fix: harden speech and document interactions

src/main/context-manager.test.ts

@@ -1,6 +1,7 @@
 import { mkdtemp, rm, writeFile } from 'node:fs/promises'
 import { tmpdir } from 'node:os'
 import { basename, join } from 'node:path'
+import { strToU8, zipSync } from 'fflate'
 import { afterEach, describe, expect, it, vi } from 'vitest'
 
 const { createFromBuffer, getSources, showOpenDialog } = vi.hoisted(() => ({
@@ -175,6 +176,59 @@ describe('ContextManager', () => {
     )
   })
 
+  it('extracts explicitly selected Office documents into bounded text context', async () => {
+    const directory = await mkdtemp(join(tmpdir(), 'goodbuddy-context-'))
+    temporaryDirectories.push(directory)
+    const filePath = join(directory, '需求说明.docx')
+    await writeFile(
+      filePath,
+      Buffer.from(
+        zipSync({
+          'word/document.xml': strToU8(
+            '<w:document><w:p><w:t>Word 需求正文</w:t></w:p></w:document>'
+          )
+        })
+      )
+    )
+    showOpenDialog.mockResolvedValue({
+      canceled: false,
+      filePaths: [filePath]
+    })
+    const manager = new ContextManager()
+
+    const [attachment] = await manager.selectFiles({} as BrowserWindow)
+
+    expect(attachment).toMatchObject({
+      name: '需求说明.docx',
+      kind: 'text',
+      preview: '[正文] Word 需求正文'
+    })
+    expect(showOpenDialog).toHaveBeenCalledWith(
+      expect.anything(),
+      expect.objectContaining({
+        filters: expect.arrayContaining([
+          expect.objectContaining({
+            name: 'PDF 和 Office 文档',
+            extensions: expect.arrayContaining([
+              'docx',
+              'pdf',
+              'pptx',
+              'xlsx'
+            ])
+          })
+        ])
+      })
+    )
+    const prompt = manager.enrichRequest({
+      requestId: '1f6a37b6-e0a3-449f-8878-b10d353fbfb4',
+      conversationId: 'conversation-1',
+      prompt: '总结文档',
+      contextIds: [attachment!.id]
+    }).prompt
+    expect(prompt).toContain('Word 需求正文')
+    expect(prompt).toContain('"content":"[正文]\\nWord 需求正文"')
+  })
+
   it('keeps all five explicitly selected images', async () => {
     const directory = await mkdtemp(join(tmpdir(), 'goodbuddy-context-'))
     temporaryDirectories.push(directory)

src/main/context-manager.ts

@@ -20,6 +20,7 @@ import type {
   AgentImage
 } from './agent/runtime'
 import { encodeBoundedJpeg } from './bounded-jpeg'
+import { parseDocument } from './knowledge/document-parser'
 
 type StoredTextContext = ContextAttachment & {
   kind: 'text'
@@ -35,6 +36,7 @@ type StoredImageContext = ContextAttachment & {
 type StoredContext = StoredTextContext | StoredImageContext
 
 const maximumFileSize = 256 * 1024
+const maximumDocumentFileSize = 20 * 1024 * 1024
 const maximumContextBytes = 12 * 1024 * 1024
 const maximumContextCount = 16
 const maximumAttachmentsPerMessage = 8
@@ -68,6 +70,36 @@ const supportedImageExtensions = new Set([
   '.png',
   '.webp'
 ])
+const supportedDocumentExtensions = new Set([
+  '.docx',
+  '.pdf',
+  '.pptx',
+  '.xlsx'
+])
+
+function truncateUtf8(value: string, maximumBytes: number): string {
+  const buffer = Buffer.from(value)
+  if (buffer.byteLength <= maximumBytes) {
+    return value
+  }
+  const marker = '\n\n[文档内容过长，已截断]'
+  const markerBytes = Buffer.byteLength(marker)
+  return `${buffer
+    .subarray(0, maximumBytes - markerBytes)
+    .toString('utf8')
+    .replace(/\uFFFD$/u, '')}${marker}`
+}
+
+function formatParsedDocument(
+  sections: Awaited<ReturnType<typeof parseDocument>>['sections']
+): string {
+  return sections
+    .map(
+      (section) =>
+        `[${section.locator}]\n${section.content}`
+    )
+    .join('\n\n')
+}
 
