fix(GATE-ISSUE-INDEX-TABLE-SHAPE): record checkers report every failure in one run, not the first

[phantomic12]

Closes the local record added in this branch, .agents/issues/GATE-ISSUE-INDEX-TABLE-SHAPE/ISSUE-LOCAL-01M3NC14GE995V9E6F7GTYSQJ3.md.

Branch name note: the canonical branch row/GATE-ISSUE-INDEX-TABLE-SHAPE on this fork is the maintainer's own long-open work on this row (42cf62bba, "a sixth grep-alternation row, from a sixth author", open since the #995 era). This PR therefore pushes as row/GATE-ISSUE-INDEX-TABLE-SHAPE-2, touches none of that branch's files, and should merge independently of it.

The masking pattern, measured

The record suite aggregates (check-agent-record.py has one errors list and reports every finding in one run — the premise this row's frozen archive already recorded), but the two layers below and beside that aggregation each re-introduce first-failure-only semantics, and that is where #3342's three defects hid:

• scripts/agent-issue-index.py load_local_files called validate_issue_record bare, so it raised at the first file whose record failed. The first file alphabetically is an _intake record nobody is editing, so the error a maintainer saw rarely belonged to the row they touched — the exact reason the orphan-row finding needed a full enumeration of every record to see past the first _intake.
• scripts/check-agent-record.py parse_claim_rows dropped any row whose shape was malformed (cells; header or must have exactly one canonical state → bare continue). The dropped row then (a) silently moved the matrix ratchet counts and (b) never reached the downstream contract checks, so a one-cell-short row reported its pipe count and nothing else. This is precisely the MODEL-DSV41-EXL3 / MODEL-DSV41-GGUF-Q1_0 shape from fix(GATE-ISSUE-INDEX-TABLE-SHAPE): declare the three rows their own specs name, so the issue index can validate a record again #3342: "The checker reports one failure at a time, so this was masking the next two."

The two fixes

load_local_files now reports every invalid file in one run. Parse and validation failures are collected per file and raised as one IssueRecordError naming every path with its reason; unreadable files join the same list. The record collection is returned only when every file is valid, so callers guarded by the exception see no new empty-collection state.

parse_claim_rows now reports a malformed row and still parses it. The ratchet counts what is on disk, duplicate detection sees the duplicate, and the per-row contract checks add their own findings to the same run. A malformed row parses with an empty state so no state-conditional contract fires on it — the reported defects stay the shape ones.

Consumers audited: audit-live-rows and check-gate-commands collect parse errors and raise before classifying rows, so their censuses only become more complete; claim-view already rejects on any collected error.

Tests

LoadLocalFilesReportsEveryInvalidFile in tests/scripts/test_agent_issue_index.py holds the two-files-one-run contract (both file paths and both reasons appear in one exception) and the clean-load path. Two cases in tests/scripts/test_agent_record.py hold that a malformed row no longer corrupts the ratchet count, and that a malformed duplicate is reported as malformed AND duplicate AND ratchet in one run.

Gates

check-agent-record            rc=0  (ENGINE=179 MODEL=384 QUANT=87 KERNEL=60 BACKEND=90)
test_agent_issue_index.py     13/13
test_audit_live_rows.py       58/58
claim-view --check            rc=0
check-commit-style --range    rc=0
check-commit-trailers --range rc=0

Baseline honesty: test_agent_record.py on a clean checkout of d15b1cc09 fails identically to this branch (26 failures, 32 errors, same case set — the deleted-helper class #3344 repairs), and test_check_gate_commands.py fails identically (13 cases). This change introduces no regression there.

The known defect this does not fix, found while probing the base tree: agent-issue-index.py --check is rc=2 on a clean checkout of d15b1cc09 — e3539d994 rewrote the ISSUE-GH-1033 record's frozen-evidence quote (../specs/ → ../../specs/) so it no longer byte-matches .agents/completed/issue-index.md:350. That is a separate one-line repair with a real question behind it (the quote cannot both byte-match the archive and link-resolve from the record's own directory) and will get its own row and PR.

FOLLOWING_AGENTS_PROTOCOL

Following-Agents-Protocol: true
AI-Assisted: true
Assisted-by: AGENT:codebuff/buffy [freebuff]

[localai-org-maint-bot]

The new recovery path can crash on the malformed input it is intended to diagnose. In scripts/check-agent-record.py:604-605, a short row now continues after the column-count error and reads cells[state_index] without checking that the cell exists. For example, under a header containing ID and State, a row containing only | MODEL-EXAMPLE | reaches that access with one cell and a state index greater than zero. Previously the shape-error branch continued to the next line; now it raises IndexError and stops the entire gate. ClaimRow.field already uses the bounds check needed here.

Please guard the state-cell access and add a regression with a row truncated before the State column, followed by another invalid row. Both diagnostics should be returned without an exception. The added tests remove only the last column and therefore do not exercise this case.

This is source review at 8a1c212, not an executed Python test. Python is absent in this environment, and downloading a standalone interpreter was denied by the release-asset host, so I cannot validate a repair here.

[mudler-agent]

Closing as superseded by #3353. The tip commit merged there has the same stable patch ID as this PR and includes the required frozen-evidence dependency stack.