chore: refactor according to semantic conventions

CHANGELOG-loongsuite.md

@@ -11,6 +11,10 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## Unreleased
 
+### Changed
+
+- Adopt upstream semantic conventions for LoongSuite instrumentation packages.
+
 ## Version 0.5.0 (2026-05-11)
 
 ### Fixed

instrumentation-genai/opentelemetry-instrumentation-vertexai/src/opentelemetry/instrumentation/vertexai/patch.py

@@ -173,7 +173,7 @@ def handle_response(
                 | None,
             ) -> None:
                 attributes = (
-                    get_server_attributes(instance.api_endpoint)  # type: ignore[reportUnknownMemberType]
+                    get_server_attributes(instance.api_endpoint)
                     | request_attributes
                     | get_genai_response_attributes(response)
                 )
@@ -257,7 +257,7 @@ def _with_default_instrumentation(
         kwargs: Any,
     ):
         params = _extract_params(*args, **kwargs)
-        api_endpoint: str = instance.api_endpoint  # type: ignore[reportUnknownMemberType]
+        api_endpoint: str = instance.api_endpoint
         span_attributes = {
             **get_genai_request_attributes(False, params),
             **get_server_attributes(api_endpoint),

instrumentation-loongsuite/loongsuite-instrumentation-agentscope/src/opentelemetry/instrumentation/agentscope/patch.py

@@ -401,8 +401,11 @@ async def wrap_tool_call(wrapped, instance, args, kwargs, handler):
     matched_skill = _match_skill_for_tool(instance, tool_args)
 
     # Create invocation object with all tool data
+    # NOTE: tool_type is set to "function" as agentscope currently only
+    # supports function-type tools. Update when other types are supported.
     invocation = ExecuteToolInvocation(
         tool_name=tool_name,
+        tool_type="function",
         tool_call_id=tool_id,
         tool_description=tool_description,
         tool_call_arguments=tool_args,

instrumentation-loongsuite/loongsuite-instrumentation-agentscope/tests/test_skill_detection.py

@@ -283,6 +283,7 @@ async def fake_tool_generator():
 
         invocation = ExecuteToolInvocation(
             tool_name="read_file",
+            tool_type="function",
             skill_name="news",
             skill_id="workspace:default:news",
             skill_description="Latest news",

instrumentation-loongsuite/loongsuite-instrumentation-claude-agent-sdk/src/opentelemetry/instrumentation/claude_agent_sdk/__init__.py

@@ -102,15 +102,14 @@ def _instrument(self, **kwargs: Any) -> None:
             wrap_function_wrapper(
                 module="claude_agent_sdk",
                 name="ClaudeSDKClient.__init__",
-                wrapper=lambda wrapped,
-                instance,
-                args,
-                kwargs: wrap_claude_client_init(
-                    wrapped,
-                    instance,
-                    args,
-                    kwargs,
-                    handler=ClaudeAgentSDKInstrumentor._handler,
+                wrapper=lambda wrapped, instance, args, kwargs: (
+                    wrap_claude_client_init(
+                        wrapped,
+                        instance,
+                        args,
+                        kwargs,
+                        handler=ClaudeAgentSDKInstrumentor._handler,
+                    )
                 ),
             )
         except Exception as e:
@@ -122,15 +121,14 @@ def _instrument(self, **kwargs: Any) -> None:
             wrap_function_wrapper(
                 module="claude_agent_sdk",
                 name="ClaudeSDKClient.query",
-                wrapper=lambda wrapped,
-                instance,
-                args,
-                kwargs: wrap_claude_client_query(
-                    wrapped,
-                    instance,
-                    args,
-                    kwargs,
-                    handler=ClaudeAgentSDKInstrumentor._handler,
+                wrapper=lambda wrapped, instance, args, kwargs: (
+                    wrap_claude_client_query(
+                        wrapped,
+                        instance,
+                        args,
+                        kwargs,
+                        handler=ClaudeAgentSDKInstrumentor._handler,
+                    )
                 ),
             )
         except Exception as e:
@@ -140,15 +138,14 @@ def _instrument(self, **kwargs: Any) -> None:
             wrap_function_wrapper(
                 module="claude_agent_sdk",
                 name="ClaudeSDKClient.receive_response",
-                wrapper=lambda wrapped,
-                instance,
-                args,
-                kwargs: wrap_claude_client_receive_response(
-                    wrapped,
-                    instance,
-                    args,
-                    kwargs,
-                    handler=ClaudeAgentSDKInstrumentor._handler,
+                wrapper=lambda wrapped, instance, args, kwargs: (
+                    wrap_claude_client_receive_response(
+                        wrapped,
+                        instance,
+                        args,
+                        kwargs,
+                        handler=ClaudeAgentSDKInstrumentor._handler,
+                    )
                 ),
             )
         except Exception as e:

instrumentation-loongsuite/loongsuite-instrumentation-claude-agent-sdk/src/opentelemetry/instrumentation/claude_agent_sdk/patch.py

