Complete ConfigLoader system implementation with cache-free architecture

Comprehensive ConfigLoader system implementation providing configuration-driven agent development:

Core ConfigLoader Classes:
- AgentConfigLoader: Agent configuration with models, tools, structured output
- GraphConfigLoader: Graph-based workflows with nodes, edges, conditions
- SwarmConfigLoader: Multi-agent swarm coordination and management
- ToolConfigLoader: Tool configuration including Agent-as-Tool, Swarm-as-Tool, Graph-as-Tool

Key Features Implemented:
- Cache-free architecture: Simplified API without caching complexity
- Top-level key requirements: Enforced schema validation (agent:, graph:, swarm:, tools:)
- JSON Schema validation: Complete schema system with IDE integration
- Structured output support: Pydantic model generation and validation
- Multi-agent tools: Agents, Swarms, and Graphs as reusable tools
- Advanced patterns: Nested configurations, parameter substitution, conditional routing

Implementation Details:

Agent Configuration:
- Model configuration with provider-specific settings
- Tool integration with custom and built-in tools
- Structured output with automatic Pydantic model generation
- Schema registry for output validation and type safety
- PydanticFactory for dynamic model creation from JSON schemas

Graph Configuration:
- Node and edge definition with flexible routing
- Conditional execution with custom condition functions
- Entry point specification for workflow initiation
- Parallel and sequential execution patterns
- Complex workflow orchestration capabilities

Swarm Configuration:
- Multi-agent coordination and communication
- Agent role definition and responsibility assignment
- Swarm-level configuration and management
- Inter-agent communication patterns

Tool Configuration:
- Multi-agent tool patterns (Agent-as-Tool, Swarm-as-Tool, Graph-as-Tool)
- Tool composition and nesting capabilities
- Parameter passing and configuration inheritance
- Future MCP (Model Context Protocol) integration planning

Schema System:
- Complete JSON Schema for all configuration types
- IDE integration with VSCode, IntelliJ, Vim support
- Real-time validation and autocompletion
- Schema documentation and examples
- Comprehensive error handling and validation

Testing Infrastructure:
- 113 comprehensive tests covering all functionality
- Integration tests for multi-agent scenarios
- Structured output validation tests
- Schema registry and PydanticFactory tests
- Multi-agent tool integration tests
- Object independence validation (cache-free verification)

Breaking Changes from Cache Removal:
- Removed cache_key parameters from all load_* methods
- Removed clear_cache() methods from all loaders
- Removed get_available_* methods from all loaders
- Simplified API with predictable behavior (each load creates fresh objects)

Migration Benefits:
- Cleaner code: ~200+ lines of cache complexity removed
- Predictable behavior: No cache state management needed
- Easier debugging: No cache-related side effects
- Reduced memory usage: No indefinite object retention
- Immediate configuration change reflection
- Simplified testing and development

Production Quality:
- Comprehensive error handling and validation
- Type safety with Pydantic integration
- Extensive test coverage (113 passing tests)
- Clean, maintainable architecture
- Professional API design
- Complete documentation and examples

Future Extensibility:
- MCP integration planning for advanced tool loading
- Extensible schema system for custom configurations
- Plugin architecture for custom loaders
- Advanced workflow patterns and orchestration

This ConfigLoader system enables configuration-driven agent development with world-class developer experience, comprehensive validation, and production-ready reliability.

Author: vawsgit

.gitignore

@@ -1,3 +1,4 @@
+.DS_Store
 build
 __pycache__*
 .coverage*

pyproject.toml

@@ -31,6 +31,7 @@ dependencies = [
     "docstring_parser>=0.15,<1.0",
     "mcp>=1.11.0,<2.0.0",
     "pydantic>=2.0.0,<3.0.0",
+    "PyYAML>=6.0.0,<7.0.0",
     "typing-extensions>=4.13.2,<5.0.0",
     "watchdog>=6.0.0,<7.0.0",
     "opentelemetry-api>=1.30.0,<2.0.0",
@@ -69,6 +70,8 @@ docs = [
 ]
 litellm = [
     "litellm>=1.73.1,<2.0.0",
+    # https://github.com/BerriAI/litellm/issues/13711
+    "openai<1.100.0",
 ]
 llamaapi = [
     "llama-api-client>=0.1.0,<1.0.0",
@@ -92,7 +95,9 @@ writer = [
 sagemaker = [
     "boto3>=1.26.0,<2.0.0",
     "botocore>=1.29.0,<2.0.0",
-    "boto3-stubs[sagemaker-runtime]>=1.26.0,<2.0.0"
+    "boto3-stubs[sagemaker-runtime]>=1.26.0,<2.0.0",
+    # uses OpenAI as part of the implementation
+    "openai>=1.68.0,<2.0.0",
 ]
 
