feat(copilot): enhance OCR handling with inline tags and document serializer

- Replace HTML comment OCR metadata with inline `<OCR:...>` tags
- Implement serializer-based markdown conversion for prefix/suffix content
- Add extractTextFromOCR utility function for text extraction
- Enable Table, Diagram, and ListCheck features in MilkdownEditor
- Add periodic debug logging for document state analysis

src/plugins/copilotPlugin.ts

@@ -1,14 +1,15 @@
 import { Plugin, PluginKey, Selection } from '@milkdown/prose/state'
 import { $prose, $ctx, $markSchema } from '@milkdown/kit/utils'
-import { parserCtx } from '@milkdown/kit/core'
+import { parserCtx, serializerCtx } from '@milkdown/kit/core'
 import { Node as ProseNode, Fragment } from '@milkdown/prose/model'
 import type { Ctx } from '@milkdown/kit/core'
 import type { EditorView } from '@milkdown/prose/view'
-import { getOcrCache, checkSizeLimit as checkOcrSizeLimit, OCR_SIZE_LIMIT } from '../utils/ocrCache'
+import { getOcrCache, checkSizeLimit as checkOcrSizeLimit, OCR_SIZE_LIMIT, extractTextFromOCR } from '../utils/ocrCache'
 
 const COPILOT_PLUGIN_KEY = new PluginKey('milkdown-copilot')
 const DEBOUNCE_MS = 1000
 const SIZE_LIMIT = OCR_SIZE_LIMIT
+const DEBUG = true
 const IMAGE_NODE_TYPES = new Set(['image', 'image-block', 'imageBlock'])
 
 interface CopilotState {
@@ -279,27 +280,57 @@ function extractImageFilenames(doc: ProseNode): string[] {
   return filenames
 }
 
-function buildPrefixWithOCR(prefix: string, doc: ProseNode, cursorPos: number): string {
-  const ocrEntries: string[] = []
+function buildPrefixWithOCRFromMarkdown(
+  doc: ProseNode,
+  cursorPos: number,
+  prefixMarkdown: string,
+  serializer: any,
+  schema: any
+): string {
+  const imageNodes: Array<{pos: number, src: string, label: string}> = []
 
   doc.descendants((node: ProseNode, pos) => {
-    if (pos >= cursorPos) return false
-    if (!isImageNodeWithSrc(node)) return true
-
+    if (!isImageNodeWithSrc(node)) return pos < cursorPos
     const src = getImageSrc(node)
-    const ocrText = getOcrCache(src)
-    if (!ocrText) return true
-
     const label = getImageLabel(node)
-    const safeOcrText = ocrText.replace(/<!--|-->/g, '').trim()
-    if (!safeOcrText) return true
-
-    ocrEntries.push(`image(${label}): ${safeOcrText}`)
-    return true
+    imageNodes.push({ pos, src, label })
+    return pos < cursorPos
   })
 
-  if (!ocrEntries.length) return prefix
-  return `${prefix}\n\n<!--OCR:\n${ocrEntries.join('\n')}\n-->`
+  if (imageNodes.length === 0) {
+    return prefixMarkdown
+  }
+
+  imageNodes.sort((a, b) => a.pos - b.pos)
+
+  const parts: string[] = []
+  let lastPos = 0
+
+  for (const img of imageNodes) {
+    if (img.pos > lastPos) {
+      const slice = doc.slice(lastPos, img.pos)
+      const sliceDoc = schema.topNodeType.createAndFill(undefined, slice.content)
+      parts.push(sliceDoc ? serializer(sliceDoc) : doc.textBetween(lastPos, img.pos))
+    }
+    const imageSyntax = `![${img.label}](${img.src})`
+    parts.push(imageSyntax)
+    const ocrText = getOcrCache(img.src)
+    if (ocrText) {
+      const textOnly = extractTextFromOCR(ocrText, 100)
+      if (textOnly) {
+        parts.push(` <OCR:${textOnly}>`)
+      }
+    }
+    lastPos = img.pos + 1
+  }
+
+  if (lastPos < cursorPos) {
+    const slice = doc.slice(lastPos, cursorPos)
+    const sliceDoc = schema.topNodeType.createAndFill(undefined, slice.content)
+    parts.push(sliceDoc ? serializer(sliceDoc) : doc.textBetween(lastPos, cursorPos))
+  }
+
+  return parts.join('')
 }
 
