@musashishao/agent-kit 1.3.0 → 1.4.1

package/.agent/scripts/ak_cli.py

@@ -179,17 +179,25 @@ def cmd_init(args: argparse.Namespace) -> int:
     print(f"\n[{steps_completed + 1}/{total_steps}] Generating dependency graph...")
     graph_script = kit_path / "skills" / "graph-mapper" / "scripts" / "generate_graph.py"
     
-    src_dir = project_root / "src"
-    if not src_dir.exists():
-        src_dir = project_root / "app"
+    # Smart source detection
+    src_dir = None
+    for folder in ["src", "app", "lib", "scripts", "components", "docs"]:
+        if (project_root / folder).exists():
+            src_dir = project_root / folder
+            break
+    
+    if src_dir is None:
+        # Fallback to project root
+        src_dir = project_root
     
-    if src_dir.exists() and graph_script.exists():
+    if graph_script.exists():
         result = subprocess.run(
             [
                 "python3", str(graph_script),
                 "--src", str(src_dir),
                 "--output", str(agent_dir / "graph.json"),
                 "--format", "both",
+                "--lang", "universal",
             ],
             capture_output=True,
             text=True,
@@ -199,7 +207,7 @@ def cmd_init(args: argparse.Namespace) -> int:
         else:
             print("  ⚠️ No source code found or script failed")
     else:
-        print("  ⚠️ No src/ or app/ directory found, skipping")
+        print("  ⚠️ Graph script not found, skipping")
     steps_completed += 1
     
     # Step 5: Configure AI hosts
@@ -258,6 +266,29 @@ def cmd_init(args: argparse.Namespace) -> int:
 # Command: sync
 # ============================================================================
 
+def find_source_dir(project_root: Path) -> Path:
+    """Intelligently find the source directory."""
+    # Priority folders
+    for folder in ["src", "app", "lib", "scripts", "components"]:
+        if (project_root / folder).exists():
+            return project_root / folder
+    
+    # Check if there are source files in root
+    source_extensions = {".ts", ".tsx", ".js", ".jsx", ".py", ".go", ".c", ".cpp", ".cs", ".md"}
+    has_source_files = False
+    for item in project_root.iterdir():
+        if item.is_file() and item.suffix in source_extensions:
+            has_source_files = True
+            break
+    
+    # Check for docs folder
+    if (project_root / "docs").exists():
+        return project_root / "docs"
+    
+    # Fallback: use project root if it has any content, otherwise return root anyway
+    return project_root
+
+
 def cmd_sync(args: argparse.Namespace) -> int:
     """Sync AI infrastructure data."""
     project_root = Path(args.project_root).resolve()
@@ -271,6 +302,10 @@ def cmd_sync(args: argparse.Namespace) -> int:
         print("❌ .agent directory not found. Run 'ak init' first.")
         return 1
     
+    # Determine source directory
+    src_dir = find_source_dir(project_root)
+    print(f"🔍 Detected source directory: {src_dir.relative_to(project_root) if src_dir != project_root else '.'}")
+
     # Determine what to sync
     targets = []
     if args.target == "all":
@@ -285,17 +320,19 @@ def cmd_sync(args: argparse.Namespace) -> int:
         print("\n📊 Updating dependency graph...")
         graph_script = kit_path / "skills" / "graph-mapper" / "scripts" / "generate_graph.py"
         
-        src_dir = project_root / "src"
-        if not src_dir.exists():
-            src_dir = project_root / "app"
-        
-        if src_dir.exists() and graph_script.exists():
+        if not graph_script.exists():
+            print(f"  ❌ Graph script not found at: {graph_script}")
+            success = False
+        elif not src_dir.exists():
+            print(f"  ⚠️ Source directory {src_dir} not found")
+        else:
             result = subprocess.run(
                 [
                     "python3", str(graph_script),
                     "--src", str(src_dir),
                     "--output", str(agent_dir / "graph.json"),
                     "--format", "both",
+                    "--lang", "universal",
                 ],
                 capture_output=True,
                 text=True,
@@ -303,26 +340,26 @@ def cmd_sync(args: argparse.Namespace) -> int:
             if result.returncode == 0:
                 print("  ✅ Graph updated")
             else:
-                print(f"  ❌ Graph sync failed: {result.stderr[:200]}")
+                print(f"  ❌ Graph sync failed: {result.stderr}")
                 success = False
-        else:
-            print("  ⚠️ Source directory or script not found")
     
