Enhance event handling for tool usage and agent execution

- Add new events for tool usage: ToolSelectionErrorEvent, ToolValidateInputErrorEvent
- Improve error tracking and event emission in ToolUsage and LLM classes
- Update AgentExecutionStartedEvent to use task_prompt instead of inputs
- Add comprehensive test coverage for new event types and error scenarios
This commit is contained in:
Lorenze Jay
2025-02-18 14:13:18 -08:00
parent e9dc68723f
commit 1b5cc08abe
12 changed files with 531 additions and 120 deletions

View File

@@ -7,7 +7,8 @@ from pydantic import BaseModel
from crewai.agents.agent_builder.utilities.base_token_process import TokenProcess
from crewai.llm import LLM
from crewai.tools import tool
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
@@ -291,32 +292,36 @@ def anthropic_llm():
"""Fixture providing an Anthropic LLM instance."""
return LLM(model="anthropic/claude-3-sonnet")
@pytest.fixture
def system_message():
"""Fixture providing a system message."""
return {"role": "system", "content": "test"}
@pytest.fixture
def user_message():
"""Fixture providing a user message."""
return {"role": "user", "content": "test"}
def test_anthropic_message_formatting_edge_cases(anthropic_llm):
"""Test edge cases for Anthropic message formatting."""
# Test None messages
with pytest.raises(TypeError, match="Messages cannot be None"):
anthropic_llm._format_messages_for_provider(None)
# Test empty message list
formatted = anthropic_llm._format_messages_for_provider([])
assert len(formatted) == 1
assert formatted[0]["role"] == "user"
assert formatted[0]["content"] == "."
# Test invalid message format
with pytest.raises(TypeError, match="Invalid message format"):
anthropic_llm._format_messages_for_provider([{"invalid": "message"}])
def test_anthropic_model_detection():
"""Test Anthropic model detection with various formats."""
models = [
@@ -327,11 +332,12 @@ def test_anthropic_model_detection():
("", False),
("anthropomorphic", False), # Should not match partial words
]
for model, expected in models:
llm = LLM(model=model)
assert llm.is_anthropic == expected, f"Failed for model: {model}"
def test_anthropic_message_formatting(anthropic_llm, system_message, user_message):
"""Test Anthropic message formatting with fixtures."""
# Test when first message is system
@@ -371,3 +377,51 @@ def test_deepseek_r1_with_open_router():
result = llm.call("What is the capital of France?")
assert isinstance(result, str)
assert "Paris" in result
@pytest.mark.vcr(filter_headers=["authorization"])
def test_tool_execution_error_event():
llm = LLM(model="gpt-4o-mini")
def failing_tool(param: str) -> str:
"""This tool always fails."""
raise Exception("Tool execution failed!")
tool_schema = {
"type": "function",
"function": {
"name": "failing_tool",
"description": "This tool always fails.",
"parameters": {
"type": "object",
"properties": {
"param": {"type": "string", "description": "A test parameter"}
},
"required": ["param"],
},
},
}
received_events = []
@crewai_event_bus.on(ToolExecutionErrorEvent)
def event_handler(source, event):
received_events.append(event)
available_functions = {"failing_tool": failing_tool}
messages = [{"role": "user", "content": "Use the failing tool"}]
llm.call(
messages,
tools=[tool_schema],
available_functions=available_functions,
)
assert len(received_events) == 1
event = received_events[0]
assert isinstance(event, ToolExecutionErrorEvent)
assert event.tool_name == "failing_tool"
assert event.tool_args == {"param": "test"}
assert event.tool_class == failing_tool
assert "Tool execution failed!" in event.error