 function doFetchSuggestion(view: EditorView, runtime: CopilotRuntime, pos: number, prefix: string, suffix: string) {
@@ -339,6 +370,7 @@ function scheduleFetch(view: EditorView, runtime: CopilotRuntime, pos: number, p
   if (!runtime.enabled) return
 
   const doc = view.state.doc
+  const schema = view.state.schema
   const imageFilenames = extractImageFilenames(doc)
   const { overLimit } = checkOcrSizeLimit(doc.content.size, imageFilenames)
 
@@ -347,7 +379,61 @@ function scheduleFetch(view: EditorView, runtime: CopilotRuntime, pos: number, p
     return
   }
 
-  const prefixWithOCR = buildPrefixWithOCR(prefix, doc, pos)
+  const serializer = runtime.ctx.get(serializerCtx)
+  
+  // 尝试使用 serializer 将文档切片转换为 Markdown
+  let prefixMarkdown = ''
+  let suffixMarkdown = ''
+  
+  try {
+    // 方法1: 使用 slice 创建文档节点
+    const prefixSlice = doc.slice(0, pos)
+    if (prefixSlice.content.size > 0) {
+      const prefixDoc = schema.topNodeType.createAndFill(undefined, prefixSlice.content)
+      if (prefixDoc) {
+        prefixMarkdown = serializer(prefixDoc)
+      }
+    }
+    if (!prefixMarkdown) {
+      // 方法2: 直接序列化整个文档然后截取
+      const fullMarkdown = serializer(doc)
+      const fullDoc = view.state.doc
+      const totalLen = fullDoc.content.size
+      if (totalLen > 0 && pos < totalLen) {
+        // 简单估算位置
+        prefixMarkdown = fullMarkdown.substring(0, Math.floor(fullMarkdown.length * pos / totalLen))
+      }
+    }
+    if (!prefixMarkdown) {
+      // 回退到 textBetween 但添加换行符
+      prefixMarkdown = doc.textBetween(0, pos, '\n', '\n')
+    }
+    
+    // Suffix
+    const suffixSlice = doc.slice(pos)
+    if (suffixSlice.content.size > 0) {
+      const suffixDoc = schema.topNodeType.createAndFill(undefined, suffixSlice.content)
+      if (suffixDoc) {
+        suffixMarkdown = serializer(suffixDoc)
+      }
+    }
+    if (!suffixMarkdown) {
+      suffixMarkdown = doc.textBetween(pos, doc.content.size, '\n', '\n')
+    }
+  } catch (e) {
+    console.error('[Copilot] Serializer error:', e)
+    prefixMarkdown = doc.textBetween(0, pos, '\n', '\n')
+    suffixMarkdown = doc.textBetween(pos, doc.content.size, '\n', '\n')
+  }
+
+  const prefixWithOCR = buildPrefixWithOCRFromMarkdown(doc, pos, prefixMarkdown, serializer, schema)
+
+  if (DEBUG) {
+    console.log('[Copilot] ===== LLM Request =====')
+    console.log('[Copilot] PREFIX:', prefixWithOCR)
+    console.log('[Copilot] SUFFIX:', suffixMarkdown)
+    console.log('[Copilot] ======================')
+  }
 
   if (runtime.debounceTimer) {
     clearTimeout(runtime.debounceTimer)
@@ -357,7 +443,7 @@ function scheduleFetch(view: EditorView, runtime: CopilotRuntime, pos: number, p
   const debounceMs = runtime.ctx.get(copilotConfigCtx.key).debounceMs ?? DEBOUNCE_MS
   runtime.debounceTimer = setTimeout(() => {
     runtime.debounceTimer = null
-    doFetchSuggestion(view, runtime, pos, prefixWithOCR, suffix)
+    doFetchSuggestion(view, runtime, pos, prefixWithOCR, suffixMarkdown)
   }, debounceMs)
 }