     # Sync RAG
     if "rag" in targets:
         print("\n📚 Updating RAG chunks...")
         rag_script = kit_path / "skills" / "rag-engineering" / "scripts" / "chunk_code.py"
         
-        src_dir = project_root / "src"
-        if not src_dir.exists():
-            src_dir = project_root / "app"
-        
-        if src_dir.exists() and rag_script.exists():
+        if not rag_script.exists():
+            print(f"  ❌ RAG script not found at: {rag_script}")
+            success = False
+        elif not src_dir.exists():
+            print(f"  ⚠️ Source directory {src_dir} not found")
+        else:
             result = subprocess.run(
                 [
                     "python3", str(rag_script),
                     "--src", str(src_dir),
                     "--output", str(agent_dir / "rag" / "chunks.json"),
+                    "--lang", "universal",
                 ],
                 capture_output=True,
                 text=True,
@@ -330,10 +367,8 @@ def cmd_sync(args: argparse.Namespace) -> int:
             if result.returncode == 0:
                 print("  ✅ RAG chunks updated")
             else:
-                print(f"  ❌ RAG sync failed: {result.stderr[:200]}")
+                print(f"  ❌ RAG sync failed: {result.stderr}")
                 success = False
-        else:
-            print("  ⚠️ Source directory or script not found")
     
