@@ -147,8 +147,10 @@ class CliCallTemplate(CallTemplate):
         commands: A list of CommandStep objects defining the commands to execute
             in order. Each command can contain UTCP_ARG_argname_UTCP_END placeholders
             that will be replaced with values from tool_args during execution.
-            Placeholders are shell-quoted and therefore expand to exactly one
-            shell token (see class docstring).
+            Each placeholder becomes a shell-variable reference whose value
+            reaches the subprocess through a per-invocation environment
+            variable, so it expands to exactly one shell token and cannot be
+            reinterpreted as shell syntax (see class docstring).
         env_vars: A dictionary of environment variables to set for the command's
             execution context. Values can be static strings or placeholders for
             variables from the UTCP client's variable substitutor. Always
@@ -238,9 +240,11 @@ class CliCallTemplate(CallTemplate):
     Security Considerations:
         - Commands are executed in a subprocess. Ensure that the commands
           specified are from a trusted source.
-        - `tool_args` values are shell-quoted on substitution, but the
-          *command template itself* is not — never assemble it from
-          untrusted input.
+        - `tool_args` values never touch the command text: they reach the
+          subprocess through per-invocation environment variables and the
+          shell expands them only after it has parsed the script. The
+          *command template itself* has no such protection — never assemble
+          it from untrusted input.
         - The host environment is restricted; secrets are not propagated
           unless explicitly named in `env_vars` or `inherit_env_vars`.
         - Commands should use the appropriate syntax for the target platform
 
@@ -63,7 +63,7 @@ class HttpCommunicationProtocol(CommunicationProtocol):
 
     Attributes:
         _session: Optional aiohttp ClientSession for connection reuse.
-        _oauth_tokens: Cache of OAuth2 tokens keyed by the full credential configuration (``OAuth2Auth.cache_key``).
+        _oauth_tokens: Cache of OAuth2 tokens keyed by the full credential configuration (OAuth2Auth's ``cache_key``).
         _log: Logger function for debugging and error reporting.
     """
 
 
@@ -63,7 +63,7 @@ class WebSocketCommunicationProtocol(CommunicationProtocol):
     Attributes:
         _connections: Active WebSocket connections by provider key.
         _sessions: aiohttp ClientSessions for connection management.
-        _oauth_tokens: Cache of OAuth2 tokens keyed by the full credential configuration (``OAuth2Auth.cache_key``).
+        _oauth_tokens: Cache of OAuth2 tokens keyed by the full credential configuration (OAuth2Auth's ``cache_key``).
     """
 
     def __init__(self, logger_func: Optional[Callable[[str], None]] = None):
 
@@ -148,44 +148,49 @@ def process_section_content(content_lines):
                 line = line.replace('{', '\\{').replace('}', '\\}')
                 stripped = line.strip()
 