 a2a = [
@@ -104,50 +109,7 @@ a2a = [
     "starlette>=0.46.2,<1.0.0",
 ]
 all = [
-    # anthropic
-    "anthropic>=0.21.0,<1.0.0",
-
-    # dev
-    "commitizen>=4.4.0,<5.0.0",
-    "hatch>=1.0.0,<2.0.0",
-    "moto>=5.1.0,<6.0.0",
-    "mypy>=1.15.0,<2.0.0",
-    "pre-commit>=3.2.0,<4.2.0",
-    "pytest>=8.0.0,<9.0.0",
-    "pytest-asyncio>=0.26.0,<0.27.0",
-    "pytest-cov>=4.1.0,<5.0.0",
-    "pytest-xdist>=3.0.0,<4.0.0",
-    "ruff>=0.4.4,<0.5.0",
-
-    # docs
-    "sphinx>=5.0.0,<6.0.0",
-    "sphinx-rtd-theme>=1.0.0,<2.0.0",
-    "sphinx-autodoc-typehints>=1.12.0,<2.0.0",
-
-    # litellm
-    "litellm>=1.72.6,<1.73.0",
-
-    # llama
-    "llama-api-client>=0.1.0,<1.0.0",
-
-    # mistral
-    "mistralai>=1.8.2",
-
-    # ollama
-    "ollama>=0.4.8,<1.0.0",
-
-    # openai
-    "openai>=1.68.0,<2.0.0",
-
-    # otel
-    "opentelemetry-exporter-otlp-proto-http>=1.30.0,<2.0.0",
-
-    # a2a
-    "a2a-sdk[sql]>=0.3.0,<0.4.0",
-    "uvicorn>=0.34.2,<1.0.0",
-    "httpx>=0.28.1,<1.0.0",
-    "fastapi>=0.115.12,<1.0.0",
-    "starlette>=0.46.2,<1.0.0",
+    "strands-agents[a2a,anthropic,dev,docs,litellm,llamaapi,mistral,ollama,openai,otel]",
 ]
 
 [tool.hatch.version]
@@ -159,7 +121,7 @@ features = ["anthropic", "litellm", "llamaapi", "ollama", "openai", "otel", "mis
 dependencies = [
   "mypy>=1.15.0,<2.0.0",
   "ruff>=0.11.6,<0.12.0",
-  "strands-agents @ {root:uri}"
+  "strands-agents @ {root:uri}",
 ]
 
 [tool.hatch.envs.hatch-static-analysis.scripts]

src/strands/agent/agent.py

@@ -642,8 +642,11 @@ def _record_tool_execution(
             tool_result: The result returned by the tool.
             user_message_override: Optional custom message to include.
         """
+        # Filter tool input parameters to only include those defined in tool spec
+        filtered_input = self._filter_tool_parameters_for_recording(tool["name"], tool["input"])
+
         # Create user message describing the tool call
-        input_parameters = json.dumps(tool["input"], default=lambda o: f"<<non-serializable: {type(o).__qualname__}>>")
+        input_parameters = json.dumps(filtered_input, default=lambda o: f"<<non-serializable: {type(o).__qualname__}>>")
 
         user_msg_content: list[ContentBlock] = [
             {"text": (f"agent.tool.{tool['name']} direct tool call.\nInput parameters: {input_parameters}\n")}
@@ -653,14 +656,21 @@ def _record_tool_execution(
         if user_message_override:
             user_msg_content.insert(0, {"text": f"{user_message_override}\n"})
 
+        # Create filtered tool use for message history
+        filtered_tool: ToolUse = {
+            "toolUseId": tool["toolUseId"],
+            "name": tool["name"],
+            "input": filtered_input,
+        }
+
         # Create the message sequence
         user_msg: Message = {
             "role": "user",
             "content": user_msg_content,
         }
         tool_use_msg: Message = {
             "role": "assistant",
-            "content": [{"toolUse": tool}],
+            "content": [{"toolUse": filtered_tool}],
         }
         tool_result_msg: Message = {
             "role": "user",
@@ -717,6 +727,25 @@ def _end_agent_trace_span(
 
             self.tracer.end_agent_span(**trace_attributes)
 