     # Update timestamp cache
     cache_file = agent_dir / ".cache" / "timestamps.json"

package/.agent/skills/graph-mapper/scripts/generate_graph.py

@@ -311,6 +311,165 @@ class PythonAnalyzer:
         return nodes
 
 
+class MarkdownAnalyzer:
+    """Analyzes Markdown files for internal links and references."""
+    
+    EXTENSIONS = {'.md', '.mdx', '.markdown'}
+    
+    # Regex patterns for link detection
+    LINK_PATTERNS = [
+        # Standard markdown link: [text](path)
+        r'\[([^\]]+)\]\(([^)]+)\)',
+        # Wikilink: [[path]] or [[path|text]]
+        r'\[\[([^\]|]+)(?:\|[^\]]+)?\]\]',
+    ]
+    
+    # Pattern for image refs (also a form of dependency)
+    IMAGE_PATTERN = r'!\[([^\]]*)\]\(([^)]+)\)'
+    
+    def __init__(self, base_path: Path, exclude_patterns: List[str]):
+        self.base_path = base_path
+        self.exclude_patterns = exclude_patterns
+    
+    def should_exclude(self, path: Path) -> bool:
+        """Check if path should be excluded."""
+        path_str = str(path)
+        for pattern in self.exclude_patterns:
+            if pattern in path_str:
+                return True
+        return False
+    
+    def analyze_file(self, file_path: Path) -> Optional[Node]:
+        """Analyze a single Markdown file for links."""
+        if self.should_exclude(file_path):
+            return None
+            
+        try:
+            content = file_path.read_text(encoding='utf-8')
+        except (UnicodeDecodeError, PermissionError):
+            return None
+        
+        relative_path = str(file_path.relative_to(self.base_path))
+        node_type = self._detect_type(relative_path, content)
+        
+        # Extract internal links
+        links = []
+        
+        # Standard markdown links
+        for match in re.findall(self.LINK_PATTERNS[0], content):
+            link_path = match[1]
+            resolved = self._resolve_link(link_path, file_path)
+            if resolved:
+                links.append(resolved)
+        
+        # Wikilinks
+        for match in re.findall(self.LINK_PATTERNS[1], content):
+            resolved = self._resolve_wikilink(match)
+            if resolved:
+                links.append(resolved)
+        
+        # Extract "exports" (main topics/headings)
+        exports = []
+        # Get main title (first h1)
+        title_match = re.search(r'^#\s+(.+)$', content, re.MULTILINE)
+        if title_match:
+            exports.append(title_match.group(1).strip())
+        
+        # Get h2 sections as additional exports
+        for match in re.findall(r'^##\s+(.+)$', content, re.MULTILINE):
+            exports.append(match.strip())
+        
+        return Node(
+            id=relative_path,
+            type=node_type,
+            path=relative_path,
+            imports=list(set(links)),  # Links = imports in docs context
+            exports=exports[:10]  # Limit to first 10 headings
+        )
+    
+    def _detect_type(self, path: str, content: str) -> str:
+        """Detect the type of documentation file."""
+        path_lower = path.lower()
+        
+        if 'readme' in path_lower:
+            return 'readme'
+        elif 'changelog' in path_lower or 'history' in path_lower:
+            return 'changelog'
+        elif 'contributing' in path_lower:
+            return 'contributing'
+        elif 'license' in path_lower:
+            return 'license'
+        elif '/docs/' in path_lower or path_lower.startswith('docs/'):
+            return 'documentation'
+        elif '/guides/' in path_lower or '/tutorials/' in path_lower:
+            return 'guide'
+        elif '/api/' in path_lower or 'api' in path_lower:
+            return 'api_doc'
+        elif 'plan' in path_lower or 'roadmap' in path_lower:
+            return 'plan'
+        else:
+            return 'document'
+    
+    def _resolve_link(self, link_path: str, from_file: Path) -> Optional[str]:
+        """Resolve a markdown link to a relative path."""
+        # Skip external links
+        if link_path.startswith(('http://', 'https://', 'mailto:', '#')):
+            return None
+        
+        # Skip anchor-only links
+        if link_path.startswith('#'):
+            return None
+        
+        # Remove anchor from path
+        if '#' in link_path:
+            link_path = link_path.split('#')[0]
+        
+        if not link_path:
+            return None
+        
+        # Resolve relative path
+        from_dir = from_file.parent
+        resolved = (from_dir / link_path).resolve()
+        
+        try:
+            relative = str(resolved.relative_to(self.base_path))
+            # Check if file exists
+            if resolved.exists():
+                return relative
+            return None
+        except ValueError:
+            return None
+    
+    def _resolve_wikilink(self, link_name: str) -> Optional[str]:
+        """Resolve a wikilink to a file path."""
+        # Search for file matching the wikilink name
+        search_name = link_name.strip()
+        
+        # Try exact match with .md extension
+        for ext in self.EXTENSIONS:
+            for file_path in self.base_path.rglob(f"*{ext}"):
+                if file_path.stem.lower() == search_name.lower():
+                    try:
+                        return str(file_path.relative_to(self.base_path))
+                    except ValueError:
+                        continue
+        
+        return None
+    
+    def analyze_directory(self, directory: Path) -> List[Node]:
+        """Analyze all Markdown files in a directory."""
+        nodes = []
+        
+        for ext in self.EXTENSIONS:
+            for file_path in directory.rglob(f'*{ext}'):
+                if file_path.is_file():
+                    node = self.analyze_file(file_path)
+                    if node:
+                        nodes.append(node)
+        
+        return nodes
+
+
 def build_edges(nodes: List[Node]) -> List[Edge]:
     """Build edges from node imports."""
     edges = []
@@ -424,13 +583,16 @@ def generate_markdown(graph: Graph, output_path: Path):
 
 
 def main():
-    parser = argparse.ArgumentParser(description='Generate dependency graph')
+    parser = argparse.ArgumentParser(
+        description='Universal Dependency Graph Generator - Code and Documentation'
+    )
     parser.add_argument('--src', default='./src', help='Source directory')
     parser.add_argument('--output', default='.agent/graph.json', help='Output file')
     parser.add_argument('--format', choices=['json', 'markdown', 'both'], default='both')
-    parser.add_argument('--lang', choices=['typescript', 'python', 'auto'], default='auto')
+    parser.add_argument('--lang', choices=['typescript', 'python', 'auto', 'universal'], 
+                       default='universal', help='Language mode (universal = Code + Markdown)')
     parser.add_argument('--depth', type=int, default=3, help='Max depth for impact analysis')
-    parser.add_argument('--exclude', default='node_modules,__pycache__,.git,dist,build',
+    parser.add_argument('--exclude', default='node_modules,__pycache__,.git,dist,build,.agent',
                        help='Comma-separated patterns to exclude')
     
     args = parser.parse_args()
@@ -443,37 +605,77 @@ def main():
         print(f"Error: Source directory '{src_path}' does not exist")
         return 1
     
