fix(pen-core): reject invalid level in buildHeading with a clear error

ab-v4 partial sweep (2026-05-01) caught gpt-5.4 emitting
`add_heading_v0({"content":"Pending invitations","level":"caption"})`
as the 12th tag of a 13-tag composite multi-tool response. Even
though the MCP tool def has `enum: ['display','h1','h2','h3']`, the
ab-corpus harness and the in-process production dispatcher both call
buildHeading() with raw JSON args (no jsonschema gate), so the model's
invented "caption" reached the preset lookup. LATIN_PRESETS["caption"]
is undefined, and the next line `fontSize: preset.fontSize` crashed
the WHOLE batch with `undefined is not an object (evaluating
'preset.fontSize')` — the 11 valid tags ahead of it never landed.

Adds an entry-point validation in buildHeading: if `level` is set and
not in the {display, h1, h2, h3} set, throw with a clear message.
The dispatch loops in apps/web/element-tools-dispatcher and
scripts/ab-corpus/apply both catch per-shape and keep running the
remaining tags, so a single bad level on tag 12 no longer kills tags
1-11 + 13.

3 new edge-case tests in element-builders-edge-cases.test.ts
cover the throw + the four valid levels + the omitted-default case.
3772 vitest pass (+3), format clean, tsc silent.

Other element builders likely have the same pattern (preset lookup
on a string enum without runtime validation) — separate sweep, not
shotgunning here.

packages/pen-core/src/__tests__/element-builders-edge-cases.test.ts

@@ -147,3 +147,28 @@ describe('calendar_grid boundary cases', () => {
     expect(hasTodayOrSelected).toBe(false);
   });
 });
+
+describe('buildHeading — invalid level rejection', () => {
+  // ab-v4 (2026-05-01) caught gpt-5.4 inventing `level: "caption"` on a
+  // composite multi-tool sub-task. The preset lookup silently returned
+  // undefined and the next line crashed with the cryptic message
+  // `undefined is not an object (evaluating 'preset.fontSize')`,
+  // failing the WHOLE composite apply mid-batch. Builder now rejects
+  // unknown levels at the entry boundary so the surrounding dispatch
+  // loop captures a clear per-shape error and keeps applying the rest.
+  it('throws with a helpful message when level is not a known enum value', () => {
+    expect(() =>
+      buildHeading({ content: 'Pending invitations', level: 'caption' as never }),
+    ).toThrow(/invalid level "caption"; expected one of: display, h1, h2, h3/);
+  });
+
+  it('does not throw for any of the four valid levels', () => {
+    for (const level of ['display', 'h1', 'h2', 'h3'] as const) {
+      expect(() => buildHeading({ content: 'ok', level })).not.toThrow();
+    }
+  });
+
+  it('does not throw when level is omitted (defaults to h2)', () => {
+    expect(() => buildHeading({ content: 'no level' })).not.toThrow();
+  });
+});

packages/pen-core/src/element-builders/heading.ts

@@ -3,8 +3,13 @@ import type { ElementTree } from './helpers.js';
 
 export type HeadingLevel = 'display' | 'h1' | 'h2' | 'h3';
 
+const VALID_HEADING_LEVELS = new Set<string>(['display', 'h1', 'h2', 'h3']);
+
 export interface HeadingParams {
   content: string;
+  /** TS compile-time enum. Runtime callers (MCP / harness raw JSON)
+   *  may pass any string; an unknown value throws with a helpful
+   *  message rather than crashing inside the preset lookup. */
   level?: HeadingLevel;
 }
 
@@ -40,7 +45,20 @@ const CJK_BASE: Record<HeadingLevel, Omit<HeadingPreset, 'fontFamily'>> = {
  * no "应拆尽拆" violation (enum controls typography, not structure).
  */
 export function buildHeading(params: HeadingParams): ElementTree {
-  const level = params.level ?? 'h2';
+  const requestedLevel = (params.level ?? 'h2') as string;
+  // ab-v4 caught gpt-5.4 inventing `level: "caption"` on a composite
+  // multi-tool sub-task — the preset lookup returned undefined and the
+  // builder crashed mid-batch with a cryptic
+  // `undefined is not an object (evaluating 'preset.fontSize')`.
+  // Reject the value at the entry boundary so the caller (apply.ts /
+  // dispatch loop) records a clear "invalid level" error per shape
+  // and the surrounding history batch still commits the rest.
+  if (!VALID_HEADING_LEVELS.has(requestedLevel)) {
+    throw new Error(
+      `add_heading_v0: invalid level "${requestedLevel}"; expected one of: display, h1, h2, h3`,
+    );
+  }
+  const level = requestedLevel as HeadingLevel;
   const script = detectCjkScript(params.content);
   const cjkFont = cjkFontFamily(script);
   const preset: HeadingPreset = cjkFont