eInteg testing

.github/workflows/pr-and-push.yml

@@ -3,7 +3,7 @@ name: Pull Request and Push Action
 on:
   pull_request:  # Safer than pull_request_target for untrusted code
     branches: [ main ]
-    types: [opened, synchronize, reopened, ready_for_review, review_requested, review_request_removed]
+    types: [opened, synchronize, reopened, ready_for_review]
   push:
     branches: [ main ]  # Also run on direct pushes to main
 concurrency:

README.md

@@ -21,6 +21,9 @@
     <a href="https://pypi.org/project/strands-agents/"><img alt="PyPI version" src="https://img.shields.io/pypi/v/strands-agents"/></a>
     <a href="https://python.org"><img alt="Python versions" src="https://img.shields.io/pypi/pyversions/strands-agents"/></a>
   </div>
+
+
+
 
   <p>
     <a href="https://strandsagents.com/">Documentation</a>
@@ -91,6 +94,17 @@ agent = Agent(tools=[word_count])
 response = agent("How many words are in this sentence?")
 ```
 
+**Hot Reloading from Directory:**
+Enable automatic tool loading and reloading from the `./tools/` directory:
+
+```python
+from strands import Agent
+
+# Agent will watch ./tools/ directory for changes
+agent = Agent(load_tools_from_directory=True)
+response = agent("Use any tools you find in the tools directory")
+```
+
 ### MCP Support
 
 Seamlessly integrate Model Context Protocol (MCP) servers:

pyproject.toml

@@ -29,7 +29,7 @@ dependencies = [
     "boto3>=1.26.0,<2.0.0",
     "botocore>=1.29.0,<2.0.0",
     "docstring_parser>=0.15,<1.0",
-    "mcp>=1.8.0,<2.0.0",
+    "mcp>=1.11.0,<2.0.0",
     "pydantic>=2.0.0,<3.0.0",
     "typing-extensions>=4.13.2,<5.0.0",
     "watchdog>=6.0.0,<7.0.0",
@@ -89,8 +89,15 @@ writer = [
     "writer-sdk>=2.2.0,<3.0.0"
 ]
 
+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"
+]
+
 a2a = [
-    "a2a-sdk[sql]>=0.2.16,<1.0.0",
+    "a2a-sdk>=0.3.0,<0.4.0",
+    "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",
@@ -136,7 +143,7 @@ all = [
     "opentelemetry-exporter-otlp-proto-http>=1.30.0,<2.0.0",
 
     # a2a
-    "a2a-sdk[sql]>=0.2.16,<1.0.0",
+    "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",
@@ -148,7 +155,7 @@ all = [
 source = "vcs"
 
 [tool.hatch.envs.hatch-static-analysis]
-features = ["anthropic", "litellm", "llamaapi", "ollama", "openai", "otel", "mistral", "writer", "a2a"]
+features = ["anthropic", "litellm", "llamaapi", "ollama", "openai", "otel", "mistral", "writer", "a2a", "sagemaker"]
 dependencies = [
   "mypy>=1.15.0,<2.0.0",
   "ruff>=0.11.6,<0.12.0",
@@ -171,7 +178,7 @@ lint-fix = [
 ]
 
 [tool.hatch.envs.hatch-test]
-features = ["anthropic", "litellm", "llamaapi", "ollama", "openai", "otel", "mistral", "writer", "a2a"]
+features = ["anthropic", "litellm", "llamaapi", "ollama", "openai", "otel", "mistral", "writer", "a2a", "sagemaker"]
 extra-dependencies = [
     "moto>=5.1.0,<6.0.0",
     "pytest>=8.0.0,<9.0.0",
@@ -187,7 +194,7 @@ extra-args = [
 
 [tool.hatch.envs.dev]
 dev-mode = true
-features = ["dev", "docs", "anthropic", "litellm", "llamaapi", "ollama", "otel", "mistral", "writer", "a2a"]
+features = ["dev", "docs", "anthropic", "litellm", "llamaapi", "ollama", "otel", "mistral", "writer", "a2a", "sagemaker"]
 
 [[tool.hatch.envs.hatch-test.matrix]]
 python = ["3.13", "3.12", "3.11", "3.10"]

src/strands/agent/conversation_manager/conversation_manager.py

@@ -36,7 +36,7 @@ def restore_from_session(self, state: dict[str, Any]) -> Optional[list[Message]]
         Args:
             state: Previous state of the conversation manager
         Returns:
-            Optional list of messages to prepend to the agents messages. By defualt returns None.
+            Optional list of messages to prepend to the agents messages. By default returns None.
         """
         if state.get("__name__") != self.__class__.__name__:
             raise ValueError("Invalid conversation manager state.")

src/strands/event_loop/_recover_message_on_max_tokens_reached.py

@@ -0,0 +1,71 @@
+"""Message recovery utilities for handling max token limit scenarios.
+
+This module provides functionality to recover and clean up incomplete messages that occur
+when model responses are truncated due to maximum token limits being reached. It specifically
+handles cases where tool use blocks are incomplete or malformed due to truncation.
+"""
+
+import logging
+
+from ..types.content import ContentBlock, Message
+from ..types.tools import ToolUse
+
+logger = logging.getLogger(__name__)
+
+
+def recover_message_on_max_tokens_reached(message: Message) -> Message:
+    """Recover and clean up messages when max token limits are reached.
+
+    When a model response is truncated due to maximum token limits, all tool use blocks
