improve task guardrails to address issue #2177

* Update constants.py

This PR updates the list of foundation models available in Amazon Bedrock to reflect the latest offerings.

* Update constants.py with inference profiles

Add the cross-region inference profiles to increase throughput and improve resiliency by routing your requests across multiple AWS Regions during peak utilization bursts.

* Update constants.py

Fix the model order

---------

Co-authored-by: Brandon Hancock (bhancock_ai) <109994880+bhancockio@users.noreply.github.com>

docs/concepts/memory.mdx

@@ -506,7 +506,7 @@ my_crew = Crew(
 )
 ```
 
-### Resetting Memory
+### Resetting Memory via cli
 
 ```shell
 crewai reset-memories [OPTIONS]
@@ -520,8 +520,46 @@ crewai reset-memories [OPTIONS]
 | `-s`, `--short`    | Reset SHORT TERM memory.         | Flag (boolean) | False   |
 | `-e`, `--entities` | Reset ENTITIES memory.           | Flag (boolean) | False   |
 | `-k`, `--kickoff-outputs` | Reset LATEST KICKOFF TASK OUTPUTS. | Flag (boolean) | False   |
+| `-kn`, `--knowledge` | Reset KNOWLEDEGE storage | Flag (boolean) | False   |
 | `-a`, `--all`      | Reset ALL memories.              | Flag (boolean) | False   |
 
+Note: To use the cli command you need to have your crew in a file called crew.py in the same directory.
+
+
+
+
+### Resetting Memory via crew object
+
+```python
+
+my_crew = Crew(
+    agents=[...],
+    tasks=[...],
+    process=Process.sequential,
+    memory=True,
+    verbose=True,
+    embedder={
+        "provider": "custom",
+        "config": {
+            "embedder": CustomEmbedder()
+        }
+    }
+)
+
+my_crew.reset_memories(command_type = 'all') # Resets all the memory
+```
+
+#### Resetting Memory Options
+
+| Command Type       | Description                      |
+| :----------------- | :------------------------------- |
+| `long`             | Reset LONG TERM memory.          | 
+| `short`            | Reset SHORT TERM memory.         | 
+| `entities`         | Reset ENTITIES memory.           | 
+| `kickoff_outputs`  | Reset LATEST KICKOFF TASK OUTPUTS. |
+| `knowledge`        | Reset KNOWLEDGE memory.          |
+| `all`              | Reset ALL memories.              |
+
 
 ## Benefits of Using CrewAI's Memory System
 

docs/how-to/kickoff-async.mdx

@@ -54,7 +54,8 @@ coding_agent = Agent(
 # Create a task that requires code execution
 data_analysis_task = Task(
     description="Analyze the given dataset and calculate the average age of participants. Ages: {ages}",
-    agent=coding_agent
+    agent=coding_agent,
+    expected_output="The average age of the participants."
 )
 
