@bendyline/squisq-formats 1.4.1 → 2.0.0

package/src/ooxml/writer.ts

@@ -20,6 +20,7 @@ import {
   REL_CORE_PROPERTIES,
 } from './namespaces.js';
 import { xmlDeclaration, escapeXml } from './xmlUtils.js';
+import { assertSafeZipPath } from '../shared/zipSafety.js';
 
 // ============================================
 // Package Builder
@@ -99,14 +100,17 @@ export function createPackage(): OoxmlPackageBuilder {
 
   return {
     addPart(path, content, contentType) {
+      assertSafeZipPath(path);
       parts.push({ path, content, contentType });
     },
 
     addBinaryPart(path, data, contentType) {
+      assertSafeZipPath(path);
       parts.push({ path, binaryContent: data, contentType });
     },
 
     addRelationship(sourcePart, rel) {
+      if (sourcePart) assertSafeZipPath(sourcePart);
       relationships.push({ sourcePart, relationship: rel });
     },
 
@@ -139,6 +143,10 @@ function assemble(
   coreProps?: CoreProperties,
 ): JSZip {
   const zip = new JSZip();
+  // Assembly must be idempotent: callers may request both Blob and
+  // ArrayBuffer representations from one builder. Never append generated
+  // relationships to the builder-owned array.
+  const assembledRelationships = relationships.slice();
 
   // Write content parts
   for (const part of parts) {
@@ -154,22 +162,32 @@ function assemble(
     const coreXml = buildCorePropertiesXml(coreProps);
     zip.file('docProps/core.xml', coreXml);
 
-    // Add relationship to core properties
-    relationships.push({
-      sourcePart: '',
-      relationship: {
-        id: `rId${relationships.length + 100}`,
-        type: REL_CORE_PROPERTIES,
-        target: 'docProps/core.xml',
-      },
-    });
+    // Add the generated relationship unless the caller already supplied it.
+    if (
+      !assembledRelationships.some(
+        ({ sourcePart, relationship }) =>
+          sourcePart === '' && relationship.type === REL_CORE_PROPERTIES,
+      )
+    ) {
+      const ids = new Set(assembledRelationships.map(({ relationship }) => relationship.id));
+      let nextId = 100;
+      while (ids.has(`rId${nextId}`)) nextId++;
+      assembledRelationships.push({
+        sourcePart: '',
+        relationship: {
+          id: `rId${nextId}`,
+          type: REL_CORE_PROPERTIES,
+          target: 'docProps/core.xml',
+        },
+      });
+    }
   }
 
   // Build and write [Content_Types].xml
   zip.file('[Content_Types].xml', buildContentTypesXml(parts, coreProps));
 
   // Build and write _rels/*.rels files
-  const relsBySource = groupRelationshipsBySource(relationships);
+  const relsBySource = groupRelationshipsBySource(assembledRelationships);
   for (const [sourcePart, rels] of relsBySource) {
     const relsPath = sourcePart === '' ? '_rels/.rels' : buildRelsPath(sourcePart);
     zip.file(relsPath, buildRelationshipsXml(rels));

package/src/pdf/export.ts

@@ -21,7 +21,7 @@
 
 import { PDFDocument, StandardFonts, rgb, PDFFont, PDFPage } from 'pdf-lib';
 
-import type { Doc } from '@bendyline/squisq/schemas';
+import type { Doc, ThemeRegistry } from '@bendyline/squisq/schemas';
 import { docToMarkdown, resolveThemeForDoc } from '@bendyline/squisq/doc';
 import type {
   MarkdownDocument,
@@ -110,6 +110,8 @@ export interface PdfExportOptions {
    * Font changes are not supported (pdf-lib uses standard 14 PDF fonts only).
    */
   themeId?: string;
+  /** Explicit caller-owned registry for non-document custom themes. */
+  themeRegistry?: ThemeRegistry;
 }
 
