feat: refine citations format and improve content presentation

Backend:
- Simplify citations prompt format and rules
- Add clear distinction between chat responses and file content
- Enforce full URL usage in markdown links, prohibit [cite-1] format
- Require content-first approach: write full content, then add citations at end

Frontend:
- Hide <citations> block in both chat messages and markdown preview
- Remove top-level Citations/Sources list for cleaner UI
- Auto-remove <citations> block in code editor view for markdown files
- Keep inline citation hover cards for reference details

This ensures citations are presented like Claude: clean content with inline reference badges.

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

backend/src/agents/lead_agent/prompt.py

@@ -123,31 +123,33 @@ You have access to skills that provide optimized workflows for specific tasks. E
 </response_style>
 
 <citations_format>
-**AUTOMATIC CITATION REQUIREMENT**: After using web_search tool, you MUST include citations in your response.
-
-**FORMAT** - Your response MUST start with a citations block, then content with inline links:
+**FORMAT** - After web_search, ALWAYS include citations in your output:
+**For chat responses:**
+Your visible response MUST start with citations block, then content with inline links:
 <citations>
 {{"id": "cite-1", "title": "Page Title", "url": "https://example.com/page", "snippet": "Brief description"}}
-{{"id": "cite-2", "title": "Another Source", "url": "https://another.com/article", "snippet": "What this covers"}}
 </citations>
+Content with inline links...
 
-Then your content: According to [Source Name](url), the findings show... [Another Source](url2) also reports...
+**For files (write_file):**
+File content MUST start with citations block, then content with inline links:
+<citations>
+{{"id": "cite-1", "title": "Page Title", "url": "https://example.com/page", "snippet": "Brief description"}}
+</citations>
+# Document Title
+Content with inline [Source Name](full_url) links...
 
 **RULES:**
-- DO NOT put citations in your thinking/reasoning - output them in your VISIBLE RESPONSE
-- DO NOT wait for user to ask - output citations AUTOMATICALLY after web search
-- DO NOT use number format like [1] or [2] - use source name like [Reuters](url)
-- The `<citations>` block MUST be FIRST in your response (before any other text)
-- Use source domain/brand name as link text (e.g., "Reuters", "TechCrunch", "智源研究院")
-- The URL in markdown link must match a URL in your citations block
-
-**IF writing markdown files**: When user asks you to create a report/document and you use write_file, use `[Source Name](url)` links in the file content (no <citations> block needed in files).
+- `<citations>` block MUST be FIRST (in both chat response AND file content)
+- Write full content naturally, add [Source Name](full_url) at end of sentence/paragraph
+- NEVER use "According to [Source]" format - write content first, then add citation link at end
+- Example: "AI agents will transform digital work ([Microsoft](url))" NOT "According to [Microsoft](url), AI agents will..."
 
 **Example:**
 <citations>
 {{"id": "cite-1", "title": "AI Trends 2026", "url": "https://techcrunch.com/ai-trends", "snippet": "Tech industry predictions"}}
 </citations>
-Based on [TechCrunch](https://techcrunch.com/ai-trends), the key AI trends for 2026 include...
+The key AI trends for 2026 include enhanced reasoning capabilities, multimodal integration, and improved efficiency [TechCrunch](https://techcrunch.com/ai-trends).
 </citations_format>
 
 <critical_reminders>

backend/src/agents/middlewares/uploads_middleware.py

@@ -1,6 +1,7 @@
 """Middleware to inject uploaded files information into agent context."""
 
 import os
+import re
 from pathlib import Path
 from typing import NotRequired, override
 
@@ -47,14 +48,15 @@ class UploadsMiddleware(AgentMiddleware[UploadsMiddlewareState]):
         """
         return Path(self._base_dir) / THREAD_DATA_BASE_DIR / thread_id / "user-data" / "uploads"
 
-    def _list_uploaded_files(self, thread_id: str) -> list[dict]:
-        """List all files in the uploads directory.
+    def _list_newly_uploaded_files(self, thread_id: str, last_message_files: set[str]) -> list[dict]:
+        """List only newly uploaded files that weren't in the last message.
 
         Args:
             thread_id: The thread ID.
+            last_message_files: Set of filenames that were already shown in previous messages.
 
         Returns:
-            List of file information dictionaries.
+            List of new file information dictionaries.
         """
         uploads_dir = self._get_uploads_dir(thread_id)
 
@@ -63,7 +65,7 @@ class UploadsMiddleware(AgentMiddleware[UploadsMiddlewareState]):
 
         files = []
         for file_path in sorted(uploads_dir.iterdir()):
-            if file_path.is_file():
+            if file_path.is_file() and file_path.name not in last_message_files:
                 stat = file_path.stat()
                 files.append(
                     {
@@ -106,10 +108,41 @@ class UploadsMiddleware(AgentMiddleware[UploadsMiddlewareState]):
 
         return "\n".join(lines)
 