-    # Detect language if auto
+    # Collect all nodes
+    all_nodes = []
     lang = args.lang
-    if lang == 'auto':
+    
+    if lang == 'universal':
+        print("Universal mode: Analyzing Code + Markdown files")
+        
+        # Analyze TypeScript/JavaScript
+        ts_analyzer = TypeScriptAnalyzer(src_path, exclude_patterns)
+        ts_nodes = ts_analyzer.analyze_directory(src_path)
+        all_nodes.extend(ts_nodes)
+        print(f"  TypeScript/JS: {len(ts_nodes)} files")
+        
+        # Analyze Python
+        py_analyzer = PythonAnalyzer(src_path, exclude_patterns)
+        py_nodes = py_analyzer.analyze_directory(src_path)
+        all_nodes.extend(py_nodes)
+        print(f"  Python: {len(py_nodes)} files")
+        
+        # Analyze Markdown
+        md_analyzer = MarkdownAnalyzer(src_path, exclude_patterns)
+        md_nodes = md_analyzer.analyze_directory(src_path)
+        all_nodes.extend(md_nodes)
+        print(f"  Markdown: {len(md_nodes)} files")
+        
+    elif lang == 'auto':
         ts_files = list(src_path.rglob('*.ts')) + list(src_path.rglob('*.tsx'))
         py_files = list(src_path.rglob('*.py'))
         lang = 'typescript' if len(ts_files) >= len(py_files) else 'python'
         print(f"Auto-detected language: {lang}")
-    
-    # Analyze based on language
-    if lang == 'typescript':
-        analyzer = TypeScriptAnalyzer(src_path, exclude_patterns)
+        
+        if lang == 'typescript':
+            analyzer = TypeScriptAnalyzer(src_path, exclude_patterns)
+        else:
+            analyzer = PythonAnalyzer(src_path, exclude_patterns)
+        
+        all_nodes = analyzer.analyze_directory(src_path)
     else:
-        analyzer = PythonAnalyzer(src_path, exclude_patterns)
+        # Specific language
+        if lang == 'typescript':
+            analyzer = TypeScriptAnalyzer(src_path, exclude_patterns)
+        else:
+            analyzer = PythonAnalyzer(src_path, exclude_patterns)
+        
+        all_nodes = analyzer.analyze_directory(src_path)
     
-    print(f"Analyzing {src_path}...")
-    nodes = analyzer.analyze_directory(src_path)
-    print(f"Found {len(nodes)} files")
+    print(f"Total: {len(all_nodes)} files")
+    
+    if len(all_nodes) == 0:
+        print("Warning: No files found. Check source directory and exclude patterns.")
+        print("Supported: .ts, .tsx, .js, .jsx, .py, .md, .mdx")
     
     # Build graph
-    edges = build_edges(nodes)
-    print(f"Found {len(edges)} dependencies")
+    edges = build_edges(all_nodes)
+    print(f"Found {len(edges)} dependencies/links")
+    
+    # Categorize by type
+    type_counts = {}
+    for node in all_nodes:
+        type_counts[node.type] = type_counts.get(node.type, 0) + 1
     
     graph = Graph(
-        nodes=nodes,
+        nodes=all_nodes,
         edges=edges,
         metadata={
             "generated_at": datetime.now().isoformat(),
             "source_path": str(src_path),
-            "language": lang,
-            "total_files": len(nodes),
-            "total_edges": len(edges)
+            "mode": lang,
+            "total_files": len(all_nodes),
+            "total_edges": len(edges),
+            "file_types": type_counts
         }
     )
     

package/.agent/skills/rag-engineering/scripts/chunk_code.py

@@ -505,29 +505,320 @@ class PythonChunker:
         )
 
