Merge branch 'master' into webb/toxfix

Author: alexander-alderman-webb

.github/workflows/ai-integration-test.yml

@@ -34,7 +34,7 @@ jobs:
           token: ${{ secrets.GITHUB_TOKEN }}
 
       - name: Run Python SDK Tests
-        uses: getsentry/testing-ai-sdk-integrations@121da677853244cedfe11e95184b2b431af102eb
+        uses: getsentry/testing-ai-sdk-integrations@285c012e522f241581534dfc89bd99ec3b1da4f6
         env:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
         with:

sentry_sdk/integrations/pydantic_ai/patches/tools.py

@@ -50,6 +50,7 @@ async def wrapped_execute_tool_call(
         call = validated.call
         name = call.tool_name
         tool = self.tools.get(name) if self.tools else None
+        selected_tool_definition = getattr(tool, "tool_def", None)
 
         # Determine tool type by checking tool.toolset
         tool_type = "function"
@@ -73,6 +74,7 @@ async def wrapped_execute_tool_call(
                     args_dict,
                     agent,
                     tool_type=tool_type,
+                    tool_definition=selected_tool_definition,
                 ) as span:
                     try:
                         result = await original_execute_tool_call(
@@ -127,6 +129,7 @@ async def wrapped_call_tool(
         # Extract tool info before calling original
         name = call.tool_name
         tool = self.tools.get(name) if self.tools else None
+        selected_tool_definition = getattr(tool, "tool_def", None)
 
         # Determine tool type by checking tool.toolset
         tool_type = "function"  # default
@@ -150,6 +153,7 @@ async def wrapped_call_tool(
                     args_dict,
                     agent,
                     tool_type=tool_type,
+                    tool_definition=selected_tool_definition,
                 ) as span:
                     try:
                         result = await original_call_tool(

sentry_sdk/integrations/pydantic_ai/spans/execute_tool.py

@@ -9,10 +9,15 @@
 
 if TYPE_CHECKING:
     from typing import Any, Optional
+    from pydantic_ai._tool_manager import ToolDefinition  # type: ignore
 
 
 def execute_tool_span(
-    tool_name: str, tool_args: "Any", agent: "Any", tool_type: str = "function"
+    tool_name: str,
+    tool_args: "Any",
+    agent: "Any",
+    tool_type: str = "function",
+    tool_definition: "Optional[ToolDefinition]" = None,
 ) -> "sentry_sdk.tracing.Span":
     """Create a span for tool execution.
 
@@ -21,6 +26,7 @@ def execute_tool_span(
         tool_args: The arguments passed to the tool
         agent: The agent executing the tool
         tool_type: The type of tool ("function" for regular tools, "mcp" for MCP services)
+        tool_definition: The definition of the tool, if available
     """
     span = sentry_sdk.start_span(
         op=OP.GEN_AI_EXECUTE_TOOL,
@@ -32,6 +38,12 @@ def execute_tool_span(
     span.set_data(SPANDATA.GEN_AI_TOOL_TYPE, tool_type)
     span.set_data(SPANDATA.GEN_AI_TOOL_NAME, tool_name)
 
+    if tool_definition is not None and hasattr(tool_definition, "description"):
+        span.set_data(
+            SPANDATA.GEN_AI_TOOL_DESCRIPTION,
+            tool_definition.description,
+        )
+
     _set_agent_data(span, agent)
 
     if _should_send_prompts() and tool_args is not None:

tests/integrations/openai_agents/test_openai_agents.py

@@ -1231,7 +1231,7 @@ def simple_test_tool(message: str) -> str:
                 "type": "object",
                 "additionalProperties": False,
             },
-            "on_invoke_tool": "<function agents.tool.function_tool.<locals>._create_function_tool.<locals>._on_invoke_tool>",
+            "on_invoke_tool": mock.ANY,
             "strict_json_schema": True,
             "is_enabled": True,
         }
@@ -1259,16 +1259,17 @@ def simple_test_tool(message: str) -> str:
             }
         )
 
-    available_tools = safe_serialize(available_tools)
-
     assert transaction["transaction"] == "test_agent workflow"
     assert transaction["contexts"]["trace"]["origin"] == "auto.ai.openai_agents"
 
