feat(config): add OCR URL configuration and improve image node handling

- Add VITE_OCR_URL environment variable with fallback URL construction
- Define IMAGE_NODE_TYPES constant to support 'image', 'image-block', and 'imageBlock' node types
- Add helper functions for safer image attribute access (getImageSrc, isImageNodeWithSrc, getImageLabel)
- Improve OCR error handling with HTTP status checking and error details
- Wrap OCR context in HTML comments to prevent prompt injection issues
- Update MilkdownEditor to use centralized OCR_URL configuration

src/plugins/copilotPlugin.ts

@@ -9,6 +9,7 @@ import { getOcrCache, checkSizeLimit as checkOcrSizeLimit, OCR_SIZE_LIMIT } from
 const COPILOT_PLUGIN_KEY = new PluginKey('milkdown-copilot')
 const DEBOUNCE_MS = 1000
 const SIZE_LIMIT = OCR_SIZE_LIMIT
+const IMAGE_NODE_TYPES = new Set(['image', 'image-block', 'imageBlock'])
 
 interface CopilotState {
   from: number
@@ -251,11 +252,28 @@ function insertPlainText(view: EditorView, suggestion: string, from: number, mar
   view.dispatch(tr)
 }
 
+function getImageSrc(node: ProseNode): string {
+  const src = node.attrs?.src
+  return typeof src === 'string' ? src : ''
+}
+
+function isImageNodeWithSrc(node: ProseNode): boolean {
+  return IMAGE_NODE_TYPES.has(node.type.name) && Boolean(getImageSrc(node))
+}
+
+function getImageLabel(node: ProseNode): string {
+  const candidates = [node.attrs?.alt, node.attrs?.title, node.attrs?.caption]
+  for (const value of candidates) {
+    if (typeof value === 'string' && value.trim()) return value.trim()
+  }
+  return 'untitled'
+}
+
 function extractImageFilenames(doc: ProseNode): string[] {
   const filenames: string[] = []
   doc.descendants((node: ProseNode) => {
-    if (node.type.name === 'image' && node.attrs.src) {
-      filenames.push(node.attrs.src)
+    if (isImageNodeWithSrc(node)) {
+      filenames.push(getImageSrc(node))
     }
   })
   return filenames
@@ -266,18 +284,22 @@ function buildPrefixWithOCR(prefix: string, doc: ProseNode, cursorPos: number):
 
   doc.descendants((node: ProseNode, pos) => {
     if (pos >= cursorPos) return false
-    if (node.type.name !== 'image' || !node.attrs.src) return true
+    if (!isImageNodeWithSrc(node)) return true
 
-    const ocrText = getOcrCache(node.attrs.src)
+    const src = getImageSrc(node)
+    const ocrText = getOcrCache(src)
     if (!ocrText) return true
 
-    const altText = typeof node.attrs.alt === 'string' ? node.attrs.alt : ''
-    ocrEntries.push(`image(${altText || 'untitled'}): ${ocrText}`)
+    const label = getImageLabel(node)
+    const safeOcrText = ocrText.replace(/<!--|-->/g, '').trim()
+    if (!safeOcrText) return true
+
+    ocrEntries.push(`image(${label}): ${safeOcrText}`)
     return true
   })
 
   if (!ocrEntries.length) return prefix
-  return `${prefix}\n\n[OCR Context]\n${ocrEntries.join('\n')}`
+  return `${prefix}\n\n<!--OCR:\n${ocrEntries.join('\n')}\n-->`
 }
 
 function doFetchSuggestion(view: EditorView, runtime: CopilotRuntime, pos: number, prefix: string, suffix: string) {