-                # Check if this looks like a parameter/item definition (name: description)
+                # Check if this looks like a parameter/item definition (name: description).
+                # A line with a colon that is NOT one -- prose such as "(note: ...)" or a
+                # quoted URL -- is ordinary text and falls through to the branches below;
+                # it must never be dropped.
+                param_match = None
                 if ':' in stripped and not stripped.endswith(':'):
                     colon_pos = stripped.find(':')
-                    param_name = stripped[:colon_pos].strip()
-                    param_desc = stripped[colon_pos + 1:].strip()
-                    
-                    # Check if param_name looks like a parameter (no spaces, reasonable length)
-                    if ' ' not in param_name and len(param_name) <= 50 and param_name.replace('_', '').isalnum():
-                        # This is likely a parameter definition
-                        processed.append(f"- **`{param_name}`**: {param_desc}")
-                        
-                        # Check for continuation lines (indented more than the parameter line)
-                        base_indent = len(line) - len(line.lstrip())
-                        i += 1
-                        while i < len(content_lines):
-                            next_line = content_lines[i]
-                            next_stripped = next_line.strip()
-                            next_indent = len(next_line) - len(next_line.lstrip()) if next_stripped else 0
-                            
-                            # Check if we hit a code block
-                            if next_stripped.startswith('```'):
-                                break
-                            
-                            if not next_stripped:
-                                # Empty line - add it and continue
-                                processed.append('')
-                                i += 1
-                            elif next_indent > base_indent:
-                                # Continuation line - add with proper spacing
-                                processed.append(f"  {next_stripped}")
-                                i += 1
-                            else:
-                                # Not a continuation, back up and break
-                                break
-                        continue
-                
+                    candidate = stripped[:colon_pos].strip()
+                    # A parameter name has no spaces and is a plain identifier
+                    if ' ' not in candidate and len(candidate) <= 50 and candidate.replace('_', '').isalnum():
+                        param_match = (candidate, stripped[colon_pos + 1:].strip())
+
+                if param_match is not None:
+                    param_name, param_desc = param_match
+                    processed.append(f"- **`{param_name}`**: {param_desc}")
+
+                    # Check for continuation lines (indented more than the parameter line)
+                    base_indent = len(line) - len(line.lstrip())
+                    i += 1
+                    while i < len(content_lines):
+                        next_line = content_lines[i]
+                        next_stripped = next_line.strip()
+                        next_indent = len(next_line) - len(next_line.lstrip()) if next_stripped else 0
+
+                        # Check if we hit a code block
+                        if next_stripped.startswith('```'):
+                            break
+
+                        if not next_stripped:
+                            # Empty line - add it and continue
+                            processed.append('')
+                            i += 1
+                        elif next_indent > base_indent:
+                            # Continuation line - add with proper spacing
+                            processed.append(f"  {next_stripped}")
+                            i += 1
+                        else:
+                            # Not a continuation, back up and break
+                            break
+                    continue
+
                 # Check if line starts with a list marker
-                elif stripped.startswith(('- ', '* ', '+ ')):
+                if stripped.startswith(('- ', '* ', '+ ')):
                     # This is already a markdown list item
                     processed.append(stripped)
                 elif stripped.startswith(('1. ', '2. ', '3. ', '4. ', '5. ', '6. ', '7. ', '8. ', '9. ')):
@@ -204,9 +209,13 @@ def process_section_content(content_lines):
         # Parse the docstring line by line
         for line in lines:
             stripped_lower = line.strip().lower()
-            
-            # Check if this line is a section header
-            if stripped_lower in section_headers or stripped_lower.endswith(':'):
+
+            # Check if this line is a section header: a known Google-style header, or a
+            # short title made of words only ("Security Considerations:"). A sentence
+            # that merely ends in a colon ("Inheritance is controlled by `x`:") is
+            # content -- treating it as a header would title-case it, code span included.
+            is_custom_header = re.fullmatch(r'[A-Za-z0-9][A-Za-z0-9 \-]{0,40}:', line.strip()) is not None
+            if stripped_lower in section_headers or is_custom_header:
                 # Save previous section if it exists
                 if current_section:
                     processed_content = process_section_content(current_section_content)
@@ -231,6 +240,10 @@ def process_section_content(content_lines):
             if processed_content:
                 result.append(f"\n**{current_section.title()}**\n")
                 result.extend(processed_content)
+        else:
+            # No section header anywhere: the whole docstring is the preamble,
+            # which is otherwise only flushed when a header follows it.
+            result.extend(process_section_content(current_section_content))
 
         # Clean up the result
         final_result = []
@@ -530,18 +543,46 @@ def add_cross_references_post_generation(self, text: str, current_output_file: s
                         class_anchor = re.sub(r'[^\w\-_]', '-', class_name.lower()).strip('-')
                         link = f"[{class_name}](./{relative_path_str}#{class_anchor})"
 
-                # Don't replace matches that are in code blocks
+                # Don't replace matches that are in code blocks or inline code spans
                 lines = modified_text.split('\n')
                 in_code_block = False
                 for i, line in enumerate(lines):
                     if line.strip().startswith('```'):
                         in_code_block = not in_code_block
                     elif not in_code_block:
-                        lines[i] = re.sub(pattern, link, line)
+                        lines[i] = self._sub_outside_inline_code(pattern, link, line)
                 modified_text = '\n'.join(lines)
-        
+
         return modified_text
+
+    @staticmethod
+    def _sub_outside_inline_code(pattern: str, replacement: str, line: str) -> str:
+        """Substitute only outside `...` / ``...`` spans.
+
+        Markdown renders a code span literally, so a link inserted inside one shows up
+        as raw brackets instead of a link.
+        """
+        parts = re.split(r'(`+[^`]*`+)', line)
+        return ''.join(part if part.startswith('`') else re.sub(pattern, replacement, part) for part in parts)
 