 
+class MarkdownChunker:
+    """Chunk Markdown files by heading sections."""
+    
+    EXTENSIONS = {'.md', '.mdx', '.markdown'}
+    
+    def __init__(self, max_chunk_size: int = 2000, overlap: int = 100):
+        self.max_chunk_size = max_chunk_size
+        self.overlap = overlap
+    
+    def chunk_file(self, file_path: Path, base_path: Path) -> List[Chunk]:
+        """Chunk a Markdown file by headings."""
+        try:
+            content = file_path.read_text(encoding='utf-8')
+        except (UnicodeDecodeError, PermissionError):
+            return []
+        
+        relative_path = str(file_path.relative_to(base_path))
+        chunks = []
+        
+        # Extract sections by heading
+        sections = self._extract_sections(content)
+        
+        if sections:
+            for section in sections:
+                chunk_content = section['content']
+                
+                # If section is too large, split it
+                if len(chunk_content) > self.max_chunk_size:
+                    sub_chunks = self._split_by_paragraphs(chunk_content)
+                    for i, sub in enumerate(sub_chunks):
+                        chunks.append(self._create_chunk(
+                            content=sub,
+                            file_path=relative_path,
+                            section_title=f"{section['title']}_part{i+1}",
+                            heading_level=section['level'],
+                            parent_headings=section['parents'],
+                            start_line=section['start_line'],
+                            end_line=section['end_line']
+                        ))
+                else:
+                    chunks.append(self._create_chunk(
+                        content=chunk_content,
+                        file_path=relative_path,
+                        section_title=section['title'],
+                        heading_level=section['level'],
+                        parent_headings=section['parents'],
+                        start_line=section['start_line'],
+                        end_line=section['end_line']
+                    ))
+        else:
+            # No headings found, treat entire file as one chunk or split by paragraphs
+            if len(content) <= self.max_chunk_size:
+                chunks.append(self._create_chunk(
+                    content=content,
+                    file_path=relative_path,
+                    section_title=file_path.stem,
+                    heading_level=0,
+                    parent_headings=[],
+                    start_line=1,
+                    end_line=content.count('\n') + 1
+                ))
+            else:
+                sub_chunks = self._split_by_paragraphs(content)
+                for i, sub in enumerate(sub_chunks):
+                    chunks.append(self._create_chunk(
+                        content=sub,
+                        file_path=relative_path,
+                        section_title=f"{file_path.stem}_part{i+1}",
+                        heading_level=0,
+                        parent_headings=[],
+                        start_line=1,
+                        end_line=content.count('\n') + 1
+                    ))
+        
+        return chunks
+    
+    def _extract_sections(self, content: str) -> List[Dict]:
+        """Extract sections based on Markdown headings."""
+        sections = []
+        lines = content.split('\n')
+        
+        heading_pattern = re.compile(r'^(#{1,6})\s+(.+)$')
+        
+        current_section = None
+        current_lines = []
+        parent_stack = []  # Track parent headings for context
+        
+        for i, line in enumerate(lines):
+            match = heading_pattern.match(line)
+            
+            if match:
+                # Save previous section
+                if current_section:
+                    current_section['content'] = '\n'.join(current_lines).strip()
+                    current_section['end_line'] = i
+                    if current_section['content']:
+                        sections.append(current_section)
+                
+                # Update parent stack
+                level = len(match.group(1))
+                title = match.group(2).strip()
+                
+                # Pop parents that are same or deeper level
+                while parent_stack and parent_stack[-1]['level'] >= level:
+                    parent_stack.pop()
+                
+                parents = [p['title'] for p in parent_stack]
+                
+                # Start new section
+                current_section = {
+                    'title': title,
+                    'level': level,
+                    'parents': parents.copy(),
+                    'start_line': i + 1,
+                    'end_line': i + 1,
+                    'content': ''
+                }
+                current_lines = [line]
+                
+                # Add this heading to parent stack
+                parent_stack.append({'level': level, 'title': title})
+            elif current_section:
+                current_lines.append(line)
+        
+        # Don't forget last section
+        if current_section:
+            current_section['content'] = '\n'.join(current_lines).strip()
+            current_section['end_line'] = len(lines)
+            if current_section['content']:
+                sections.append(current_section)