 /**
@@ -219,7 +221,7 @@ async function createExportContext(
   // including inline custom themes (resolved doc-scoped).
   const themeId = options.themeId ?? readFrontmatterThemeId(doc.frontmatter);
   if (themeId) {
-    const theme = resolveThemeForDoc(doc, themeId);
+    const theme = resolveThemeForDoc(doc, themeId, options.themeRegistry);
     if (theme.colors) {
       colorText = hexToRgb(theme.colors.text) ?? COLOR_TEXT;
       colorHeading = hexToRgb(theme.colors.primary) ?? COLOR_HEADING;

package/src/pdf/import.ts

@@ -107,16 +107,21 @@ export async function pdfToMarkdownDoc(
         ? new Uint8Array(data)
         : data;
 
-  const textLines = await extractTextLines(bytes);
+  const loaded = await loadPdfDocument(bytes);
+  try {
+    const textLines = await extractTextLines(loaded.pdf);
 
-  if (textLines.length === 0) {
-    return { type: 'document', children: [] };
-  }
+    if (textLines.length === 0) {
+      return { type: 'document', children: [] };
+    }
 
-  const bodySize = options.bodyFontSize ?? detectBodyFontSize(textLines);
-  const blocks = classifyLines(textLines, bodySize, options);
+    const bodySize = options.bodyFontSize ?? detectBodyFontSize(textLines);
+    const blocks = classifyLines(textLines, bodySize, options);
 
-  return { type: 'document', children: blocks };
+    return { type: 'document', children: blocks };
+  } finally {
+    await loaded.pdf.destroy?.();
+  }
 }
 
 /**
@@ -157,8 +162,14 @@ export async function pdfToContainer(
         ? new Uint8Array(data)
         : data;
 
-  const textLines = await extractTextLines(bytes);
-  const images = await extractImages(bytes);
+  const loaded = await loadPdfDocument(bytes);
+  let textLines: TextLine[];
+  let images: ExtractedImage[];
+  try {
+    [textLines, images] = await Promise.all([extractTextLines(loaded.pdf), extractImages(loaded)]);
+  } finally {
+    await loaded.pdf.destroy?.();
+  }
 
   // Under Node there is no DOM/canvas, so extractImages() returns [] via its
   // environment guard and embedded images can't be decoded. The registry's
@@ -208,32 +219,14 @@ export interface ExtractedImage {
  * Extract embedded images from a PDF using pdfjs-dist operator list API.
  * Requires browser canvas for PNG encoding.
  */
-async function extractImages(data: Uint8Array): Promise<ExtractedImage[]> {
+async function extractImages(loaded: LoadedPdf): Promise<ExtractedImage[]> {
   // Canvas is required for PNG encoding — skip in non-browser environments
   if (typeof document === 'undefined') return [];
 
-  let pdfjsLib: PdfjsLib & { OPS?: Record<string, number> };
-  try {
-    pdfjsLib = (await import('pdfjs-dist/legacy/build/pdf.mjs')) as unknown as PdfjsLib & {
-      OPS?: Record<string, number>;
-    };
-  } catch {
-    pdfjsLib = (await import('pdfjs-dist')) as unknown as PdfjsLib & {
-      OPS?: Record<string, number>;
-    };
-  }
-
-  await applyWorkerConfig(pdfjsLib);
+  const { pdfjsLib, pdf } = loaded;
 
   const OPS_paintImageXObject = pdfjsLib.OPS?.paintImageXObject ?? 85;
 
-  const loadingTask = pdfjsLib.getDocument({
-    data,
-    isEvalSupported: false,
-    useSystemFonts: true,
-  });
-
-  const pdf = await loadingTask.promise;
   const images: ExtractedImage[] = [];
   let counter = 0;
 
