fix(parsers): keep date fields, drop file-name image alt text, fast-path plain cells

Only the slide-number field is layout text; date and time fields outside
a dt placeholder are content, and skipping them emptied a deck made of
them. Image alternative text that is a bare file name or an auto caption
is noise, so the HTML, PresentationML, and OpenDocument walkers share
one filter. A table cell with no element children is read directly
instead of running the block-spacing and nested-table queries.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

Author: waleedlatif1

apps/sim/lib/file-parsers/html-parser.test.ts

@@ -159,6 +159,18 @@ describe('HtmlParser', () => {
       expect(result.metadata?.tableCount).toBe(2)
     })
 
+    it('keeps descriptive image alt text and drops file-name alt text', async () => {
+      const buffer = Buffer.from(
+        `<body><img alt="Org chart"><img alt="python-logo.gif"><img alt="Image 2"><p>Body</p></body>`
+      )
+
+      const result = await parser.parseBuffer(buffer)
+
+      expect(result.content).toContain('[Image: Org chart]')
+      expect(result.content).not.toContain('python-logo')
+      expect(result.content).not.toContain('Image 2')
+    })
+
     it('separates block elements inside a list item', async () => {
       const buffer = Buffer.from(
         `<body><ul><li><div><p>Versions</p><p>Release Information</p></div></li></ul></body>`

apps/sim/lib/file-parsers/html-parser.ts

@@ -3,6 +3,7 @@ import { createLogger } from '@sim/logger'
 import { getErrorMessage } from '@sim/utils/errors'
 import * as cheerio from 'cheerio'
 import { FileParserError } from '@/lib/file-parsers/errors'
+import { imageAltText } from '@/lib/file-parsers/office-text'
 import type { FileParseResult, FileParser } from '@/lib/file-parsers/types'
 import { decodeTextBuffer, sanitizeTextForUTF8 } from '@/lib/file-parsers/utils'
 
@@ -259,10 +260,8 @@ function processElement(
       }
 
       case 'img': {
-        const alt = $node.attr('alt')
-        if (alt) {
-          contentParts.push(`[Image: ${alt}]`)
-        }
+        const image = imageAltText($node.attr('alt'))
+        if (image) contentParts.push(image)
         break
       }
 
@@ -372,14 +371,19 @@ function flattenedTableCells($: cheerio.CheerioAPI, table: cheerio.Cheerio<AnyNo
 }
 
 /**
- * One cell's text on a single line. Block elements inside the cell get a space
+ * One cell's text on a single line. A text-only cell — the common case in a
+ * data table — is read directly. Block elements inside the cell get a space
  * so adjacent paragraphs do not glue together — this mutates the live DOM, and
  * runs before `extractHeadings`/`extractLinks`, so heading or link text inside a
  * cell gains those spaces too. A nested table is rendered on a clone of the cell
  * as its cells joined with ` / `, without `[Table]` markers or pipes, so the
  * outer row stays one line and the inner text appears exactly once.
  */
 function cellText($: cheerio.CheerioAPI, cell: cheerio.Cheerio<AnyNode>): string {
+  if (cell.children().length === 0) {
+    return cell.text().replace(/\s+/g, ' ').trim()
+  }
+
   const nested = topLevelNestedTables($, cell)
   if (nested.length === 0) {
     cell.find(CELL_BLOCK_SELECTOR).after(' ')

apps/sim/lib/file-parsers/odf-text.test.ts

@@ -127,6 +127,14 @@ describe('extractOpenDocumentText', () => {
     expect(await extractOpenDocumentText(buffer)).toBe('[Image: Org chart]')
   })
 