+        
+        return sections
+    
+    def _split_by_paragraphs(self, content: str) -> List[str]:
+        """Split content by paragraphs when too large."""
+        chunks = []
+        paragraphs = re.split(r'\n\s*\n', content)
+        
+        current_chunk = []
+        current_size = 0
+        
+        for para in paragraphs:
+            para_size = len(para) + 2  # +2 for paragraph break
+            
+            if current_size + para_size > self.max_chunk_size and current_chunk:
+                chunks.append('\n\n'.join(current_chunk))
+                current_chunk = []
+                current_size = 0
+            
+            current_chunk.append(para)
+            current_size += para_size
+        
+        if current_chunk:
+            chunks.append('\n\n'.join(current_chunk))
+        
+        return chunks
+    
+    def _create_chunk(
+        self,
+        content: str,
+        file_path: str,
+        section_title: str,
+        heading_level: int,
+        parent_headings: List[str],
+        start_line: int,
+        end_line: int
+    ) -> Chunk:
+        """Create a Chunk object with rich context."""
+        # Build context string for better retrieval
+        context_path = ' > '.join(parent_headings + [section_title]) if parent_headings else section_title
+        
+        return Chunk(
+            id=generate_chunk_id(file_path, content),
+            content=content,
+            metadata={
+                'file_path': file_path,
+                'file_type': 'markdown',
+                'chunk_type': f'heading_{heading_level}' if heading_level > 0 else 'paragraph',
+                'name': section_title,
+                'context_path': context_path,
+                'heading_level': heading_level,
+                'parent_headings': parent_headings,
+                'start_line': start_line,
+                'end_line': end_line,
+                'char_count': len(content),
+                'line_count': content.count('\n') + 1
+            }
+        )
+
+
+class TextChunker:
+    """Chunk plain text files by paragraphs."""
+    
+    EXTENSIONS = {'.txt', '.text', '.log'}
+    
+    def __init__(self, max_chunk_size: int = 1500, overlap: int = 100):
+        self.max_chunk_size = max_chunk_size
+        self.overlap = overlap
+    
+    def chunk_file(self, file_path: Path, base_path: Path) -> List[Chunk]:
+        """Chunk a text file by paragraphs."""
+        try:
+            content = file_path.read_text(encoding='utf-8')
+        except (UnicodeDecodeError, PermissionError):
+            return []
+        
+        relative_path = str(file_path.relative_to(base_path))
+        chunks = []
+        
+        if len(content) <= self.max_chunk_size:
+            chunks.append(self._create_chunk(
+                content=content,
+                file_path=relative_path,
+                name=file_path.stem,
+                start_line=1,
+                end_line=content.count('\n') + 1
+            ))
+        else:
+            # Split by paragraphs
+            paragraphs = re.split(r'\n\s*\n', content)
+            current_chunk = []
+            current_size = 0
+            chunk_index = 0
+            
+            for para in paragraphs:
+                para_size = len(para) + 2
+                
+                if current_size + para_size > self.max_chunk_size and current_chunk:
+                    chunk_index += 1
+                    chunks.append(self._create_chunk(
+                        content='\n\n'.join(current_chunk),
+                        file_path=relative_path,
+                        name=f"{file_path.stem}_part{chunk_index}",
+                        start_line=1,
+                        end_line=content.count('\n') + 1
+                    ))
+                    current_chunk = []
+                    current_size = 0
+                
+                current_chunk.append(para)
+                current_size += para_size
+            
+            if current_chunk:
+                chunk_index += 1
+                chunks.append(self._create_chunk(
+                    content='\n\n'.join(current_chunk),
+                    file_path=relative_path,
+                    name=f"{file_path.stem}_part{chunk_index}",
+                    start_line=1,
+                    end_line=content.count('\n') + 1
+                ))
+        
+        return chunks
+    
+    def _create_chunk(
+        self,
+        content: str,
+        file_path: str,
+        name: str,
+        start_line: int,
+        end_line: int
+    ) -> Chunk:
+        """Create a Chunk object."""
+        return Chunk(
+            id=generate_chunk_id(file_path, content),
+            content=content,
+            metadata={
+                'file_path': file_path,
+                'file_type': 'text',
+                'chunk_type': 'paragraph',
+                'name': name,
+                'start_line': start_line,
+                'end_line': end_line,
+                'char_count': len(content),
+                'line_count': content.count('\n') + 1
+            }
+        )
+
+
 def chunk_directory(
     src_path: Path,
     lang: str,
     exclude_patterns: List[str],
     max_chunk_size: int = 1500
 ) -> List[Chunk]:
-    """Chunk all files in a directory."""
+    """Chunk all files in a directory - Universal support for Code, Docs, and Text."""
     all_chunks = []
     