@@ -159,6 +159,7 @@ def _create_tool_spans_from_message(
             try:
                 tool_invocation = ExecuteToolInvocation(
                     tool_name=tool_name,
+                    tool_type="function",
                     tool_call_id=tool_use_id,
                     tool_call_arguments=tool_input,
                     tool_description=tool_name,

instrumentation-loongsuite/loongsuite-instrumentation-claw-eval/src/opentelemetry/instrumentation/claw_eval/internal/wrappers.py

@@ -62,6 +62,38 @@
 # string instead of reading it from ``gen_ai_attributes``.
 GEN_AI_TOOL_DEFINITIONS = "gen_ai.tool.definitions"
 
+
+def _infer_provider_name(provider: Any) -> str:
+    """Infer gen_ai.provider.name from the provider instance."""
+    if provider is None:
+        return "claw-eval"
+    # Try provider_name attribute
+    pn = getattr(provider, "provider_name", None)
+    if pn:
+        return str(pn)
+    # Infer from class name
+    cls_name = type(provider).__name__.lower()
+    if "openai" in cls_name:
+        return "openai"
+    if "anthropic" in cls_name:
+        return "anthropic"
+    if "dashscope" in cls_name:
+        return "dashscope"
+    if "gemini" in cls_name or "google" in cls_name:
+        return "google"
+    # Infer from model_id
+    model_id = str(getattr(provider, "model_id", "") or "")
+    if model_id.startswith(("gpt-", "o1-", "o3-", "chatgpt-")):
+        return "openai"
+    if model_id.startswith(("claude-",)):
+        return "anthropic"
+    if model_id.startswith(("qwen",)):
+        return "dashscope"
+    if model_id.startswith(("gemini",)):
+        return "google"
+    return "unknown"
+
+
 # ---------------------------------------------------------------------------
 # ContextVars for STEP lifecycle & compact-depth tracking
 # ---------------------------------------------------------------------------
@@ -514,6 +546,10 @@ def __call__(self, wrapped, instance, args, kwargs):
             span.set_attribute(GenAI.GEN_AI_AGENT_NAME, "claw-eval")
             span.set_attribute("claw_eval.task_id", str(task_id))
 
+            # Set gen_ai.provider.name (Required attribute)
+            _provider_name = _infer_provider_name(provider)
+            span.set_attribute(GenAI.GEN_AI_PROVIDER_NAME, _provider_name)
+
             model_id = ""
             if provider is not None:
                 model_id = str(getattr(provider, "model_id", "") or "")

instrumentation-loongsuite/loongsuite-instrumentation-dashscope/src/opentelemetry/instrumentation/dashscope/patch/embedding.py

@@ -52,6 +52,8 @@ def wrap_text_embedding_call(wrapped, instance, args, kwargs, handler=None):
         # Create embedding invocation object
         invocation = EmbeddingInvocation(request_model=model)
         invocation.provider = "dashscope"
+        invocation.server_address = "dashscope.aliyuncs.com"
+        invocation.server_port = 443
 
         # Extract parameters from kwargs or kwargs["parameters"] dict
         parameters = kwargs.get("parameters", {})

instrumentation-loongsuite/loongsuite-instrumentation-dashscope/src/opentelemetry/instrumentation/dashscope/utils/common.py

@@ -88,6 +88,48 @@ def _extract_usage(response: Any) -> tuple[Optional[int], Optional[int]]:
         return None, None
 