+    def _filter_tool_parameters_for_recording(self, tool_name: str, input_params: dict[str, Any]) -> dict[str, Any]:
+        """Filter input parameters to only include those defined in the tool specification.
+
+        Args:
+            tool_name: Name of the tool to get specification for
+            input_params: Original input parameters
+
+        Returns:
+            Filtered parameters containing only those defined in tool spec
+        """
+        all_tools_config = self.tool_registry.get_all_tools_config()
+        tool_spec = all_tools_config.get(tool_name)
+
+        if not tool_spec or "inputSchema" not in tool_spec:
+            return input_params.copy()
+
+        properties = tool_spec["inputSchema"]["json"]["properties"]
+        return {k: v for k, v in input_params.items() if k in properties}
+
     def _append_message(self, message: Message) -> None:
         """Appends a message to the agent's list of messages and invokes the callbacks for the MessageCreatedEvent."""
         self.messages.append(message)

src/strands/event_loop/streaming.py

@@ -40,10 +40,12 @@ def remove_blank_messages_content_text(messages: Messages) -> Messages:
         # only modify assistant messages
         if "role" in message and message["role"] != "assistant":
             continue
-
         if "content" in message:
             content = message["content"]
             has_tool_use = any("toolUse" in item for item in content)
+            if len(content) == 0:
+                content.append({"text": "[blank text]"})
+                continue
 
             if has_tool_use:
                 # Remove blank 'text' items for assistant messages
@@ -194,16 +196,18 @@ def handle_content_block_stop(state: dict[str, Any]) -> dict[str, Any]:
         state["text"] = ""
 
     elif reasoning_text:
-        content.append(
-            {
-                "reasoningContent": {
-                    "reasoningText": {
-                        "text": state["reasoningText"],
-                        "signature": state["signature"],
-                    }
+        content_block: ContentBlock = {
+            "reasoningContent": {
+                "reasoningText": {
+                    "text": state["reasoningText"],
                 }
             }
-        )
+        }
+
+        if "signature" in state:
+            content_block["reasoningContent"]["reasoningText"]["signature"] = state["signature"]
+
+        content.append(content_block)
         state["reasoningText"] = ""
 
     return state
@@ -263,7 +267,6 @@ async def process_stream(chunks: AsyncIterable[StreamEvent]) -> AsyncGenerator[d
         "text": "",
         "current_tool_use": {},
         "reasoningText": "",
-        "signature": "",
     }
     state["content"] = state["message"]["content"]
 
@@ -272,7 +275,6 @@ async def process_stream(chunks: AsyncIterable[StreamEvent]) -> AsyncGenerator[d
 
     async for chunk in chunks:
         yield {"callback": {"event": chunk}}
-
         if "messageStart" in chunk:
             state["message"] = handle_message_start(chunk["messageStart"], state["message"])
         elif "contentBlockStart" in chunk:
@@ -312,7 +314,6 @@ async def stream_messages(
     logger.debug("model=<%s> | streaming messages", model)
 
     messages = remove_blank_messages_content_text(messages)
-
     chunks = model.stream(messages, tool_specs if tool_specs else None, system_prompt)
 
     async for event in process_stream(chunks):

src/strands/experimental/config_loader/__init__.py

@@ -0,0 +1,8 @@
+"""Contains logic that loads agent configurations from YAML files."""
+
+from .agent import AgentConfigLoader
+from .graph import GraphConfigLoader
+from .swarm import SwarmConfigLoader
+from .tools import AgentAsToolWrapper, ToolConfigLoader
+
+__all__ = ["AgentConfigLoader", "ToolConfigLoader", "AgentAsToolWrapper", "GraphConfigLoader", "SwarmConfigLoader"]

src/strands/experimental/config_loader/agent/__init__.py

@@ -0,0 +1,25 @@
+"""Agent configuration loader module."""
+
+from .agent_config_loader import AgentConfigLoader
+from .pydantic_factory import PydanticModelFactory
+from .schema_registry import SchemaRegistry
+from .structured_output_errors import (
+    ModelCreationError,
+    OutputValidationError,
+    SchemaImportError,
+    SchemaRegistryError,
+    SchemaValidationError,
+    StructuredOutputError,
+)
+
+__all__ = [
+    "AgentConfigLoader",
+    "PydanticModelFactory",
+    "SchemaRegistry",
+    "StructuredOutputError",
+    "SchemaValidationError",
+    "ModelCreationError",
+    "OutputValidationError",
+    "SchemaRegistryError",
+    "SchemaImportError",
+]