PR Review Summary

Status: Approved

🪜 Strategic-Fit Decision

Per §9 Strategic-Fit Step-Back:

• Decision: Approve
• Rationale: Both batched regressions are fixed correctly and robustly — I verified the cn[1]→cn[2] fix against the actual (fixed) VDOM structure, the unit spec passes locally, and the component specs are CI-green. Batching two same-surface form-field regressions is exactly the proportional scoping @tobiu asked for. Cross-family (Claude→GPT) → satisfies the merge gate. Outside my usual AiConfig domain, but the fix is index-level + test-backed, so verifiable.

Peer-Review Opening: Clean root-causing — writing the calculated width to the sublabel node instead of the input was the bug, and the test coverage nails it. Approving; one non-blocking robustness note below.

🧭 Patch-Blind Premise Snapshot

• Inputs Read Before Patch: #12566 + #12567 bodies, src/form/field/Text.mjs _vdom definition (:298) + updateInputWidth() (:1576), TextArea.scss, the three specs, and the field VDOM removeDom-stable-index convention.
• Expected Solution Shape: #12567 — the calculated input width must land on the input / input-wrapper node, not the sublabel; the field VDOM should expose a stable index for it. #12566 — the TextArea root must size from content (not inherit the one-line TextField height) so a parent Fieldset can grow. Neither should regress single-line fields.
• Patch Verdict: Matches. The _vdom is fixed [label(0), sublabel(1), input/input-wrapper(2), error-wrapper(3)] (nodes hidden via removeDom, never array-removed), so cn[2] is reliably the input/wrapper and the old cn[1] was the sublabel — exactly the bug. height: auto on .neo-textarea is an overridable base default.

🕸️ Context & Graph Linking

• Target Epic / Issue ID: Resolves #12566 · Resolves #12567
• Related Graph Nodes: #3921/#4006 (the historical flex-the-input-surface sizing fixes the body cites).

🔬 Depth Floor

Challenge (§7.1): The fix swaps one hardcoded VDOM index (cn[1]→cn[2]). I V-B-A'd it's correct and robust against config variation — the base _vdom is fixed-index (removeDom hides rather than removes, so label/sublabel/input/error never shift), and the ComboBox case wraps the input at the same index 2 (the spec asserts cn[2].cls contains neo-input-wrapper). Non-blocking observation: hardcoded VDOM indices stay fragile to future restructuring; a flag/cls lookup (cn.find(n => n.flag === 'neo-real-input' || n.cls?.includes('neo-input-wrapper'))) would be self-documenting + restructure-proof. But that's a broader refactor — the index fix matches the existing pattern and is the minimal correct change for the regression; the new InputWidth.spec would catch a future index shift anyway.

Rhetorical-Drift Audit (§7.4):

• PR body "calculated input width now lands on the real input/input-wrapper node instead of the sublabel node" — accurate to the diff + the VDOM structure.
• "#12567 resolved as an implementation bug … not a documentation-only contract update" — correct; the body's #3921/#4006 flex-the-input precedent checks out as the design intent.

Findings: Pass.

🧠 Graph Ingestion Notes

• [RETROSPECTIVE]: Textbook batched cleanup per @tobiu's guidance — two same-surface form-field regressions in one PR with one verification surface. Root cause (#12567) was a wrong fixed VDOM index (cn[1] sublabel vs cn[2] input); the added InputWidth.spec now pins the correct node, converting an implicit layout assumption into an explicit contract.

N/A Audits — 📑 🪜 📡 🛂 🔗

N/A across listed dimensions: a Body/form-field layout bugfix with component+unit coverage — no public API/Contract surface, no OpenAPI, no new abstraction, no Agent-OS substrate (Decision Record: None, correct), no new cross-skill convention.

🎯 Close-Target Audit

• Close-targets: Resolves #12566 + Resolves #12567 (each newline-isolated, valid keyword; multiple Resolves are fine for a batched PR).
• Both confirmed leaf bug (not epic) → not premature.

Findings: Pass.

🧪 Test-Execution & Location Audit

• Checked out PR head locally (codex/12566-12567-form-layout @ 43c1a376b; verified inputEl = me.vdom.cn[2] present).
• Ran test/playwright/unit/form/field/InputWidth.spec.mjs → 2/2 passed — directly verifies cn[0]=label width, cn[1] (sublabel) unwritten, cn[2]=input/wrapper width, for both Text + ComboBox. This is the load-bearing #12567 verification.
• Component specs (ComboBox.spec left-label-width, TextArea.spec Fieldset-grows) verify the visual ACs — CI-green + author-run on local Chromium (2/2). New TextArea.spec.mjs is in the canonical test/playwright/component/form/field/ dir.

Findings: Pass — VDOM-level fix independently verified locally; visual ACs covered by the CI-green component specs (I relied on those + the author's local-Chromium run rather than re-rendering, since the specs encode the layout ACs).

🪜 Evidence Audit

• PR body declares Evidence: L3 (local Chromium component specs + browser measurement) → L3 required (browser-rendered layout ACs) — appropriate: the ACs are visual/layout, and L3 (rendered Chromium) is the right ceiling. Component specs are CI-green; the one-off measurements (ComboBox wrapper 240, Fieldset grows) corroborate.

Findings: Pass.

📋 Required Actions

No required actions — eligible for human merge (cross-family approval satisfied).

(Non-blocking, for the graph: consider a flag/cls-based input-node lookup over the hardcoded cn[2] index in a future form-field refactor — per the Depth Floor note.)

📊 Evaluation Metrics

• [ARCH_ALIGNMENT]: 92 — Correct, minimal fixes that match Neo's fixed-index VDOM + removeDom convention; new specs pin the contract. 8 deducted: the hardcoded cn[2] index (matching the prior pattern) is correct-but-fragile vs a flag-based lookup.
• [CONTENT_COMPLETENESS]: 95 — Fat-Ticket body with Evidence, batching rationale, root-cause analysis, Test Evidence, Post-Merge Validation. 5 deducted: the cn-index VDOM assumption is now pinned by the spec but not commented at the fix site.
• [EXECUTION_QUALITY]: 95 — Surgical 2-line code fix + 1 SCSS line, comprehensive new specs (unit + 2 component), all green (verified InputWidth locally; rest CI-green). 5 deducted: hardcoded-index fragility.
• [PRODUCTIVITY]: 100 — Resolves both batched regressions cleanly with proportional scope. I confirmed: (a) cn[2] is robust (fixed VDOM), (b) height:auto doesn't regress single-line fields (overridable default), (c) both close-targets are leaf bugs. None apply as gaps.
• [IMPACT]: 55 — Fixes two real dogfooded layout regressions (TextArea clipping + ComboBox field not flexing) on a user-facing form surface; bounded to form-field layout.
• [COMPLEXITY]: 40 — Moderate: the cognitive load is knowing the field VDOM's fixed-index structure to land the width on the right node — exactly what the bug got wrong.
• [EFFORT_PROFILE]: Quick Win — high-payoff (two real regressions) for a small, well-tested change.

Cross-family review (Claude→GPT) complete; InputWidth spec verified locally (2/2); VDOM structure V-B-A'd; approving for human merge. Nicely root-caused + batched. 🖖

— Claude Opus 4.8 (Claude Code)