@@ -501,6 +494,12 @@ interface PdfjsLib {
 interface PdfjsDocument {
   numPages: number;
   getPage(pageNum: number): Promise<PdfjsPage>;
+  destroy?(): Promise<void>;
+}
+
+interface LoadedPdf {
+  pdfjsLib: PdfjsLib & { OPS?: Record<string, number> };
+  pdf: PdfjsDocument;
 }
 
 interface PdfjsPage {
@@ -528,14 +527,18 @@ async function applyWorkerConfig(pdfjsLib: PdfjsLib): Promise<void> {
   // Node.js the caller must have called configurePdfWorker() first.
 }
 
-async function extractTextLines(data: Uint8Array): Promise<TextLine[]> {
+async function loadPdfDocument(data: Uint8Array): Promise<LoadedPdf> {
   // Dynamic import — the legacy build bundles a fake-worker fallback
   // that avoids a real Web Worker in environments that don't support it.
-  let pdfjsLib: PdfjsLib;
+  let pdfjsLib: PdfjsLib & { OPS?: Record<string, number> };
   try {
-    pdfjsLib = (await import('pdfjs-dist/legacy/build/pdf.mjs')) as unknown as PdfjsLib;
+    pdfjsLib = (await import('pdfjs-dist/legacy/build/pdf.mjs')) as unknown as PdfjsLib & {
+      OPS?: Record<string, number>;
+    };
   } catch {
-    pdfjsLib = (await import('pdfjs-dist')) as unknown as PdfjsLib;
+    pdfjsLib = (await import('pdfjs-dist')) as unknown as PdfjsLib & {
+      OPS?: Record<string, number>;
+    };
   }
 
   await applyWorkerConfig(pdfjsLib);
@@ -546,7 +549,10 @@ async function extractTextLines(data: Uint8Array): Promise<TextLine[]> {
     useSystemFonts: true,
   });
 
-  const pdf = await loadingTask.promise;
+  return { pdfjsLib, pdf: await loadingTask.promise };
+}
+
+async function extractTextLines(pdf: PdfjsDocument): Promise<TextLine[]> {
   const allLines: TextLine[] = [];
 
   for (let pageNum = 1; pageNum <= pdf.numPages; pageNum++) {

package/src/pptx/export.ts

@@ -29,7 +29,7 @@
 
 import type { Doc, Transition, TransitionDirection } from '@bendyline/squisq/schemas';
 import { resolveFontFamily } from '@bendyline/squisq/schemas';
-import type { Theme } from '@bendyline/squisq/schemas';
+import type { Theme, ThemeRegistry } from '@bendyline/squisq/schemas';
 import { docToMarkdown, resolveThemeForDoc } from '@bendyline/squisq/doc';
 import type {
   MarkdownDocument,
@@ -132,6 +132,8 @@ export interface PptxExportOptions {
    * Falls back to the document's frontmatter `themeId` if not set here.
    */
   themeId?: string;
+  /** Explicit caller-owned registry for non-document custom themes. */
+  themeRegistry?: ThemeRegistry;
   /**
    * Image data keyed by the path/URL used in markdown.
    * When provided, images are embedded as picture shapes instead of
@@ -224,7 +226,7 @@ function resolveSlideStyle(
     };
   }
 
-  const theme: Theme = resolveThemeForDoc(doc, themeId);
+  const theme: Theme = resolveThemeForDoc(doc, themeId, options.themeRegistry);
   const c = theme.colors;
 
   return {

package/src/pptx/import.ts

@@ -7,9 +7,20 @@
  * (or "Slide N"), the remaining text as a bullet list, and any slide tables
  * (`<a:tbl>`) as markdown tables. Text lives in the DrawingML namespace
  * (`a:p` / `a:r` / `a:t`) inside PresentationML shapes (`p:sp`).
+ *
+ * Theme + layout inference (default ON): the deck's theme part is compiled
+ * into a Squisq custom theme and carried in the returned document's
+ * frontmatter (`squisq-custom-themes` + `squisq-theme`); slide layouts are
+ * classified against the built-in template set, distinctive ones become
+ * `squisq-custom-templates` definitions, and each slide's heading is
+ * annotated (`{[templateName …]}`) per its layout's verdict. Pass
+ * `inferTheme: false` / `inferLayouts: false` for legacy plain imports.
+ * Inference never fails an import — extraction errors degrade to plain
+ * output with a console warning.
  */
 