+  it('drops a file-name image title', async () => {
+    const buffer = await text(
+      `<text:p><draw:frame draw:name="Image1"><draw:image xlink:href="Pictures/a.png" xmlns:xlink="http://www.w3.org/1999/xlink"/><svg:title xmlns:svg="urn:oasis:names:tc:opendocument:xmlns:svg-compatible:1.0">python-icon.jpeg</svg:title></draw:frame>after</text:p>`
+    )
+
+    expect(await extractOpenDocumentText(buffer)).toBe('after')
+  })
+
   it('rejects a content part above the per-part size cap before parsing it', async () => {
     const buffer = await text('<text:p>Small</text:p>')
     const zip = await JSZip.loadAsync(buffer)

apps/sim/lib/file-parsers/odf-text.ts

@@ -5,6 +5,7 @@ import {
   collapseWhitespace,
   findFirst,
   formatTableRow,
+  imageAltText,
   isXmlElement,
   joinBlocks,
   NOTES_MARKER,
@@ -246,8 +247,7 @@ function imageFrameText(frame: XmlElement, state: WalkState): string | null {
   const children = childElements(frame)
   if (!children.some((child) => child.name === 'draw:image')) return null
   const alt = children.find((child) => child.name === 'svg:title' || child.name === 'svg:desc')
-  const text = alt ? collapseWhitespace(inlineText(alt, state)) : ''
-  return text ? `[Image: ${text}]` : ''
+  return alt ? (imageAltText(inlineText(alt, state)) ?? '') : ''
 }
 
 function emitNotes(notes: XmlElement, state: WalkState): void {

apps/sim/lib/file-parsers/office-text.ts

@@ -88,6 +88,22 @@ export function findAll(node: XmlDocument | XmlElement, tagName: string): XmlEle
   return DomUtils.findAll((element) => element.name === tagName, node.children)
 }
 
+/** A bare file name (`python-logo.gif`) rather than a description. */
+const FILENAME_LIKE = /^[^\s]+\.[a-z0-9]{2,4}$/i
+
+/** Auto-generated captions that name the object, not its content (`Picture 3`). */
+const AUTO_CAPTION = /^(?:picture|image|graphic|photo|figure|chart|diagram|screenshot)\s*\d*$/i
+
+/**
+ * Renders an image's alternative text as `[Image: …]`, or `null` when the text
+ * is a file name or an auto-generated caption, which would only add noise.
+ */
+export function imageAltText(raw: string | undefined): string | null {
+  const text = raw ? collapseWhitespace(raw) : ''
+  if (!text || FILENAME_LIKE.test(text) || AUTO_CAPTION.test(text)) return null
+  return `[Image: ${text}]`
+}
+
 /** Collapses internal whitespace so a cell or list item occupies a single line. */
 export function collapseWhitespace(text: string): string {
   return text.replace(/\s+/g, ' ').trim()

apps/sim/lib/file-parsers/ooxml-presentation.test.ts

@@ -145,14 +145,24 @@ describe('extractPresentationText', () => {
     )
   })
 
-  it('skips slide-number and date fields outside their placeholders', async () => {
-    const spTree = `<p:sp><p:nvSpPr><p:cNvPr id="1" name="s"/><p:cNvSpPr/></p:nvSpPr><p:txBody><a:p><a:r><a:t>Page </a:t></a:r><a:fld type="slidenum"><a:t>369</a:t></a:fld><a:fld type="datetime1"><a:t>1/1/2026</a:t></a:fld><a:fld type="custom"><a:t>kept</a:t></a:fld></a:p></p:txBody></p:sp>`
+  it('skips slide-number fields outside their placeholder but keeps date fields', async () => {
+    const spTree = `<p:sp><p:nvSpPr><p:cNvPr id="1" name="s"/><p:cNvSpPr/></p:nvSpPr><p:txBody><a:p><a:r><a:t>Page </a:t></a:r><a:fld type="slidenum"><a:t>369</a:t></a:fld><a:fld type="datetime1"><a:t>6/29/2021</a:t></a:fld><a:fld type="custom"><a:t>kept</a:t></a:fld></a:p></p:txBody></p:sp>`
 
