feat: chat uploads, vision turn isolation, and MindSpace agent improvements

Add chat file/image upload UX, attachment proxying, vision thumbnails, and per-turn image scoping so agents only use the current upload. Extend MindSpace asset context, billing token state, OA/scenario verify scripts, and related runtime config.

Co-authored-by: Cursor <cursoragent@cursor.com>

tkmind-proxy-vision.test.mjs

@@ -41,6 +41,23 @@ test('extractImageUrlsFromMessage deduplicates image urls', () => {
   assert.deepEqual(urls, [url]);
 });
 
+test('extractImageUrlsFromMessage ignores content aliases when metadata is present', () => {
+  const urls = extractImageUrlsFromMessage({
+    metadata: {
+      imageUrls: ['/api/mindspace/v1/assets/asset-4/download?inline=1'],
+    },
+    content: [
+      {
+        type: 'text',
+        text:
+          '[图片1]: https://m.tkmind.cn/MindSpace/user/public/images/2026-07-10/asset-4.jpg',
+      },
+    ],
+  });
+
+  assert.deepEqual(urls, ['/api/mindspace/v1/assets/asset-4/download?inline=1']);
+});
+
 test('buildVisionPayload marks one billable image analysis when vision succeeds', async () => {
   const result = await buildVisionPayload({
     userId: 'user-1',
@@ -59,6 +76,7 @@ test('buildVisionPayload marks one billable image analysis when vision succeeds'
 
   assert.equal(result?.billableImageCount, 1);
   assert.match(result?.userMessage?.content?.[0]?.text ?? '', /Qwen VL 图片描述/);
+  assert.match(result?.userMessage?.content?.[0]?.text ?? '', /本轮图片主题/);
   assert.equal(
     result?.userMessage?.metadata?.displayText,
     '根据图片生成页面\n\n[图片1]: /api/mindspace/v1/assets/asset-1/download?inline=1',