+    should be replaced with informative error messages since they may be incomplete or
+    unreliable. This function inspects the message content and:
+
+    1. Identifies all tool use blocks (regardless of validity)
+    2. Replaces all tool uses with informative error messages
+    3. Preserves all non-tool content blocks (text, images, etc.)
+    4. Returns a cleaned message suitable for conversation history
+
+    This recovery mechanism ensures that the conversation can continue gracefully even when
+    model responses are truncated, providing clear feedback about what happened and preventing
+    potentially incomplete or corrupted tool executions.
+
+    Args:
+        message: The potentially incomplete message from the model that was truncated
+                due to max token limits.
+
+    Returns:
+        A cleaned Message with all tool uses replaced by explanatory text content.
+        The returned message maintains the same role as the input message.
+
+    Example:
+        If a message contains any tool use (complete or incomplete):
+        ```
+        {"toolUse": {"name": "calculator", "input": {"expression": "2+2"}, "toolUseId": "123"}}
+        ```
+
+        It will be replaced with:
+        ```
+        {"text": "The selected tool calculator's tool use was incomplete due to maximum token limits being reached."}
+        ```
+    """
+    logger.info("handling max_tokens stop reason - replacing all tool uses with error messages")
+
+    valid_content: list[ContentBlock] = []
+    for content in message["content"] or []:
+        tool_use: ToolUse | None = content.get("toolUse")
+        if not tool_use:
+            valid_content.append(content)
+            continue
+
+        # Replace all tool uses with error messages when max_tokens is reached
+        display_name = tool_use.get("name") or "<unknown>"
+        logger.warning("tool_name=<%s> | replacing with error message due to max_tokens truncation.", display_name)
+
+        valid_content.append(
+            {
+                "text": f"The selected tool {display_name}'s tool use was incomplete due "
+                f"to maximum token limits being reached."
+            }
+        )
+
+    return {"content": valid_content, "role": message["role"]}

src/strands/event_loop/event_loop.py

@@ -28,9 +28,15 @@
 from ..telemetry.tracer import get_tracer
 from ..tools.executor import run_tools, validate_and_prepare_tools
 from ..types.content import Message
-from ..types.exceptions import ContextWindowOverflowException, EventLoopException, ModelThrottledException
+from ..types.exceptions import (
+    ContextWindowOverflowException,
+    EventLoopException,
+    MaxTokensReachedException,
+    ModelThrottledException,
+)
 from ..types.streaming import Metrics, StopReason
 from ..types.tools import ToolChoice, ToolChoiceAuto, ToolConfig, ToolGenerator, ToolResult, ToolUse
+from ._recover_message_on_max_tokens_reached import recover_message_on_max_tokens_reached
 from .streaming import stream_messages
 
 if TYPE_CHECKING:
@@ -151,6 +157,9 @@ async def event_loop_cycle(agent: "Agent", invocation_state: dict[str, Any]) ->
                     )
                 )
 
+                if stop_reason == "max_tokens":
+                    message = recover_message_on_max_tokens_reached(message)
+
                 if model_invoke_span:
                     tracer.end_model_invoke_span(model_invoke_span, message, usage, stop_reason)
                 break  # Success! Break out of retry loop
@@ -200,6 +209,22 @@ async def event_loop_cycle(agent: "Agent", invocation_state: dict[str, Any]) ->
         agent.event_loop_metrics.update_usage(usage)
         agent.event_loop_metrics.update_metrics(metrics)
 
+        if stop_reason == "max_tokens":
+            """
+            Handle max_tokens limit reached by the model.
+
+            When the model reaches its maximum token limit, this represents a potentially unrecoverable
+            state where the model's response was truncated. By default, Strands fails hard with an
+            MaxTokensReachedException to maintain consistency with other failure types.
+            """
+            raise MaxTokensReachedException(
+                message=(
+                    "Agent has reached an unrecoverable state due to max_tokens limit. "
+                    "For more information see: "
+                    "https://strandsagents.com/latest/user-guide/concepts/agents/agent-loop/#maxtokensreachedexception"
+                )
+            )
+
         # If the model is requesting to use tools
         if stop_reason == "tool_use":
             # Handle tool execution
@@ -231,7 +256,8 @@ async def event_loop_cycle(agent: "Agent", invocation_state: dict[str, Any]) ->
