docs: salvage focused stale PR contributions

- add Vite and Redis pattern skills from closed stale PRs

- add frontend-slides support assets

- port skill-comply runner fixes and LLM prompt/provider regressions

- harden agent frontmatter validation and sync catalog counts

src/llm/prompt/__init__.py

@@ -1,13 +1,23 @@
 """Prompt module for prompt building and normalization."""
 
 from llm.prompt.builder import PromptBuilder, adapt_messages_for_provider, get_provider_builder
-from llm.prompt.templates import TEMPLATES, get_template, get_template_or_default
+from llm.prompt.templates import (
+    TEMPLATES,
+    clear_templates,
+    deregister_template,
+    get_template,
+    get_template_or_default,
+    register_template,
+)
 
 __all__ = (
     "PromptBuilder",
     "TEMPLATES",
+    "clear_templates",
+    "deregister_template",
     "adapt_messages_for_provider",
     "get_provider_builder",
     "get_template",
     "get_template_or_default",
+    "register_template",
 )

src/llm/prompt/builder.py

@@ -19,9 +19,32 @@ class PromptConfig:
     tool_format: str = "native"
 
 
-class PromptBuilder:
-    def __init__(self, config: PromptConfig | None = None) -> None:
-        self.config = config or PromptConfig()
+class PromptBuilder:
+    def __init__(
+        self,
+        config: PromptConfig | None = None,
+        *,
+        system_template: str | None = None,
+        user_template: str | None = None,
+        include_tools_in_system: bool | None = None,
+        tool_format: str | None = None,
+    ) -> None:
+        if config is not None and any(
+            value is not None
+            for value in (system_template, user_template, include_tools_in_system, tool_format)
+        ):
+            raise ValueError("Pass either config or PromptBuilder keyword options, not both")
+
+        if config is None:
+            overrides = {
+                "system_template": system_template,
+                "user_template": user_template,
+                "include_tools_in_system": include_tools_in_system,
+                "tool_format": tool_format,
+            }
+            config = PromptConfig(**{key: value for key, value in overrides.items() if value is not None})
+
+        self.config = config
 
     def build(self, messages: list[Message], tools: list[ToolDefinition] | None = None) -> list[Message]:
         if not messages:

src/llm/prompt/templates/__init__.py

@@ -1 +1,41 @@
-# Templates module for provider-specific prompt templates
+"""Provider-specific prompt template helpers."""
+
+from __future__ import annotations
+
+_TEMPLATE_REGISTRY: dict[str, str] = {}
+TEMPLATES = _TEMPLATE_REGISTRY
+
+
+def _validate_template_input(name: str, template: str | None = None) -> None:
+    """Validate template registry inputs before mutating the registry."""
+    if not isinstance(name, str) or not name.strip():
+        raise ValueError("Template name must be a non-empty string")
+    if template is not None and (not isinstance(template, str) or not template.strip()):
+        raise ValueError("Template content must be a non-empty string")
+
+
+def register_template(name: str, template: str) -> None:
+    """Register or replace a named prompt template."""
+    _validate_template_input(name, template)
+    _TEMPLATE_REGISTRY[name] = template
+
+
+def deregister_template(name: str) -> None:
+    """Remove a named prompt template if it is registered."""
+    _validate_template_input(name)
+    _TEMPLATE_REGISTRY.pop(name, None)
+
+
+def clear_templates() -> None:
+    """Remove all registered prompt templates."""
+    _TEMPLATE_REGISTRY.clear()
+
+
+def get_template(name: str) -> str | None:
+    """Return a named prompt template when one is registered."""
+    return _TEMPLATE_REGISTRY.get(name)
+
+
+def get_template_or_default(name: str, default: str = "") -> str:
+    """Return a named prompt template or a caller-provided default."""
+    return _TEMPLATE_REGISTRY.get(name, default)

src/llm/providers/claude.py

@@ -57,27 +57,39 @@ class ClaudeProvider(LLMProvider):
             }
             if input.max_tokens:
                 params["max_tokens"] = input.max_tokens
-            else:
-                params["max_tokens"] = 8192  # required by Anthropic API
-            if input.tools:
+            else:
+                params["max_tokens"] = 8192  # required by Anthropic API
+            if input.tools:
                 params["tools"] = [tool.to_anthropic_tool() for tool in input.tools]
 
             response = self.client.messages.create(**params)
 
-            tool_calls = None
-            if response.content and hasattr(response.content[0], "type"):
-                if response.content[0].type == "tool_use":
-                    tool_calls = [
-                        ToolCall(
-                            id=getattr(response.content[0], "id", ""),
-                            name=getattr(response.content[0], "name", ""),
-                            arguments=getattr(response.content[0].input, "__dict__", {}),
-                        )
-                    ]
-
-            return LLMOutput(
-                content=response.content[0].text if response.content else "",
-                tool_calls=tool_calls,
+            text_parts: list[str] = []
+            tool_calls: list[ToolCall] = []
+            for block in response.content or []:
+                block_type = getattr(block, "type", None)
+                if block_type == "text":
+                    text = getattr(block, "text", "")
+                    if text:
+                        text_parts.append(text)
+                elif block_type == "tool_use":
+                    raw_arguments = getattr(block, "input", {})
+                    arguments = (
+                        raw_arguments.copy()
+                        if isinstance(raw_arguments, dict)
+                        else getattr(raw_arguments, "__dict__", {}).copy()
+                    )
+                    tool_calls.append(
+                        ToolCall(
+                            id=getattr(block, "id", ""),
+                            name=getattr(block, "name", ""),
+                            arguments=arguments,
+                        )
+                    )
+
+            return LLMOutput(
+                content="".join(text_parts),
+                tool_calls=tool_calls or None,
                 model=response.model,
                 usage={
                     "input_tokens": response.usage.input_tokens,