+    def _render_methods(self, content: List[str], methods: List[DocEntry], file_path: str) -> None:
+        """Append a class's documented methods to ``content``."""
+        content.extend(["#### Methods:", ""])
+        for method in methods:
+            # Add cross-references to method signature
+            linked_signature = self.add_cross_references(method.signature, file_path)
+            docstrings = method.docstring if method.docstring else "*No method documentation available*"
+            content.extend(
+                [
+                    "<details>",
+                    f"<summary>{linked_signature}</summary>",
+                    "",
+                    docstrings,
+                    "</details>",
+                    "",
+                ]
+            )
+
     def generate_module_markdown(self, file_path: str, file_data: Dict[str, List[DocEntry]]) -> str:
         """Generate markdown content for a single module/file."""
         if not any(file_data.values()):
@@ -632,34 +673,27 @@ def generate_module_markdown(self, file_path: str, file_data: Dict[str, List[Doc
 
                 # Add methods for this class
                 if class_entry.name in methods_by_class:
-                    content.extend(["#### Methods:", ""])
-                    
-                    for method in methods_by_class[class_entry.name]:
-                        method_anchor = re.sub(r'[^\w\-_]', '-', f"{class_entry.name}-{method.name}".lower()).strip('-')
-                        
-                        # Add cross-references to method signature
-                        linked_signature = self.add_cross_references(method.signature, file_path)
-                        
-                        docstrings = ""
-                        
-                        if method.docstring:
-                            docstrings = method.docstring
-                        else:
-                            docstrings = "*No method documentation available*"
-                            
-                        content.extend(
-                            [
-                                "<details>",
-                                f"<summary>{linked_signature}</summary>",
-                                "",
-                                docstrings,
-                                "</details>",
-                                "",
-                            ]
-                        )
-                
+                    self._render_methods(content, methods_by_class[class_entry.name], file_path)
+
                 content.extend(["---", ""])
-        
+
+        # A class whose own docstring is not REQUIRED can still have REQUIRED
+        # methods. They are required documentation and the index counts them,
+        # so they are rendered under a bare class heading rather than lost.
+        documented_classes = {class_entry.name for class_entry in file_data['classes']}
+        for class_name, methods in methods_by_class.items():
+            if class_name in documented_classes:
+                continue
+            class_anchor = re.sub(r'[^\w\-_]', '-', class_name.lower()).strip('-')
+            content.extend([
+                f"### class {class_name} {{#{class_anchor}}}",
+                "",
+                "*No class documentation available*",
+                "",
+            ])
+            self._render_methods(content, methods, file_path)
+            content.extend(["---", ""])
+
         # Add standalone functions
         if file_data['functions']:
             for func_entry in file_data['functions']:
@@ -748,7 +782,7 @@ def generate_index_file(self, modules: Dict[str, Dict[str, List[DocEntry]]], out
                     index_path = output_path / "index.md"
                     target_path = Path(output_file_path)
                     try:
-                        relative_path = target_path.relative_to(output_path)
+                        relative_path = target_path.relative_to(output_path).as_posix()
                         link_path = f"./{relative_path}"
                     except ValueError:
                         # Fallback to simple filename if relative path calculation fails
@@ -805,7 +839,7 @@ def generate_index_file(self, modules: Dict[str, Dict[str, List[DocEntry]]], out
                     index_path = output_path / "index.md"
                     target_path = Path(output_file_path)
                     try:
-                        relative_path = target_path.relative_to(output_path)
+                        relative_path = target_path.relative_to(output_path).as_posix()
                         link_path = f"./{relative_path}"
                     except ValueError:
                         # Fallback to simple filename if relative path calculation fails
@@ -900,10 +934,11 @@ def generate_docs(self, output_dir: str) -> None:
 
         # Second pass: Add cross-references and write files
         for file_path, (content, output_file_path) in generated_files.items():
-            # Post-process content to add proper cross-references
-            processed_content = self.add_cross_references_post_generation(content, str(output_file_path).replace('\\', '/'))
-            # Also process field references
-            lines = processed_content.split('\n')
+            # Field lines were emitted as "- `name: type`" -- the backticks are a
+            # placeholder (see format_field_with_references), not a code span.
+            # Unwrap them BEFORE cross-referencing: the cross-reference pass
+            # leaves real code spans alone, and these must receive links.
+            lines = content.split('\n')
             processed_lines = []
             for line in lines:
                 if line.strip().startswith('- `') and ':' in line:
@@ -916,8 +951,9 @@ def generate_docs(self, output_dir: str) -> None:
                         processed_lines.append(line)
                 else:
                     processed_lines.append(line)
-            
-            final_content = '\n'.join(processed_lines)
+
+            # Post-process content to add proper cross-references
+            final_content = self.add_cross_references_post_generation('\n'.join(processed_lines), str(output_file_path).replace('\\', '/'))
 
             with open(output_file_path, 'w', encoding='utf-8') as f:
                 f.write(final_content)