         # Don't yield or log the exception - we already did it when we
         # raised the exception and we don't need that duplication.
         raise
-    except ContextWindowOverflowException as e:
+    except (ContextWindowOverflowException, MaxTokensReachedException) as e:
+        # Special cased exceptions which we want to bubble up rather than get wrapped in an EventLoopException
         if cycle_span:
             tracer.end_span_with_error(cycle_span, str(e), e)
         raise e

src/strands/models/anthropic.py

@@ -414,7 +414,7 @@ async def structured_output(
         stop_reason, messages, _, _ = event["stop"]
 
         if stop_reason != "tool_use":
-            raise ValueError("No valid tool use or tool use input was found in the Anthropic response.")
+            raise ValueError(f'Model returned stop_reason: {stop_reason} instead of "tool_use".')
 
         content = messages["content"]
         output_response: dict[str, Any] | None = None

src/strands/models/bedrock.py

@@ -17,10 +17,10 @@
 
 from ..event_loop import streaming
 from ..tools import convert_pydantic_to_tool_spec
-from ..types.content import Messages
+from ..types.content import ContentBlock, Message, Messages
 from ..types.exceptions import ContextWindowOverflowException, ModelThrottledException
 from ..types.streaming import StreamEvent
-from ..types.tools import ToolSpec
+from ..types.tools import ToolResult, ToolSpec
 from .model import Model
 
 logger = logging.getLogger(__name__)
@@ -181,7 +181,7 @@ def format_request(
         """
         return {
             "modelId": self.config["model_id"],
-            "messages": messages,
+            "messages": self._format_bedrock_messages(messages),
             "system": [
                 *([{"text": system_prompt}] if system_prompt else []),
                 *([{"cachePoint": {"type": self.config["cache_prompt"]}}] if self.config.get("cache_prompt") else []),
@@ -246,6 +246,53 @@ def format_request(
             ),
         }
 
+    def _format_bedrock_messages(self, messages: Messages) -> Messages:
+        """Format messages for Bedrock API compatibility.
+
+        This function ensures messages conform to Bedrock's expected format by:
+        - Cleaning tool result content blocks by removing additional fields that may be
+          useful for retaining information in hooks but would cause Bedrock validation
+          exceptions when presented with unexpected fields
+        - Ensuring all message content blocks are properly formatted for the Bedrock API
+
+        Args:
+            messages: List of messages to format
+
+        Returns:
+            Messages formatted for Bedrock API compatibility
+
+        Note:
+            Bedrock will throw validation exceptions when presented with additional
+            unexpected fields in tool result blocks.
+            https://docs.aws.amazon.com/bedrock/latest/APIReference/API_runtime_ToolResultBlock.html
+        """
+        cleaned_messages = []
+
+        for message in messages:
+            cleaned_content: list[ContentBlock] = []
+
+            for content_block in message["content"]:
+                if "toolResult" in content_block:
+                    # Create a new content block with only the cleaned toolResult
+                    tool_result: ToolResult = content_block["toolResult"]
+
+                    # Keep only the required fields for Bedrock
+                    cleaned_tool_result = ToolResult(
+                        content=tool_result["content"], toolUseId=tool_result["toolUseId"], status=tool_result["status"]
+                    )
+
+                    cleaned_block: ContentBlock = {"toolResult": cleaned_tool_result}
+                    cleaned_content.append(cleaned_block)
+                else:
+                    # Keep other content blocks as-is
+                    cleaned_content.append(content_block)
+
+            # Create new message with cleaned content
+            cleaned_message: Message = Message(content=cleaned_content, role=message["role"])
+            cleaned_messages.append(cleaned_message)
+
+        return cleaned_messages
+
     def _has_blocked_guardrail(self, guardrail_data: dict[str, Any]) -> bool:
         """Check if guardrail data contains any blocked policies.
 
@@ -584,7 +631,7 @@ async def structured_output(
         stop_reason, messages, _, _ = event["stop"]
 
         if stop_reason != "tool_use":
-            raise ValueError("No valid tool use or tool use input was found in the Bedrock response.")
+            raise ValueError(f'Model returned stop_reason: {stop_reason} instead of "tool_use".')
 
         content = messages["content"]
         output_response: dict[str, Any] | None = None