 export class ContextManager {
   private readonly contexts = new Map<string, StoredContext>()
@@ -161,6 +193,12 @@ export class ContextManager {
           extensions: [...supportedImageExtensions].map((extension) =>
             extension.slice(1)
           )
+        },
+        {
+          name: 'PDF 和 Office 文档',
+          extensions: [...supportedDocumentExtensions].map((extension) =>
+            extension.slice(1)
+          )
         }
       ]
     })
@@ -178,7 +216,8 @@ export class ContextManager {
         const extension = extname(canonicalPath).toLowerCase()
         if (
           !supportedExtensions.has(extension) &&
-          !supportedImageExtensions.has(extension)
+          !supportedImageExtensions.has(extension) &&
+          !supportedDocumentExtensions.has(extension)
         ) {
           throw new Error(`不支持的文件类型：${extension || '未知'}`)
         }
@@ -204,6 +243,33 @@ export class ContextManager {
           }
           continue
         }
+        if (supportedDocumentExtensions.has(extension)) {
+          try {
+            const fileStat = await handle.stat()
+            if (
+              !fileStat.isFile() ||
+              fileStat.size > maximumDocumentFileSize
+            ) {
+              throw new Error('PDF 或 Office 文档必须小于 20MB 且不能是目录')
+            }
+            const parsed = await parseDocument(
+              basename(canonicalPath),
+              await handle.readFile()
+            )
+            attachments.push(
+              this.storeText(
+                basename(canonicalPath),
+                truncateUtf8(
+                  formatParsedDocument(parsed.sections),
+                  maximumFileSize
+                )
+              )
+            )
+          } finally {
+            await handle.close()
+          }
+          continue
+        }
         let content: string
         try {
           const fileStat = await handle.stat()

src/main/knowledge/document-parser.test.ts

@@ -2,6 +2,33 @@ import { strToU8, zipSync } from 'fflate'
 import { describe, expect, it } from 'vitest'
 import { chunkDocument, parseDocument } from './document-parser'
 
+function createPdfFixture(text: string): Buffer {
+  const stream = `BT /F1 18 Tf 50 100 Td (${text}) Tj ET`
+  const objects = [
+    '<< /Type /Catalog /Pages 2 0 R >>',
+    '<< /Type /Pages /Kids [3 0 R] /Count 1 >>',
+    '<< /Type /Page /Parent 2 0 R /MediaBox [0 0 300 200] /Resources << /Font << /F1 4 0 R >> >> /Contents 5 0 R >>',
+    '<< /Type /Font /Subtype /Type1 /BaseFont /Helvetica >>',
+    `<< /Length ${Buffer.byteLength(stream)} >>\nstream\n${stream}\nendstream`
+  ]
+  let content = '%PDF-1.4\n'
+  const offsets = [0]
+  for (const [index, object] of objects.entries()) {
+    offsets.push(Buffer.byteLength(content))
+    content += `${index + 1} 0 obj\n${object}\nendobj\n`
+  }
+  const xrefOffset = Buffer.byteLength(content)
+  content += `xref\n0 ${objects.length + 1}\n`
+  content += '0000000000 65535 f \n'
+  content += offsets
+    .slice(1)
+    .map((offset) => `${String(offset).padStart(10, '0')} 00000 n \n`)
+    .join('')
+  content += `trailer\n<< /Size ${objects.length + 1} /Root 1 0 R >>\n`
+  content += `startxref\n${xrefOffset}\n%%EOF\n`
+  return Buffer.from(content)
+}
+
 describe('document parser', () => {
   it('parses text and creates overlapping bounded chunks', async () => {
     const parsed = await parseDocument(
@@ -66,6 +93,21 @@ describe('document parser', () => {
     }
   })
 
+  it('extracts page text and locators from PDF files', async () => {
+    const parsed = await parseDocument(
+      'sample.pdf',
+      createPdfFixture('PDF body text')
+    )
+
+    expect(parsed.content).toContain('PDF body text')
+    expect(parsed.sections).toEqual([
+      {
+        locator: '第 1 页',
+        content: 'PDF body text'
+      }
+    ])
+  })
+
   it('rejects unsupported or oversized content', async () => {
     await expect(
       parseDocument('archive.zip', Buffer.from('not supported'))