 # Create a crew and add the task
@@ -116,4 +117,4 @@ async def async_multiple_crews():
 
 # Run the async function
 asyncio.run(async_multiple_crews())
-```
+```

src/crewai/cli/constants.py

@@ -216,10 +216,43 @@ MODELS = {
         "watsonx/ibm/granite-3-8b-instruct",
     ],
     "bedrock": [
+        "bedrock/us.amazon.nova-pro-v1:0",
+        "bedrock/us.amazon.nova-micro-v1:0",
+        "bedrock/us.amazon.nova-lite-v1:0",
+        "bedrock/us.anthropic.claude-3-5-sonnet-20240620-v1:0",
+        "bedrock/us.anthropic.claude-3-5-haiku-20241022-v1:0",
+        "bedrock/us.anthropic.claude-3-5-sonnet-20241022-v2:0",
+        "bedrock/us.anthropic.claude-3-7-sonnet-20250219-v1:0",
+        "bedrock/us.anthropic.claude-3-sonnet-20240229-v1:0",
+        "bedrock/us.anthropic.claude-3-opus-20240229-v1:0",
+        "bedrock/us.anthropic.claude-3-haiku-20240307-v1:0",
+        "bedrock/us.meta.llama3-2-11b-instruct-v1:0",
+        "bedrock/us.meta.llama3-2-3b-instruct-v1:0",
+        "bedrock/us.meta.llama3-2-90b-instruct-v1:0",
+        "bedrock/us.meta.llama3-2-1b-instruct-v1:0",
+        "bedrock/us.meta.llama3-1-8b-instruct-v1:0",
+        "bedrock/us.meta.llama3-1-70b-instruct-v1:0",
+        "bedrock/us.meta.llama3-3-70b-instruct-v1:0",
+        "bedrock/us.meta.llama3-1-405b-instruct-v1:0",
+        "bedrock/eu.anthropic.claude-3-5-sonnet-20240620-v1:0",
+        "bedrock/eu.anthropic.claude-3-sonnet-20240229-v1:0",
+        "bedrock/eu.anthropic.claude-3-haiku-20240307-v1:0",
+        "bedrock/eu.meta.llama3-2-3b-instruct-v1:0",
+        "bedrock/eu.meta.llama3-2-1b-instruct-v1:0",
+        "bedrock/apac.anthropic.claude-3-5-sonnet-20240620-v1:0",
+        "bedrock/apac.anthropic.claude-3-5-sonnet-20241022-v2:0",
+        "bedrock/apac.anthropic.claude-3-sonnet-20240229-v1:0",
+        "bedrock/apac.anthropic.claude-3-haiku-20240307-v1:0",
+        "bedrock/amazon.nova-pro-v1:0",
+        "bedrock/amazon.nova-micro-v1:0",
+        "bedrock/amazon.nova-lite-v1:0",
         "bedrock/anthropic.claude-3-5-sonnet-20240620-v1:0",
+        "bedrock/anthropic.claude-3-5-haiku-20241022-v1:0",
+        "bedrock/anthropic.claude-3-5-sonnet-20241022-v2:0",
+        "bedrock/anthropic.claude-3-7-sonnet-20250219-v1:0",
         "bedrock/anthropic.claude-3-sonnet-20240229-v1:0",
-        "bedrock/anthropic.claude-3-haiku-20240307-v1:0",
         "bedrock/anthropic.claude-3-opus-20240229-v1:0",
+        "bedrock/anthropic.claude-3-haiku-20240307-v1:0",
         "bedrock/anthropic.claude-v2:1",
         "bedrock/anthropic.claude-v2",
         "bedrock/anthropic.claude-instant-v1",
@@ -234,8 +267,6 @@ MODELS = {
         "bedrock/ai21.j2-mid-v1",
         "bedrock/ai21.j2-ultra-v1",
         "bedrock/ai21.jamba-instruct-v1:0",
-        "bedrock/meta.llama2-13b-chat-v1",
-        "bedrock/meta.llama2-70b-chat-v1",
         "bedrock/mistral.mistral-7b-instruct-v0:2",
         "bedrock/mistral.mixtral-8x7b-instruct-v0:1",
     ],

src/crewai/cli/utils.py

@@ -257,11 +257,11 @@ def get_crew(crew_path: str = "crew.py", require: bool = False) -> Crew | None:
         import os
 
         for root, _, files in os.walk("."):
-            if "crew.py" in files:
-                crew_path = os.path.join(root, "crew.py")
+            if crew_path in files:
+                crew_os_path = os.path.join(root, crew_path)
                 try:
                     spec = importlib.util.spec_from_file_location(
-                        "crew_module", crew_path
+                        "crew_module", crew_os_path
                     )
                     if not spec or not spec.loader:
                         continue
@@ -273,9 +273,11 @@ def get_crew(crew_path: str = "crew.py", require: bool = False) -> Crew | None:
                         for attr_name in dir(module):
                             attr = getattr(module, attr_name)
                             try:
-                                if callable(attr) and hasattr(attr, "crew"):
-                                    crew_instance = attr().crew()
-                                    return crew_instance
+                                if isinstance(attr, Crew) and hasattr(attr, "kickoff"):
+                                    print(
+                                        f"Found valid crew object in attribute '{attr_name}' at {crew_os_path}."
+                                    )
+                                    return attr
 
