From c5e746b6c619519520b56f5ecbf92413b9cfcda7 Mon Sep 17 00:00:00 2001
From: Phantom <59059173+EurFelux@users.noreply.github.com>
Date: Sat, 30 Aug 2025 18:24:44 +0800
Subject: [PATCH] fix: filter inline base64 image in messages summary (#9687)
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

* feat(markdown): 添加清理base64图片链接的功能

添加purifyMarkdownImages函数用于将Markdown中的base64图片链接替换为普通链接格式

* fix(utils): 清理markdown中的base64图片链接并应用到消息摘要

在ApiService中调用purifyMarkdownImages清理消息摘要中的base64图片链接
---
 src/renderer/src/services/ApiService.ts       |  3 +-
 .../src/utils/__tests__/markdown.test.ts      | 67 +++++++++++++++++++
 src/renderer/src/utils/markdown.ts            | 18 +++++
 3 files changed, 87 insertions(+), 1 deletion(-)

diff --git a/src/renderer/src/services/ApiService.ts b/src/renderer/src/services/ApiService.ts
index d86854acec..d1e2a18664 100644
--- a/src/renderer/src/services/ApiService.ts
+++ b/src/renderer/src/services/ApiService.ts
@@ -42,6 +42,7 @@ import { removeSpecialCharactersForTopicName, uuid } from '@renderer/utils'
 import { abortCompletion } from '@renderer/utils/abortController'
 import { isAbortError } from '@renderer/utils/error'
 import { extractInfoFromXML, ExtractResults } from '@renderer/utils/extract'
+import { purifyMarkdownImages } from '@renderer/utils/markdown'
 import { filterAdjacentUserMessaegs, filterLastAssistantMessage } from '@renderer/utils/messageUtils/filters'
 import { findFileBlocks, getMainTextContent } from '@renderer/utils/messageUtils/find'
 import {
@@ -699,7 +700,7 @@ export async function fetchMessagesSummary({ messages, assistant }: { messages:
   const structredMessages = contextMessages.map((message) => {
     const structredMessage = {
       role: message.role,
-      mainText: getMainTextContent(message)
+      mainText: purifyMarkdownImages(getMainTextContent(message))
     }
 
     // 让LLM知道消息中包含的文件，但只提供文件名
diff --git a/src/renderer/src/utils/__tests__/markdown.test.ts b/src/renderer/src/utils/__tests__/markdown.test.ts
index 3b3c4f6cd6..9418af0dc5 100644
--- a/src/renderer/src/utils/__tests__/markdown.test.ts
+++ b/src/renderer/src/utils/__tests__/markdown.test.ts
@@ -9,6 +9,7 @@ import {
   isHtmlCode,
   markdownToPlainText,
   processLatexBrackets,
+  purifyMarkdownImages,
   removeTrailingDoubleSpaces,
   updateCodeBlock
 } from '../markdown'
@@ -634,4 +635,70 @@ $$
       expect(isHtmlCode('a < b')).toBe(false)
     })
   })
+
+  describe('purifyMarkdownImages', () => {
+    it('should replace base64 image with placeholder', () => {
+      const input = '![cat](data:image/png;base64,iVBORw0KGgo)'
+      const expected = '![cat](image_url)'
+      expect(purifyMarkdownImages(input)).toBe(expected)
+    })
+
+    it('should handle multiple base64 images', () => {
+      const input = `
+      ![dog](data:image/jpeg;base64,ABC123)
+      Some text
+      ![avatar](data:image/png;base64,XYZ789)
+    `
+      const expected = `
+      ![dog](image_url)
+      Some text
+      ![avatar](image_url)
+    `
+      expect(purifyMarkdownImages(input)).toBe(expected)
+    })
+
+    it('should ignore normal image links', () => {
+      const input = '![cat](https://example.com/cat.png)'
+      expect(purifyMarkdownImages(input)).toBe(input)
+    })
+
+    it('should handle whitespace in base64 url', () => {
+      const input = '![logo](  data:image/svg+xml;base64,CONTENT  )'
+      const expected = '![logo](image_url)'
+      expect(purifyMarkdownImages(input)).toBe(expected)
+    })
+
+    it('should preserve alt text', () => {
+      const input = '![User Avatar](data:image/png;base64,xxx)'
+      const expected = '![User Avatar](image_url)'
+      expect(purifyMarkdownImages(input)).toBe(expected)
+    })
+
+    it('should handle uppercase data URL', () => {
+      const input = '![test](DATA:IMAGE/PNG;BASE64,ABC)'
+      const expected = '![test](image_url)'
+      expect(purifyMarkdownImages(input)).toBe(expected)
+    })
+
+    it('should not modify text that is not image', () => {
+      const input = 'This is a data:image/png;base64,iVBORw line of text'
+      expect(purifyMarkdownImages(input)).toBe(input)
+    })
+
+    it('should handle mixed content', () => {
+      const input = `
+      Regular: ![cat](https://example.com/cat.png)
+      Base64: ![dog](data:image/jpeg;base64,BASE64DATA)
+      Another: ![bird](https://example.com/bird.gif)
+      Inline: ![icon](  data:image/x-icon;base64,ICONDATA  )
+    `
+      const expected = `
+      Regular: ![cat](https://example.com/cat.png)
+      Base64: ![dog](image_url)
+      Another: ![bird](https://example.com/bird.gif)
+      Inline: ![icon](image_url)
+    `
+      expect(purifyMarkdownImages(input)).toBe(expected)
+    })
+  })
 })
diff --git a/src/renderer/src/utils/markdown.ts b/src/renderer/src/utils/markdown.ts
index 60bdaaac32..bbeb2a2f74 100644
--- a/src/renderer/src/utils/markdown.ts
+++ b/src/renderer/src/utils/markdown.ts
@@ -312,3 +312,21 @@ export const markdownToPlainText = (markdown: string): string => {
   // 直接用 remove-markdown 库，使用默认的 removeMarkdown 参数
   return removeMarkdown(markdown)
 }
+
+/**
+ * 清理 Markdown 中的 base64 图片链接
+ *
+ * 将 Markdown 中的 base64 格式图片链接替换为普通链接格式。
+ *
+ * @param {string} markdown - 包含图片链接的 Markdown 文本
+ * @returns {string} 处理后的 Markdown 文本，所有 base64 图片链接都被替换为普通链接
+ * @example
+ * - 输入: `![image](data:image/png;base64,iVBORw0...)`
+ * - 输出: `![image](image_url)`
+ */
+export const purifyMarkdownImages = (markdown: string): string => {
+  return markdown.replace(
+    /(!\[[^\]]*\]\()\s*data:image\/[\w+.-]+;base64\s*,[\w+/=]+(?:\s*[\w+/=]+)*\s*\)/gi,
+    '$1image_url)'
+  )
+}