 import type {
+  HeadingTemplateAnnotation,
   MarkdownBlockNode,
   MarkdownDocument,
   MarkdownImage,
@@ -19,14 +30,19 @@ import type {
   MarkdownTableRow,
 } from '@bendyline/squisq/markdown';
 import { stringifyMarkdown } from '@bendyline/squisq/markdown';
+import type { CustomTemplateDefinition, Theme } from '@bendyline/squisq/schemas';
 import { getPartBinary, getPartRelationships, getPartXml, openPackage } from '../ooxml/reader.js';
+import type { OoxmlOpenOptions } from '../ooxml/reader.js';
 import type { OoxmlPackage } from '../ooxml/types.js';
 import { NS_DRAWINGML, NS_PML, NS_R } from '../ooxml/namespaces.js';
+import { attrNS, baseDirOf, resolveTarget } from '../ooxml/readUtils.js';
 import type { ContentContainer } from '@bendyline/squisq/storage';
 import { buildContainer } from '../shared/container.js';
 import { extToMime } from '../shared/images.js';
+import type { ExtractedFileTheme } from '../infer/types.js';
+import type { AnalyzedLayout, PptxLayoutInference } from './layouts.js';
 
-export interface PptxImportOptions {
+export interface PptxImportOptions extends OoxmlOpenOptions {
   /**
    * Whether to extract embedded slide images into the document as image nodes
    * (referencing `images/imageN.ext`). When false, pictures are ignored so the
@@ -34,6 +50,19 @@ export interface PptxImportOptions {
    * `pptxToContainer` forces this on. Default: false.
    */
   extractImages?: boolean;
+  /**
+   * Infer a Squisq theme from the deck's theme part (colors + fonts) and
+   * carry it in the returned document's frontmatter (`squisq-custom-themes`
+   * + `squisq-theme`). Default: true — pass false for a legacy import with
+   * no frontmatter.
+   */
+  inferTheme?: boolean;
+  /**
+   * Derive custom layout templates from the deck's slide layouts/masters
+   * (`squisq-custom-templates` frontmatter) and annotate slide headings
+   * with matching built-in or generated templates. Default: true.
+   */
+  inferLayouts?: boolean;
 }
 
 /**
@@ -46,20 +75,10 @@ interface ImportContext {
   /** Collected image files: `images/imageN.ext` → { data, mimeType } */
   extractedImages: Map<string, { data: ArrayBuffer; mimeType: string }>;
   imageCounter: number;
-}
-
-function attrNS(el: Element, ns: string, local: string, fallback: string): string | null {
-  return el.getAttributeNS(ns, local) ?? el.getAttribute(fallback);
-}
-
-function resolveTarget(baseDir: string, target: string): string {
-  if (target.startsWith('/')) return target.replace(/^\//, '');
-  const stack = baseDir ? baseDir.split('/') : [];
-  for (const seg of target.split('/')) {
-    if (seg === '..') stack.pop();
-    else if (seg !== '.') stack.push(seg);
-  }
-  return stack.join('/');
+  /** Layout analysis when `inferLayouts` is on. */
+  inference?: PptxLayoutInference;
+  /** Generated custom templates actually referenced by ≥1 slide annotation. */
+  usedCustomTemplates: Map<string, CustomTemplateDefinition>;
 }
 
 async function orderedSlidePaths(pkg: OoxmlPackage): Promise<string[]> {
@@ -114,12 +133,6 @@ function tableToMarkdown(tbl: Element): MarkdownTable {
   return { type: 'table', children: rows };
 }
 