                             except Exception as e:
                                 print(f"Error processing attribute {attr_name}: {e}")

src/crewai/flow/utils.py

@@ -16,7 +16,8 @@ Example
 import ast
 import inspect
 import textwrap
-from typing import Any, Dict, List, Optional, Set, Union
+from collections import defaultdict, deque
+from typing import Any, Deque, Dict, List, Optional, Set, Union
 
 
 def get_possible_return_constants(function: Any) -> Optional[List[str]]:
@@ -118,7 +119,7 @@ def calculate_node_levels(flow: Any) -> Dict[str, int]:
     - Processes router paths separately
     """
     levels: Dict[str, int] = {}
-    queue: List[str] = []
+    queue: Deque[str] = deque()
     visited: Set[str] = set()
     pending_and_listeners: Dict[str, Set[str]] = {}
 
@@ -128,28 +129,35 @@ def calculate_node_levels(flow: Any) -> Dict[str, int]:
             levels[method_name] = 0
             queue.append(method_name)
 
+    # Precompute listener dependencies
+    or_listeners = defaultdict(list)
+    and_listeners = defaultdict(set)
+    for listener_name, (condition_type, trigger_methods) in flow._listeners.items():
+        if condition_type == "OR":
+            for method in trigger_methods:
+                or_listeners[method].append(listener_name)
+        elif condition_type == "AND":
+            and_listeners[listener_name] = set(trigger_methods)
+
     # Breadth-first traversal to assign levels
     while queue:
-        current = queue.pop(0)
+        current = queue.popleft()
         current_level = levels[current]
         visited.add(current)
 
-        for listener_name, (condition_type, trigger_methods) in flow._listeners.items():
-            if condition_type == "OR":
-                if current in trigger_methods:
-                    if (
-                        listener_name not in levels
-                        or levels[listener_name] > current_level + 1
-                    ):
-                        levels[listener_name] = current_level + 1
-                        if listener_name not in visited:
-                            queue.append(listener_name)
-            elif condition_type == "AND":
+        for listener_name in or_listeners[current]:
+            if listener_name not in levels or levels[listener_name] > current_level + 1:
+                levels[listener_name] = current_level + 1
+                if listener_name not in visited:
+                    queue.append(listener_name)
+
+        for listener_name, required_methods in and_listeners.items():
+            if current in required_methods:
                 if listener_name not in pending_and_listeners:
                     pending_and_listeners[listener_name] = set()
-                if current in trigger_methods:
-                    pending_and_listeners[listener_name].add(current)
-                if set(trigger_methods) == pending_and_listeners[listener_name]:
+                pending_and_listeners[listener_name].add(current)
+
+                if required_methods == pending_and_listeners[listener_name]:
                     if (
                         listener_name not in levels
                         or levels[listener_name] > current_level + 1
@@ -159,22 +167,7 @@ def calculate_node_levels(flow: Any) -> Dict[str, int]:
                             queue.append(listener_name)
 
         # Handle router connections
-        if current in flow._routers:
-            router_method_name = current
-            paths = flow._router_paths.get(router_method_name, [])
-            for path in paths:
-                for listener_name, (
-                    condition_type,
-                    trigger_methods,
-                ) in flow._listeners.items():
-                    if path in trigger_methods:
-                        if (
-                            listener_name not in levels
-                            or levels[listener_name] > current_level + 1
-                        ):
-                            levels[listener_name] = current_level + 1
-                            if listener_name not in visited:
-                                queue.append(listener_name)
+        process_router_paths(flow, current, current_level, levels, queue)
 
     return levels
 
@@ -227,10 +220,7 @@ def build_ancestor_dict(flow: Any) -> Dict[str, Set[str]]:
 
 
 def dfs_ancestors(
-    node: str,
-    ancestors: Dict[str, Set[str]],
-    visited: Set[str],
-    flow: Any
+    node: str, ancestors: Dict[str, Set[str]], visited: Set[str], flow: Any
 ) -> None:
     """
     Perform depth-first search to build ancestor relationships.
@@ -274,7 +264,9 @@ def dfs_ancestors(
                     dfs_ancestors(listener_name, ancestors, visited, flow)
 
 
-def is_ancestor(node: str, ancestor_candidate: str, ancestors: Dict[str, Set[str]]) -> bool:
+def is_ancestor(
+    node: str, ancestor_candidate: str, ancestors: Dict[str, Set[str]]
+) -> bool:
     """
     Check if one node is an ancestor of another.
 