     assert agent_span["description"] == "invoke_agent test_agent"
     assert agent_span["origin"] == "auto.ai.openai_agents"
     assert agent_span["data"]["gen_ai.agent.name"] == "test_agent"
     assert agent_span["data"]["gen_ai.operation.name"] == "invoke_agent"
-    assert agent_span["data"]["gen_ai.request.available_tools"] == available_tools
+    assert (
+        json.loads(agent_span["data"]["gen_ai.request.available_tools"])
+        == available_tools
+    )
     assert agent_span["data"]["gen_ai.request.max_tokens"] == 100
     assert agent_span["data"]["gen_ai.request.model"] == "gpt-4"
     assert agent_span["data"]["gen_ai.request.temperature"] == 0.7
@@ -1279,7 +1280,10 @@ def simple_test_tool(message: str) -> str:
     assert ai_client_span1["data"]["gen_ai.operation.name"] == "chat"
     assert ai_client_span1["data"]["gen_ai.system"] == "openai"
     assert ai_client_span1["data"]["gen_ai.agent.name"] == "test_agent"
-    assert ai_client_span1["data"]["gen_ai.request.available_tools"] == available_tools
+    assert (
+        json.loads(ai_client_span1["data"]["gen_ai.request.available_tools"])
+        == available_tools
+    )
     assert ai_client_span1["data"]["gen_ai.request.max_tokens"] == 100
     assert ai_client_span1["data"]["gen_ai.request.messages"] == safe_serialize(
         [
@@ -1320,11 +1324,7 @@ def simple_test_tool(message: str) -> str:
     assert tool_span["data"]["gen_ai.agent.name"] == "test_agent"
     assert tool_span["data"]["gen_ai.operation.name"] == "execute_tool"
     assert (
-        re.sub(
-            "<.*>(,)",
-            r"'NOT_CHECKED'\1",
-            agent_span["data"]["gen_ai.request.available_tools"],
-        )
+        json.loads(agent_span["data"]["gen_ai.request.available_tools"])
         == available_tools
     )
     assert tool_span["data"]["gen_ai.request.max_tokens"] == 100
@@ -1342,11 +1342,7 @@ def simple_test_tool(message: str) -> str:
     assert ai_client_span2["data"]["gen_ai.agent.name"] == "test_agent"
     assert ai_client_span2["data"]["gen_ai.operation.name"] == "chat"
     assert (
-        re.sub(
-            "<.*>(,)",
-            r"'NOT_CHECKED'\1",
-            agent_span["data"]["gen_ai.request.available_tools"],
-        )
+        json.loads(agent_span["data"]["gen_ai.request.available_tools"])
         == available_tools
     )
     assert ai_client_span2["data"]["gen_ai.request.max_tokens"] == 100

tests/integrations/pydantic_ai/test_pydantic_ai.py

@@ -16,7 +16,6 @@
 from pydantic_ai import Agent
 from pydantic_ai.messages import BinaryContent, UserPromptPart
 from pydantic_ai.usage import RequestUsage
-from pydantic_ai.models.test import TestModel
 from pydantic_ai.exceptions import ModelRetry, UnexpectedModelBehavior
 
 
@@ -2386,7 +2385,9 @@ async def test_execute_tool_span_with_mcp_type(sentry_init, capture_events):
     Test execute_tool span with MCP tool type.
     """
     import sentry_sdk
-    from sentry_sdk.integrations.pydantic_ai.spans.execute_tool import execute_tool_span
+    from sentry_sdk.integrations.pydantic_ai.spans.execute_tool import (
+        execute_tool_span,
+    )
 
     sentry_init(
         integrations=[PydanticAIIntegration()],
@@ -2794,3 +2795,42 @@ async def test_set_usage_data_with_cache_tokens(sentry_init, capture_events):
     (span_data,) = event["spans"]
     assert span_data["data"][SPANDATA.GEN_AI_USAGE_INPUT_TOKENS_CACHED] == 80
     assert span_data["data"][SPANDATA.GEN_AI_USAGE_INPUT_TOKENS_CACHE_WRITE] == 20
+
+
+@pytest.mark.asyncio
+async def test_tool_description_in_execute_tool_span(sentry_init, capture_events):
+    """
+    Test that tool description from the tool's docstring is included in execute_tool spans.
+    """
+    agent = Agent(
+        "test",
+        name="test_agent",
+        system_prompt="You are a helpful test assistant.",
+    )
+
+    @agent.tool_plain
+    def multiply_numbers(a: int, b: int) -> int:
+        """Multiply two numbers and return the product."""
+        return a * b
+
+    sentry_init(
+        integrations=[PydanticAIIntegration()],
+        traces_sample_rate=1.0,
+        send_default_pii=True,
+    )
+
+    events = capture_events()
+
+    result = await agent.run("What is 5 times 3?")
+    assert result is not None
+
+    (transaction,) = events
+    spans = transaction["spans"]
+
+    tool_spans = [s for s in spans if s["op"] == "gen_ai.execute_tool"]
+    assert len(tool_spans) >= 1
+
+    tool_span = tool_spans[0]
+    assert tool_span["data"]["gen_ai.tool.name"] == "multiply_numbers"
+    assert SPANDATA.GEN_AI_TOOL_DESCRIPTION in tool_span["data"]
+    assert "Multiply two numbers" in tool_span["data"][SPANDATA.GEN_AI_TOOL_DESCRIPTION]

tests/integrations/pyramid/test_pyramid.py

@@ -6,6 +6,7 @@
 import pytest
 from pyramid.authorization import ACLAuthorizationPolicy
 from pyramid.response import Response
+from packaging.version import Version
 from werkzeug.test import Client
 
 from sentry_sdk import capture_message, add_breadcrumb
@@ -18,7 +19,7 @@
 try:
     from importlib.metadata import version
 
-    PYRAMID_VERSION = tuple(map(int, version("pyramid").split(".")))
+    PYRAMID_VERSION = Version(version("pyramid")).release
 
 except ImportError:
     # < py3.8