-/** Directory portion of a part path, e.g. `ppt/slides/slide1.xml` → `ppt/slides`. */
-function baseDirOf(path: string): string {
-  const slash = path.lastIndexOf('/');
-  return slash === -1 ? '' : path.slice(0, slash);
-}
-
 /**
  * Extract every `<p:pic>` picture in a slide as an image node, reading the
  * `<a:blip r:embed>` relationship, resolving it to the media part, and copying
@@ -168,6 +181,93 @@ async function extractSlideImages(
   return images;
 }
 
+// ── Layout-verdict annotation building ───────────────────────────────
+
+/** One non-title text shape on a slide, with its placeholder identity. */
+interface SlideTextEntry {
+  rawType: string;
+  idx: number;
+  texts: string[];
+}
+
+/** Annotation params live on a single heading line — flatten whitespace. */
+function cleanParamText(text: string): string {
+  return text.replace(/\s+/g, ' ').trim();
+}
+
+/**
+ * Turn a slide's layout verdict into a heading annotation, downgrading to
+ * plain when the slide lacks the content the template needs (a bare
+ * annotation without its essential params renders an empty card in the
+ * player path, which never derives inputs for annotated blocks).
+ */
+function buildSlideAnnotation(
+  analyzed: AnalyzedLayout | undefined,
+  entries: SlideTextEntry[],
+  images: MarkdownImage[],
+  ctx: ImportContext,
+): { annotation?: HeadingTemplateAnnotation; omitted: Set<SlideTextEntry> } {
+  const omitted = new Set<SlideTextEntry>();
+  if (!analyzed) return { omitted };
+  const verdict = analyzed.verdict;
+
+  if (verdict.kind === 'custom') {
+    ctx.usedCustomTemplates.set(verdict.def.name, verdict.def);
+    return { annotation: { template: verdict.def.name }, omitted };
+  }
+  if (verdict.kind !== 'builtin') return { omitted };
+
+  switch (verdict.paramSpec) {
+    case 'titleSubtitle': {
+      const sub = entries.find((e) => e.rawType === 'subTitle');
+      const subtitle = sub ? cleanParamText(sub.texts.join(' ')) : '';
+      // The subtitle moves into the card; keep it out of the bullets.
+      if (sub && subtitle) omitted.add(sub);
+      return {
+        annotation: {
+          template: verdict.template,
+          ...(subtitle ? { params: { subtitle } } : {}),
+        },
+        omitted,
+      };
+    }
+    case 'comparisonPairs': {
+      if (!verdict.columns) return { omitted };
+      const textAt = (idx: number | undefined): string[] =>
+        idx === undefined ? [] : (entries.find((e) => e.idx === idx)?.texts ?? []);
+      const side = (idxs: number[]): string => {
+        const header = cleanParamText(textAt(idxs[0])[0] ?? '');
+        const body = cleanParamText(textAt(idxs[1])[0] ?? '');
+        return header ? (body ? `${header}|${body}` : header) : '';
+      };
+      const left = side(verdict.columns.left);
+      const right = side(verdict.columns.right);
+      // twoColumn requires both labels — a bare annotation renders nothing.
+      if (!left || !right) return { omitted };
+      return { annotation: { template: 'twoColumn', params: { left, right } }, omitted };
+    }
+    case 'featureImage': {
+      const image = images[0];
+      if (!image) return { omitted };
+      const params: Record<string, string> = { imageSrc: image.url };
+      if (image.alt && image.alt !== 'Image') params.imageAlt = cleanParamText(image.alt);
+      return { annotation: { template: verdict.template, params }, omitted };
+    }
+    case 'photoGridGate': {
+      if (images.length < 2) return { omitted };
+      return {
+        annotation: {
+          template: 'photoGrid',
+          params: { images: images.map((i) => i.url).join(',') },
+        },
+        omitted,
+      };
+    }
+    default:
+      return { annotation: { template: verdict.template }, omitted };
+  }
+}
+
 async function convertSlide(
   path: string,
   index: number,
@@ -179,7 +279,7 @@ async function convertSlide(
   const out: MarkdownBlockNode[] = [];
 