-    # Select chunker
-    if lang == 'typescript':
-        chunker = TypeScriptChunker(max_chunk_size=max_chunk_size)
-        extensions = TypeScriptChunker.EXTENSIONS
-    else:
-        chunker = PythonChunker(max_chunk_size=max_chunk_size)
-        extensions = {'.py'}
+    # Initialize all chunkers
+    ts_chunker = TypeScriptChunker(max_chunk_size=max_chunk_size)
+    py_chunker = PythonChunker(max_chunk_size=max_chunk_size)
+    md_chunker = MarkdownChunker(max_chunk_size=max_chunk_size)
+    txt_chunker = TextChunker(max_chunk_size=max_chunk_size)
+    
+    # Map extensions to chunkers
+    extension_map = {}
+    for ext in TypeScriptChunker.EXTENSIONS:
+        extension_map[ext] = ts_chunker
+    extension_map['.py'] = py_chunker
+    for ext in MarkdownChunker.EXTENSIONS:
+        extension_map[ext] = md_chunker
+    for ext in TextChunker.EXTENSIONS:
+        extension_map[ext] = txt_chunker
+    
+    # All supported extensions
+    all_extensions = set(extension_map.keys())
     
     # Process files
     for file_path in src_path.rglob('*'):
         if not file_path.is_file():
             continue
         
-        if file_path.suffix not in extensions:
+        # Check if extension is supported
+        if file_path.suffix not in all_extensions:
             continue
         
         # Check exclusions
@@ -535,19 +826,25 @@ def chunk_directory(
         if any(pattern in path_str for pattern in exclude_patterns):
             continue
         
-        chunks = chunker.chunk_file(file_path, src_path)
-        all_chunks.extend(chunks)
+        # Select appropriate chunker
+        chunker = extension_map.get(file_path.suffix)
+        if chunker:
+            chunks = chunker.chunk_file(file_path, src_path)
+            all_chunks.extend(chunks)
     
     return all_chunks
 
 
 def main():
-    parser = argparse.ArgumentParser(description='Chunk code files')
+    parser = argparse.ArgumentParser(
+        description='Universal Chunker - Code, Markdown, and Text files'
+    )
     parser.add_argument('--src', default='./src', help='Source directory')
     parser.add_argument('--output', default='.agent/rag/chunks.json', help='Output file')
-    parser.add_argument('--lang', choices=['typescript', 'python', 'auto'], default='auto')
+    parser.add_argument('--lang', choices=['typescript', 'python', 'auto', 'universal'], 
+                       default='universal', help='Language mode (universal = all file types)')
     parser.add_argument('--max-size', type=int, default=1500, help='Max chunk size in chars')
-    parser.add_argument('--exclude', default='node_modules,__pycache__,.git,dist,build',
+    parser.add_argument('--exclude', default='node_modules,__pycache__,.git,dist,build,.agent',
                        help='Patterns to exclude')
     
     args = parser.parse_args()
@@ -560,28 +857,41 @@ def main():
         print(f"Error: Source directory '{src_path}' does not exist")
         return 1
     
-    # Auto-detect language
+    # Mode selection
     lang = args.lang
     if lang == 'auto':
         ts_files = list(src_path.rglob('*.ts')) + list(src_path.rglob('*.tsx'))
         py_files = list(src_path.rglob('*.py'))
         lang = 'typescript' if len(ts_files) >= len(py_files) else 'python'
         print(f"Auto-detected language: {lang}")
+    elif lang == 'universal':
+        print("Universal mode: Processing Code, Markdown, and Text files")
     
     print(f"Chunking {src_path}...")
     chunks = chunk_directory(src_path, lang, exclude_patterns, args.max_size)
     print(f"Created {len(chunks)} chunks")
     
+    if len(chunks) == 0:
+        print("Warning: No chunks created. Check if source directory has supported files.")
+        print("Supported: .ts, .tsx, .js, .jsx, .py, .md, .mdx, .txt")
+    
     # Save output
     output_path.parent.mkdir(parents=True, exist_ok=True)
     