@@ -339,7 +331,9 @@ def build_parent_children_dict(flow: Any) -> Dict[str, List[str]]:
     return parent_children
 
 
-def get_child_index(parent: str, child: str, parent_children: Dict[str, List[str]]) -> int:
+def get_child_index(
+    parent: str, child: str, parent_children: Dict[str, List[str]]
+) -> int:
     """
     Get the index of a child node in its parent's sorted children list.
 
@@ -360,3 +354,23 @@ def get_child_index(parent: str, child: str, parent_children: Dict[str, List[str
     children = parent_children.get(parent, [])
     children.sort()
     return children.index(child)
+
+
+def process_router_paths(flow, current, current_level, levels, queue):
+    """
+    Handle the router connections for the current node.
+    """
+    if current in flow._routers:
+        paths = flow._router_paths.get(current, [])
+        for path in paths:
+            for listener_name, (
+                condition_type,
+                trigger_methods,
+            ) in flow._listeners.items():
+                if path in trigger_methods:
+                    if (
+                        listener_name not in levels
+                        or levels[listener_name] > current_level + 1
+                    ):
+                        levels[listener_name] = current_level + 1
+                        queue.append(listener_name)

src/crewai/llm.py

@@ -64,6 +64,7 @@ LLM_CONTEXT_WINDOW_SIZES = {
     "gpt-4-turbo": 128000,
     "o1-preview": 128000,
     "o1-mini": 128000,
+    "o3-mini": 200000,  # Based on official o3-mini specifications
     # gemini
     "gemini-2.0-flash": 1048576,
     "gemini-1.5-pro": 2097152,
@@ -485,10 +486,23 @@ class LLM:
         """
         Returns the context window size, using 75% of the maximum to avoid
         cutting off messages mid-thread.
+
+        Raises:
+            ValueError: If a model's context window size is outside valid bounds (1024-2097152)
         """
         if self.context_window_size != 0:
             return self.context_window_size
 
+        MIN_CONTEXT = 1024
+        MAX_CONTEXT = 2097152  # Current max from gemini-1.5-pro
+
+        # Validate all context window sizes
+        for key, value in LLM_CONTEXT_WINDOW_SIZES.items():
+            if value < MIN_CONTEXT or value > MAX_CONTEXT:
+                raise ValueError(
+                    f"Context window for {key} must be between {MIN_CONTEXT} and {MAX_CONTEXT}"
+                )
+
         self.context_window_size = int(
             DEFAULT_CONTEXT_WINDOW_SIZE * CONTEXT_WINDOW_USAGE_RATIO
         )

src/crewai/task.py

@@ -19,6 +19,8 @@ from typing import (
     Tuple,
     Type,
     Union,
+    get_args,
+    get_origin,
 )
 
 from pydantic import (
@@ -172,15 +174,29 @@ class Task(BaseModel):
         """
         if v is not None:
             sig = inspect.signature(v)
-            if len(sig.parameters) != 1:
+            positional_args = [
+                param
+                for param in sig.parameters.values()
+                if param.default is inspect.Parameter.empty
+            ]
+            if len(positional_args) != 1:
                 raise ValueError("Guardrail function must accept exactly one parameter")
 