   let title = '';
-  const bullets: string[] = [];
+  const entries: SlideTextEntry[] = [];
   const shapes = doc.getElementsByTagNameNS(NS_PML, 'sp');
   for (let s = 0; s < shapes.length; s++) {
     const sp = shapes[s]!;
@@ -194,17 +294,32 @@ async function convertSlide(
     if (texts.length === 0) continue;
     if (isTitleShape(sp) && !title) {
       title = texts.join(' ');
-    } else {
-      bullets.push(...texts);
+      continue;
     }
+    const phs = sp.getElementsByTagNameNS(NS_PML, 'ph');
+    const rawType = phs.length ? (phs[0]!.getAttribute('type') ?? '') : '';
+    const idxRaw = phs.length ? phs[0]!.getAttribute('idx') : null;
+    const idx = idxRaw ? parseInt(idxRaw, 10) || 0 : 0;
+    entries.push({ rawType, idx, texts });
   }
 
+  // Images are extracted before annotation building so feature/photoGrid
+  // params can reference their container paths; they still land after the
+  // bullet list in the output, unchanged.
+  const images = ctx.extractImages ? await extractSlideImages(doc, path, ctx) : [];
+
+  const layoutPath = ctx.inference?.layoutPathBySlide.get(path);
+  const analyzed = layoutPath ? ctx.inference?.byLayoutPath.get(layoutPath) : undefined;
+  const { annotation, omitted } = buildSlideAnnotation(analyzed, entries, images, ctx);
+
   out.push({
     type: 'heading',
     depth: 2,
     children: [{ type: 'text', value: title || `Slide ${index + 1}` }],
+    ...(annotation ? { templateAnnotation: annotation } : {}),
   });
 
+  const bullets = entries.filter((e) => !omitted.has(e)).flatMap((e) => e.texts);
   if (bullets.length > 0) {
     const items: MarkdownListItem[] = bullets.map((text) => ({
       type: 'listItem',
@@ -214,11 +329,8 @@ async function convertSlide(
   }
 
   // Embedded pictures land after the bullet list and before any tables.
-  if (ctx.extractImages) {
-    const images = await extractSlideImages(doc, path, ctx);
-    for (const image of images) {
-      out.push({ type: 'paragraph', children: [image] });
-    }
+  for (const image of images) {
+    out.push({ type: 'paragraph', children: [image] });
   }
 
   const tbls = doc.getElementsByTagNameNS(NS_DRAWINGML, 'tbl');
@@ -227,6 +339,11 @@ async function convertSlide(
   return out;
 }
 