+    def _extract_files_from_message(self, content: str) -> set[str]:
+        """Extract filenames from uploaded_files tag in message content.
+
+        Args:
+            content: Message content that may contain <uploaded_files> tag.
+
+        Returns:
+            Set of filenames mentioned in the tag.
+        """
+        # Match <uploaded_files>...</uploaded_files> tag
+        match = re.search(r"<uploaded_files>([\s\S]*?)</uploaded_files>", content)
+        if not match:
+            return set()
+
+        files_content = match.group(1)
+
+        # Extract filenames from lines like "- filename.ext (size)"
+        # Need to capture everything before the opening parenthesis, including spaces
+        filenames = set()
+        for line in files_content.split("\n"):
+            # Match pattern: - filename with spaces.ext (size)
+            # Changed from [^\s(]+ to [^(]+ to allow spaces in filename
+            file_match = re.match(r"^-\s+(.+?)\s*\(", line.strip())
+            if file_match:
+                filenames.add(file_match.group(1).strip())
+
+        return filenames
+
     @override
     def before_agent(self, state: UploadsMiddlewareState, runtime: Runtime) -> dict | None:
         """Inject uploaded files information before agent execution.
 
+        Only injects files that weren't already shown in previous messages.
+        Prepends file info to the last human message content.
+
         Args:
             state: Current agent state.
             runtime: Runtime context containing thread_id.
@@ -117,26 +150,56 @@ class UploadsMiddleware(AgentMiddleware[UploadsMiddlewareState]):
         Returns:
             State updates including uploaded files list.
         """
+        import logging
+        logger = logging.getLogger(__name__)
+        
         thread_id = runtime.context.get("thread_id")
         if thread_id is None:
             return None
 
-        # List uploaded files
-        files = self._list_uploaded_files(thread_id)
+        messages = list(state.get("messages", []))
+        if not messages:
+            return None
+
+        # Track all filenames that have been shown in previous messages (EXCEPT the last one)
+        shown_files: set[str] = set()
+        for msg in messages[:-1]:  # Scan all messages except the last one
+            if isinstance(msg, HumanMessage):
+                content = msg.content if isinstance(msg.content, str) else ""
+                extracted = self._extract_files_from_message(content)
+                shown_files.update(extracted)
+                if extracted:
+                    logger.info(f"Found previously shown files: {extracted}")
+
+        logger.info(f"Total shown files from history: {shown_files}")
+        
+        # List only newly uploaded files
+        files = self._list_newly_uploaded_files(thread_id, shown_files)
+        logger.info(f"Newly uploaded files to inject: {[f['filename'] for f in files]}")
 
         if not files:
             return None
 
-        # Create system message with file list
+        # Find the last human message and prepend file info to it
+        last_message_index = len(messages) - 1
+        last_message = messages[last_message_index]
+
+        if not isinstance(last_message, HumanMessage):
+            return None
+
+        # Create files message and prepend to the last human message content
         files_message = self._create_files_message(files)
-        files_human_message = HumanMessage(content=files_message)
+        original_content = last_message.content if isinstance(last_message.content, str) else ""
+        
+        # Create new message with combined content
+        updated_message = HumanMessage(
+            content=f"{files_message}\n\n{original_content}",
+            id=last_message.id,
+            additional_kwargs=last_message.additional_kwargs,
+        )
 
-        # Inject the message into the message history
-        # This will be added before user messages
-        messages = list(state.get("messages", []))
-
-        insert_index = 0
-        messages.insert(insert_index, files_human_message)
+        # Replace the last message
+        messages[last_message_index] = updated_message
 
         return {
             "uploaded_files": files,

backend/src/gateway/routers/artifacts.py

@@ -1,6 +1,7 @@
 import mimetypes
 import os
 from pathlib import Path
+from urllib.parse import quote
 
 from fastapi import APIRouter, HTTPException, Request, Response
 from fastapi.responses import FileResponse, HTMLResponse, PlainTextResponse
@@ -104,9 +105,12 @@ async def get_artifact(thread_id: str, path: str, request: Request) -> FileRespo
 
     mime_type, _ = mimetypes.guess_type(actual_path)
 
+    # Encode filename for Content-Disposition header (RFC 5987)
+    encoded_filename = quote(actual_path.name)
+    
     # if `download` query parameter is true, return the file as a download
     if request.query_params.get("download"):
-        return FileResponse(path=actual_path, filename=actual_path.name, media_type=mime_type, headers={"Content-Disposition": f'attachment; filename="{actual_path.name}"'})
+        return FileResponse(path=actual_path, filename=actual_path.name, media_type=mime_type, headers={"Content-Disposition": f"attachment; filename*=UTF-8''{encoded_filename}"})
 
     if mime_type and mime_type == "text/html":
         return HTMLResponse(content=actual_path.read_text())
@@ -117,4 +121,4 @@ async def get_artifact(thread_id: str, path: str, request: Request) -> FileRespo
     if is_text_file_by_content(actual_path):
         return PlainTextResponse(content=actual_path.read_text(), media_type=mime_type)
 
-    return Response(content=actual_path.read_bytes(), media_type=mime_type, headers={"Content-Disposition": f'inline; filename="{actual_path.name}"'})
+    return Response(content=actual_path.read_bytes(), media_type=mime_type, headers={"Content-Disposition": f"inline; filename*=UTF-8''{encoded_filename}"})