fix: prefer UTF-8 encoding for shell output on Windows when detected

packages/core/src/utils/systemEncoding.test.ts

@@ -391,6 +391,19 @@ describe('Shell Command Processor - Encoding Functions', () => {
       expect(result).toBe('windows-1252');
     });
 
+    it('should prioritize UTF-8 detection over Windows system encoding', () => {
+      mockedOsPlatform.mockReturnValue('win32');
+      mockedExecSync.mockReturnValue('Active code page: 936'); // GBK
+
+      const buffer = Buffer.from('test');
+      // Mock chardet to return UTF-8
+      mockedChardetDetect.mockReturnValue('UTF-8');
+
+      const result = getCachedEncodingForBuffer(buffer);
+
+      expect(result).toBe('utf-8');
+    });
+
     it('should cache null system encoding result', () => {
       // Reset the cache specifically for this test
       resetEncodingCache();

packages/core/src/utils/systemEncoding.ts

@@ -34,6 +34,15 @@ export function getCachedEncodingForBuffer(buffer: Buffer): string {
 
   // If we have a cached system encoding, use it
   if (cachedSystemEncoding) {
+    // If the system encoding is not UTF-8 (e.g. Windows CP936), but the buffer
+    // is detected as UTF-8, prefer UTF-8. This handles tools like 'git' which
+    // often output UTF-8 regardless of the system code page.
+    if (cachedSystemEncoding !== 'utf-8') {
+      const detected = detectEncodingFromBuffer(buffer);
+      if (detected === 'utf-8') {
+        return 'utf-8';
+      }
+    }
     return cachedSystemEncoding;
   }