+function warnInferenceFailure(step: string, err: unknown): void {
+  const message = err instanceof Error ? err.message : String(err);
+  console.warn(`pptx import: ${step} failed; importing without it — ${message}`);
+}
+
 async function importDocument(
   pkg: OoxmlPackage,
   options: PptxImportOptions,
@@ -236,20 +353,85 @@ async function importDocument(
     extractImages: options.extractImages ?? false,
     extractedImages: new Map(),
     imageCounter: 0,
+    usedCustomTemplates: new Map(),
   };
+
+  const inferTheme = options.inferTheme !== false;
+  const inferLayouts = options.inferLayouts !== false;
+
+  // All inference modules load lazily so a plain import stays light, and
+  // every inference step degrades to a warning rather than failing the import.
+  let extraction: ExtractedFileTheme | null = null;
+  if (inferTheme || inferLayouts) {
+    try {
+      const { extractPptxTheme } = await import('../infer/extract.js');
+      extraction = await extractPptxTheme(pkg);
+    } catch (err: unknown) {
+      warnInferenceFailure('theme extraction', err);
+    }
+  }
+
+  let theme: Theme | undefined;
+  if (inferTheme && extraction) {
+    try {
+      const { compileExtractedTheme } = await import('../infer/mapTheme.js');
+      theme = compileExtractedTheme(extraction).theme;
+    } catch (err: unknown) {
+      warnInferenceFailure('theme compilation', err);
+    }
+  }
+
+  if (inferLayouts) {
+    try {
+      const { analyzePptxLayouts } = await import('./layouts.js');
+      const { colorHintsFromExtraction } = await import('../infer/mapTheme.js');
+      ctx.inference = await analyzePptxLayouts(pkg, {
+        colors: extraction ? colorHintsFromExtraction(extraction) : {},
+      });
+    } catch (err: unknown) {
+      warnInferenceFailure('layout inference', err);
+    }
+  }
+
   const paths = await orderedSlidePaths(pkg);
   const children: MarkdownBlockNode[] = [];
   for (let i = 0; i < paths.length; i++) {
     children.push(...(await convertSlide(paths[i]!, i, ctx)));
   }
-  return { doc: { type: 'document', children }, ctx };
+  const doc: MarkdownDocument = { type: 'document', children };
+
+  const frontmatter: Record<string, unknown> = {};
+  if (theme || ctx.usedCustomTemplates.size > 0) {
+    const {
+      writeCustomThemesToFrontmatter,
+      writeCustomTemplatesToFrontmatter,
+      FRONTMATTER_CUSTOM_THEMES_KEY,
+      FRONTMATTER_CUSTOM_TEMPLATES_KEY,
+    } = await import('@bendyline/squisq/doc');
+    if (theme) {
+      const payload = writeCustomThemesToFrontmatter([theme]);
+      if (payload) {
+        frontmatter[FRONTMATTER_CUSTOM_THEMES_KEY] = payload;
+        // The doc-level selector `resolveThemeForDoc` reads — activates the
+        // inferred theme without any global registration.
+        frontmatter['squisq-theme'] = theme.id;
+      }
+    }
+    if (ctx.usedCustomTemplates.size > 0) {
+      const payload = writeCustomTemplatesToFrontmatter([...ctx.usedCustomTemplates.values()]);
+      if (payload) frontmatter[FRONTMATTER_CUSTOM_TEMPLATES_KEY] = payload;
+    }
+  }
+  if (Object.keys(frontmatter).length > 0) doc.frontmatter = frontmatter;
+
+  return { doc, ctx };
 }
 
 export async function pptxToMarkdownDoc(
   data: ArrayBuffer | Blob,
   options: PptxImportOptions = {},
 ): Promise<MarkdownDocument> {
-  const pkg = await openPackage(data);
+  const pkg = await openPackage(data, options);
   const { doc } = await importDocument(pkg, options);
   return doc;
 }
@@ -266,7 +448,7 @@ export async function pptxToContainer(
   data: ArrayBuffer | Blob,
   options: PptxImportOptions = {},
 ): Promise<ContentContainer> {
-  const pkg = await openPackage(data);
+  const pkg = await openPackage(data, options);
   const { doc, ctx } = await importDocument(pkg, { ...options, extractImages: true });
 
   return buildContainer(stringifyMarkdown(doc), ctx.extractedImages);

package/src/pptx/index.ts

@@ -27,6 +27,20 @@ import { type PptxImportOptions, pptxToMarkdownDoc } from './import.js';
 export type { PptxImportOptions } from './import.js';
 export { pptxToMarkdownDoc, pptxToContainer } from './import.js';
 
+// Layout inference (used by the PPTX importer and the theme dialog)
+export { analyzePptxLayouts, inspectPptxLayouts } from './layouts.js';
+export type {
+  AnalyzedLayout,
+  AnalyzePptxLayoutsOptions,
+  ExtractedPlaceholder,
+  ExtractedSlideLayout,
+  InspectPptxLayoutsOptions,
+  LayoutVerdict,
+  PptxColorHints,
+  PptxLayoutInference,
+  PptxLayoutSummary,
+} from './layouts.js';
+
 /**
  * Convert a .pptx file to a squisq Doc (via the markdown model).
  */