             # Check return annotation if present, but don't require it
             return_annotation = sig.return_annotation
             if return_annotation != inspect.Signature.empty:
+
+                return_annotation_args = get_args(return_annotation)
                 if not (
-                    return_annotation == Tuple[bool, Any]
-                    or str(return_annotation) == "Tuple[bool, Any]"
+                    get_origin(return_annotation) is tuple
+                    and len(return_annotation_args) == 2
+                    and return_annotation_args[0] is bool
+                    and (
+                        return_annotation_args[1] is Any
+                        or return_annotation_args[1] is str
+                        or return_annotation_args[1] is TaskOutput
+                        or return_annotation_args[1] == Union[str, TaskOutput]
+                    )
                 ):
                     raise ValueError(
                         "If return type is annotated, it must be Tuple[bool, Any]"
@@ -435,9 +451,9 @@ class Task(BaseModel):
                 content = (
                     json_output
                     if json_output
-                    else pydantic_output.model_dump_json()
-                    if pydantic_output
-                    else result
+                    else (
+                        pydantic_output.model_dump_json() if pydantic_output else result
+                    )
                 )
                 self._save_file(content)
             crewai_event_bus.emit(self, TaskCompletedEvent(output=task_output))

src/crewai/translations/en.json

@@ -39,8 +39,8 @@
     "validation_error": "### Previous attempt failed validation: {guardrail_result_error}\n\n\n### Previous result:\n{task_output}\n\n\nTry again, making sure to address the validation error."
   },
   "tools": {
-    "delegate_work": "Delegate a specific task to one of the following coworkers: {coworkers}\nThe input to this tool should be the coworker, the task you want them to do, and ALL necessary context to execute the task, they know nothing about the task, so share absolute everything you know, don't reference things but instead explain them.",
-    "ask_question": "Ask a specific question to one of the following coworkers: {coworkers}\nThe input to this tool should be the coworker, the question you have for them, and ALL necessary context to ask the question properly, they know nothing about the question, so share absolute everything you know, don't reference things but instead explain them.",
+    "delegate_work": "Delegate a specific task to one of the following coworkers: {coworkers}\nThe input to this tool should be the coworker, the task you want them to do, and ALL necessary context to execute the task, they know nothing about the task, so share absolutely everything you know, don't reference things but instead explain them.",
+    "ask_question": "Ask a specific question to one of the following coworkers: {coworkers}\nThe input to this tool should be the coworker, the question you have for them, and ALL necessary context to ask the question properly, they know nothing about the question, so share absolutely everything you know, don't reference things but instead explain them.",
     "add_image": {
       "name": "Add image to content",
       "description": "See image to understand its content, you can optionally ask a question about the image",

src/crewai/utilities/llm_utils.py

@@ -44,6 +44,7 @@ def create_llm(
         # Extract attributes with explicit types
         model = (
             getattr(llm_value, "model_name", None)
+            or getattr(llm_value, "model", None)
             or getattr(llm_value, "deployment_name", None)
             or str(llm_value)
         )

tests/llm_test.py

@@ -6,7 +6,7 @@ import pytest
 from pydantic import BaseModel
 
 from crewai.agents.agent_builder.utilities.base_token_process import TokenProcess
-from crewai.llm import LLM
+from crewai.llm import CONTEXT_WINDOW_USAGE_RATIO, LLM
 from crewai.utilities.events import crewai_event_bus
 from crewai.utilities.events.tool_usage_events import ToolExecutionErrorEvent
 from crewai.utilities.token_counter_callback import TokenCalcHandler
@@ -285,6 +285,23 @@ def test_o3_mini_reasoning_effort_medium():
     assert isinstance(result, str)
     assert "Paris" in result
 
+def test_context_window_validation():
+    """Test that context window validation works correctly."""
+    # Test valid window size
+    llm = LLM(model="o3-mini")
+    assert llm.get_context_window_size() == int(200000 * CONTEXT_WINDOW_USAGE_RATIO)
+
+    # Test invalid window size
+    with pytest.raises(ValueError) as excinfo:
+        with patch.dict(
+            "crewai.llm.LLM_CONTEXT_WINDOW_SIZES",
+            {"test-model": 500},  # Below minimum
+            clear=True,
+        ):
+            llm = LLM(model="test-model")
+            llm.get_context_window_size()
+    assert "must be between 1024 and 2097152" in str(excinfo.value)
+
 