+    # Categorize chunks by file type
+    file_types = {}
+    for c in chunks:
+        ft = c.metadata.get('file_type', 'unknown')
+        file_types[ft] = file_types.get(ft, 0) + 1
+    
     output_data = {
         'metadata': {
             'generated_at': datetime.now().isoformat(),
             'source_path': str(src_path),
-            'language': lang,
+            'mode': lang,
             'total_chunks': len(chunks),
-            'max_chunk_size': args.max_size
+            'max_chunk_size': args.max_size,
+            'file_types': file_types
         },
         'chunks': [asdict(c) for c in chunks]
     }

package/docs/PLAN-universal-intelligence.md

@@ -0,0 +1,48 @@
+# PLAN: Universal Intelligence Engine (UIE)
+
+> Mục tiêu: Nâng cấp Agent Kit để hỗ trợ mọi loại folder làm việc (Docs, Research, Code) với khả năng thấu hiểu tri thức toàn diện.
+
+## Phase 1: Universal RAG (Smart Chunking cho mọi định dạng) ✅ COMPLETED
+- [x] **Nâng cấp `chunk_code.py` thành Universal Chunker**:
+    - [x] Thêm `MarkdownChunker`: Cắt segment theo Heading (`#`, `##`, `###`).
+    - [x] Thêm `TextChunker`: Cắt theo đoạn văn (Paragraph).
+    - [x] Giữ nguyên khả năng cắt Code (TS, Python) sử dụng AST.
+- [x] **Contextual Metadata cho Docs**:
+    - [x] Tự động trích xuất tiêu đề chính của file làm context cho từng chunk bên trong.
+    - [x] Ghi nhận `context_path` (Parent > Child heading) để AI biết ngữ cảnh.
+
+## Phase 2: Knowledge Graph cho Tài liệu (Relationship Mapping) ✅ COMPLETED
+- [x] **Nâng cấp `generate_graph.py`**:
+    - [x] Thêm `MarkdownAnalyzer`: Quét link nội bộ `[text](file.md)` và wikilink `[[file]]`.
+    - [x] Nhận diện heading exports (H1, H2) làm "API" của tài liệu.
+    - [x] Chế độ `universal`: Kết hợp Code + Markdown trong cùng một graph.
+
+## Phase 3: MCP Gateway Enhancements
+- [ ] **Content-Type Awareness**:
+    - [ ] Tool `analyze_dependencies` trả về "References" thay vì "Imports" cho docs.
+    - [ ] Tool `search_code_logic` alias thành `search_knowledge`.
+- [ ] **Auto-Detection**: gateway báo cáo loại dự án (Creative, Tech, Mixed).
+
+## Phase 4: CLI Evolution (`ak` command)
+- [ ] **`ak init` thông minh**:
+    - [ ] Không ép buộc phải có `src` hay `app`.
+    - [ ] Tự động tạo `AGENTS.md` theo template "General Knowledge" nếu không phát hiện code.
+- [ ] **`ak sync` toàn diện**:
+    - [ ] Quét mọi file (ngo trừ ignore list) để đảm bảo không bỏ sót tri thức nào.
+
+## Phase 5: Verification & Beta Test
+- [ ] **Test Case 1**: Dự án chỉ gồm 100 file Markdown (Hướng dẫn sử dụng).
+- [ ] **Test Case 2**: Dự án hỗn hợp (Next.js + Docs + API Specs).
+- [ ] **Test Case 3**: Dự án nghiên cứu (Nhiều file .txt và ghi chú rời rạc).
+
+---
+
+## Agent Assignments
+- **Python Specialist**: Xử lý logic Chunker và Graph Mapper (Phase 1 & 2).
+- **TypeScript Expert**: Cập nhật MCP Gateway (Phase 3).
+- **Orchestrator**: Cập nhật CLI và hoàn thiện Docs (Phase 4).
+
+## Verification Checklist
+- [ ] `ai status` hiển thị ✅ RAG và ✅ Graph ngay cả trong folder không có code.
+- [ ] AI có thể tìm thấy thông tin nằm sâu trong một sub-section của file Markdown dài.
+- [ ] Dependency graph hiển thị được mối liên kết giữa các file tài liệu.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@musashishao/agent-kit",
-  "version": "1.3.0",
+  "version": "1.4.1",
   "description": "AI Agent templates - Skills, Agents, Workflows, and AI-Ready Data Infrastructure Gateway",
   "main": "index.js",
   "bin": {