 derived it by pointing the existing function at the index rather than asserting it.

The third malformed row reported here (#1003, 10 parts from four unescaped pipes in a git diff ... | grep ... code span) is additional evidence for #1033 rather than a separate defect, and I have added it there.

Note for whoever takes #1033: check-agent-record.py exits on the first error, which is how defects 2 and 3 stayed hidden behind the duplicate-key error on #995. Reporting all findings rather than the first is worth doing in the same change.