 @pytest.mark.vcr(filter_headers=["authorization"])
 @pytest.fixture

tests/task_test.py

@@ -1283,3 +1283,109 @@ def test_interpolate_valid_types():
     assert parsed["optional"] is None
     assert parsed["nested"]["flag"] is True
     assert parsed["nested"]["empty"] is None
+
+
+def test_guardrail_with_new_style_annotations():
+    """Test that guardrails with new-style type annotations work correctly."""
+
+    # Define a guardrail with new-style annotation
+    def guardrail(result: TaskOutput) -> tuple[bool, str]:
+        return (True, result.raw.upper())
+
+    agent = MagicMock()
+    agent.role = "test_agent"
+    agent.execute_task.return_value = "test result"
+    agent.crew = None
+
+    task = Task(description="Test task", expected_output="Output", guardrail=guardrail)
+
+    result = task.execute_sync(agent=agent)
+    assert isinstance(result, TaskOutput)
+    assert result.raw == "TEST RESULT"
+
+
+def test_guardrail_with_specific_return_type():
+    """Test that guardrails with specific return types work correctly."""
+
+    # Define a guardrail with specific return type
+    def guardrail(result: TaskOutput) -> tuple[bool, TaskOutput]:
+        if "error" in result.raw.lower():
+            return (False, "Contains error")
+        return (True, result)
+
+    agent = MagicMock()
+    agent.role = "test_agent"
+    agent.execute_task.return_value = "success result"
+    agent.crew = None
+
+    task = Task(description="Test task", expected_output="Output", guardrail=guardrail)
+
+    result = task.execute_sync(agent=agent)
+    assert isinstance(result, TaskOutput)
+    assert result.raw == "success result"
+
+
+def test_guardrail_with_positional_and_default_args():
+    """Test that guardrails with positional and default arguments work correctly."""
+
+    # Define a guardrail with a positional argument and a default argument
+    def guardrail(result: TaskOutput, optional_arg=None) -> tuple[bool, str]:
+        return (True, result.raw.upper())
+
+    agent = MagicMock()
+    agent.role = "test_agent"
+    agent.execute_task.return_value = "test result"
+    agent.crew = None
+
+    # This should now work with the updated validator
+    task = Task(description="Test task", expected_output="Output", guardrail=guardrail)
+
+    result = task.execute_sync(agent=agent)
+    assert isinstance(result, TaskOutput)
+    assert result.raw == "TEST RESULT"
+
+
+def test_guardrail_with_multiple_positional_args():
+    """Test that guardrails with multiple positional arguments are rejected."""
+
+    # Define a guardrail with multiple positional arguments
+    def guardrail(result: TaskOutput, another_required_arg) -> tuple[bool, str]:
+        return (True, result.raw.upper())
+
+    agent = MagicMock()
+    agent.role = "test_agent"
+    agent.execute_task.return_value = "test result"
+    agent.crew = None
+
+    # This should raise a ValueError because guardrail must accept exactly one positional parameter
+    with pytest.raises(ValueError) as excinfo:
+        Task(description="Test task", expected_output="Output", guardrail=guardrail)
+
+    assert "Guardrail function must accept exactly one parameter" in str(excinfo.value)
+
+
+def test_guardrail_with_positional_and_default_args():
+    """Validate that the guardrail function has the correct signature and behavior.
+
+    While type hints provide static checking, this validator ensures runtime safety by:
+    1. Verifying the function accepts exactly one required parameter (the TaskOutput)
+       (additional parameters with default values are allowed)
+    2. Checking return type annotations match Tuple[bool, Any] or tuple[bool, Any] if present
+    3. Providing clear, immediate error messages for debugging
+    """
+
+    # Define a guardrail with a positional argument and a default argument
+    def guardrail(result: TaskOutput, optional_arg=None) -> tuple[bool, str]:
+        return (True, result.raw.upper())
+
+    agent = MagicMock()
+    agent.role = "test_agent"
+    agent.execute_task.return_value = "test result"
+    agent.crew = None
+
+    # This should now work with the updated validator
+    task = Task(description="Test task", expected_output="Output", guardrail=guardrail)
+
+    result = task.execute_sync(agent=agent)
+    assert isinstance(result, TaskOutput)
+    assert result.raw == "TEST RESULT"