diff --git a/docs/reference/memport.md b/docs/reference/memport.md
index a8c2da5a2d..882fd3dfe8 100644
--- a/docs/reference/memport.md
+++ b/docs/reference/memport.md
@@ -120,6 +120,12 @@ code spans, ensuring that `@` imports inside these regions are properly ignored.
 This provides robust handling of nested code blocks and complex Markdown
 structures.
 
+## HTML comment handling
+
+The import processor removes HTML comments from memory content before resolving
+imports. Any `@` imports inside HTML comments are ignored, and the comments do
+not appear in the final prompt.
+
 ## Import tree structure
 
 The processor returns an import tree that shows the hierarchy of imported files,
diff --git a/packages/core/src/utils/memoryImportProcessor.test.ts b/packages/core/src/utils/memoryImportProcessor.test.ts
index 3c9a74b604..57cf2feb7a 100644
--- a/packages/core/src/utils/memoryImportProcessor.test.ts
+++ b/packages/core/src/utils/memoryImportProcessor.test.ts
@@ -378,6 +378,32 @@ describe('memoryImportProcessor', () => {
       );
     });
 
+    it('should strip HTML comments and ignore imports inside them', async () => {
+      const content = [
+        'Header',
+        '<!-- user comment that should be removed -->',
+        'Real import @./real.md',
+        'Inline comment <!-- @./fake.md --> after text',
+      ].join('\n');
+      const basePath = testPath('test', 'path');
+      const importedContent = 'Real imported content';
+
+      mockedFs.access.mockResolvedValue(undefined);
+      mockedFs.readFile.mockResolvedValue(importedContent);
+
+      const result = await processImports(content, basePath, true);
+
+      expect(result.content).toContain(importedContent);
+      expect(result.content).not.toContain('user comment that should be removed');
+      expect(result.content).not.toContain('fake.md');
+
+      expect(mockedFs.readFile).toHaveBeenCalledTimes(1);
+      expect(mockedFs.readFile).toHaveBeenCalledWith(
+        path.resolve(basePath, './real.md'),
+        'utf-8',
+      );
+    });
+
     it('should handle nested tokens and non-unique content correctly', async () => {
       // This test verifies the robust findCodeRegions implementation
       // that recursively walks the token tree and handles non-unique content
diff --git a/packages/core/src/utils/memoryImportProcessor.ts b/packages/core/src/utils/memoryImportProcessor.ts
index dc4b0b8537..aa64e18c4d 100644
--- a/packages/core/src/utils/memoryImportProcessor.ts
+++ b/packages/core/src/utils/memoryImportProcessor.ts
@@ -176,6 +176,31 @@ function findCodeRegions(content: string): Array<[number, number]> {
   return regions;
 }
 
+function stripHtmlComments(content: string): string {
+  const startToken = '<!--';
+  const endToken = '-->';
+  let start = content.indexOf(startToken);
+  if (start === -1) return content;
+
+  const parts: string[] = [];
+  let cursor = 0;
+
+  while (start !== -1) {
+    parts.push(content.slice(cursor, start));
+
+    const end = content.indexOf(endToken, start + startToken.length);
+    if (end === -1) {
+      return parts.join('');
+    }
+
+    cursor = end + endToken.length;
+    start = content.indexOf(startToken, cursor);
+  }
+
+  parts.push(content.slice(cursor));
+  return parts.join('');
+}
+
 /**
  * Processes import statements in GEMINI.md content
  * Supports @path/to/file syntax for importing content from other files
@@ -200,6 +225,8 @@ export async function processImports(
   importFormat: 'flat' | 'tree' = 'tree',
   boundaryMarkers: readonly string[] = ['.git'],
 ): Promise<ProcessImportsResult> {
+  content = stripHtmlComments(content);
+
   if (!projectRoot) {
     projectRoot = await findProjectRoot(basePath, boundaryMarkers);
   }
@@ -230,6 +257,7 @@ export async function processImports(
       filePath: string,
       depth: number,
     ) {
+      const sanitizedContent = stripHtmlComments(fileContent);
       // Normalize the file path to ensure consistent comparison
       const normalizedPath = path.normalize(filePath);
 
@@ -240,11 +268,11 @@ export async function processImports(
       processedFiles.add(normalizedPath);
 
       // Add this file to the flat list
-      flatFiles.push({ path: normalizedPath, content: fileContent });
+      flatFiles.push({ path: normalizedPath, content: sanitizedContent });
 
       // Find imports in this file
-      const codeRegions = findCodeRegions(fileContent);
-      const imports = findImports(fileContent);
+      const codeRegions = findCodeRegions(sanitizedContent);
+      const imports = findImports(sanitizedContent);
 
       // Process imports in reverse order to handle indices correctly
       for (let i = imports.length - 1; i >= 0; i--) {