-    expect(await extractPresentationText(await buildDeck([{ index: 1, spTree }]))).toBe('Page kept')
+    expect(await extractPresentationText(await buildDeck([{ index: 1, spTree }]))).toBe(
+      'Page 6/29/2021kept'
+    )
+  })
+
+  it('still drops a date field inside a dt placeholder', async () => {
+    const spTree = `<p:sp><p:nvSpPr><p:cNvPr id="1" name="s"/><p:cNvSpPr/><p:nvPr><p:ph type="dt"/></p:nvPr></p:nvSpPr><p:txBody><a:p><a:fld type="datetime1"><a:t>6/29/2021</a:t></a:fld></a:p></p:txBody></p:sp>${shape('Body')}`
+
+    expect(await extractPresentationText(await buildDeck([{ index: 1, spTree }]))).toBe('Body')
   })
 
-  it('emits a picture as its alternative text', async () => {
-    const spTree = `<p:pic><p:nvPicPr><p:cNvPr id="4" name="Picture 3" descr="Org chart"/><p:cNvPicPr/><p:nvPr/></p:nvPicPr></p:pic>${shape('Caption')}`
+  it('emits a picture as its alternative text unless it is a file name or auto caption', async () => {
+    const pic = (descr: string) =>
+      `<p:pic><p:nvPicPr><p:cNvPr id="4" name="Picture 3" descr="${descr}"/><p:cNvPicPr/><p:nvPr/></p:nvPicPr></p:pic>`
+    const spTree = pic('Org chart') + pic('python-logo.gif') + pic('Picture 2') + shape('Caption')
 
     expect(await extractPresentationText(await buildDeck([{ index: 1, spTree }]))).toBe(
       '[Image: Org chart]\nCaption'

apps/sim/lib/file-parsers/ooxml-presentation.ts

@@ -4,6 +4,7 @@ import {
   findAll,
   findFirst,
   formatTableRow,
+  imageAltText,
   isXmlElement,
   joinBlocks,
   NOTES_MARKER,
@@ -31,8 +32,12 @@ const SLIDE_PART = /^ppt\/slides\/slide(\d+)\.xml$/
 const NOTES_RELATIONSHIP_SUFFIX = '/notesSlide'
 const NOTES_PART_PREFIX = 'ppt/notesSlides/'
 
-/** Fields PowerPoint fills at render time; their cached text is the layout's, not the author's. */
-const RENDER_TIME_FIELD_TYPES = /^(slidenum|datetime)/i
+/**
+ * The slide-number field's cached text is the layout's, not the author's. Date
+ * fields keep their text: outside a `dt` placeholder (already skipped) a deck's
+ * dates are content, and python-pptx keeps them too.
+ */
+const SLIDE_NUMBER_FIELD_TYPE = 'slidenum'
 
 /** Layout-chrome placeholders whose text is a field, not slide content. */
 const SKIPPED_PLACEHOLDER_TYPES = new Set(['sldNum', 'dt', 'ftr', 'hdr'])
@@ -49,7 +54,7 @@ function placeholderType(shape: XmlElement): string | null {
 
 /**
  * Concatenates a DrawingML paragraph's runs, turning `<a:br/>` into a newline
- * and skipping slide-number and date fields wherever they appear.
+ * and skipping slide-number fields wherever they appear.
  */
 function paragraphText(paragraph: XmlElement): string {
   const pieces: string[] = []
@@ -58,7 +63,7 @@ function paragraphText(paragraph: XmlElement): string {
       pieces.push('\n')
       return
     }
-    if (element.name === 'a:fld' && RENDER_TIME_FIELD_TYPES.test(element.attribs.type ?? '')) {
+    if (element.name === 'a:fld' && element.attribs.type === SLIDE_NUMBER_FIELD_TYPE) {
       return
     }
     if (element.name === 'a:t') {
@@ -124,8 +129,8 @@ function graphicFrameBlocks(frame: XmlElement): string[] {
 function pictureBlocks(picture: XmlElement): string[] {
   const nonVisual = childElements(picture).find((child) => child.name === 'p:nvPicPr')
   const properties = nonVisual ? findFirst(nonVisual, 'p:cNvPr') : null
-  const description = properties?.attribs.descr?.trim()
-  return description ? [`[Image: ${description}]`] : []
+  const image = imageAltText(properties?.attribs.descr)
+  return image ? [image] : []
 }
 
 /**