 
+def _extract_cache_tokens(
+    response: Any,
+) -> tuple[Optional[int], Optional[int]]:
+    """Extract cache token usage from DashScope response.
+
+    Args:
+        response: DashScope response object
+
+    Returns:
+        Tuple of (cache_creation_input_tokens, cache_read_input_tokens)
+    """
+    if not response:
+        return None, None
+
+    try:
+        usage = getattr(response, "usage", None)
+        if not usage:
+            return None, None
+
+        # DashScope may report cache tokens in various fields
+        cache_creation = getattr(
+            usage, "cache_creation_input_tokens", None
+        ) or getattr(usage, "cache_creation_tokens", None)
+        cache_read = (
+            getattr(usage, "cache_read_input_tokens", None)
+            or getattr(usage, "cache_read_tokens", None)
+            or getattr(usage, "prompt_cache_hit_tokens", None)
+        )
+
+        # Also check prompt_tokens_details (OpenAI-compatible format)
+        prompt_details = getattr(usage, "prompt_tokens_details", None)
+        if prompt_details and cache_read is None:
+            cache_read = getattr(prompt_details, "cached_tokens", None)
+
+        return (
+            cache_creation if cache_creation and cache_creation > 0 else None,
+            cache_read if cache_read and cache_read > 0 else None,
+        )
+    except (KeyError, AttributeError):
+        return None, None
+
+
 def _extract_task_id(task: Any) -> Optional[str]:
     """Extract task_id from task parameter (can be str or Response object).
 

instrumentation-loongsuite/loongsuite-instrumentation-dashscope/src/opentelemetry/instrumentation/dashscope/utils/generation.py

@@ -469,6 +469,10 @@ def _create_invocation_from_generation(
 
     invocation = LLMInvocation(request_model=request_model)
     invocation.provider = "dashscope"
+    # DashScope SDK always targets dashscope.aliyuncs.com; if custom endpoint
+    # support is added in the future, extract from instance/env instead.
+    invocation.server_address = "dashscope.aliyuncs.com"
+    invocation.server_port = 443
     invocation.input_messages = _extract_input_messages(kwargs)
 
     # Extract tool definitions and convert to FunctionToolDefinition objects
@@ -568,6 +572,15 @@ def _update_invocation_from_response(
         invocation.input_tokens = input_tokens
         invocation.output_tokens = output_tokens
 
+        # Extract cache token usage
+        from ..utils.common import _extract_cache_tokens  # noqa: PLC0415
+
+        cache_creation, cache_read = _extract_cache_tokens(response)
+        if cache_creation is not None:
+            invocation.usage_cache_creation_input_tokens = cache_creation
+        if cache_read is not None:
+            invocation.usage_cache_read_input_tokens = cache_read
+
         # Extract response model name (if available)
         response_model = _safe_get(response, "model")
         if response_model:

instrumentation-loongsuite/loongsuite-instrumentation-dashscope/tests/test_embedding.py

@@ -141,7 +141,7 @@ def test_text_embedding_basic(instrument, span_exporter):
     """Test basic text embedding call."""
 
     response = TextEmbedding.call(
-        model="text-embedding-v1", input="Hello, world!"
+        model="text-embedding-v4", input="Hello, world!"
     )
 
     assert response is not None
@@ -165,7 +165,7 @@ def test_text_embedding_basic(instrument, span_exporter):
     # Assert all span attributes
     _assert_embedding_span_attributes(
         span,
-        request_model="text-embedding-v1",
+        request_model="text-embedding-v4",
         response=response,
         input_tokens=input_tokens,
     )
@@ -178,7 +178,7 @@ def test_text_embedding_batch(instrument, span_exporter):
     """Test text embedding with batch input."""
 
     response = TextEmbedding.call(
-        model="text-embedding-v1", input=["Hello", "World"]
+        model="text-embedding-v4", input=["Hello", "World"]
     )
 
     assert response is not None
@@ -202,7 +202,7 @@ def test_text_embedding_batch(instrument, span_exporter):
     # Assert all span attributes
     _assert_embedding_span_attributes(
         span,
-        request_model="text-embedding-v1",
+        request_model="text-embedding-v4",
         response=response,
         input_tokens=input_tokens,
     )
@@ -215,7 +215,7 @@ def test_text_embedding_with_text_type(instrument, span_exporter):
     """Test text embedding with text_type parameter."""
 
     response = TextEmbedding.call(
-        model="text-embedding-v1",
+        model="text-embedding-v4",
         input="What is machine learning?",
         text_type="query",
     )
@@ -241,7 +241,7 @@ def test_text_embedding_with_text_type(instrument, span_exporter):
     # Assert all span attributes
     _assert_embedding_span_attributes(
         span,
-        request_model="text-embedding-v1",
+        request_model="text-embedding-v4",
         response=response,
         input_tokens=input_tokens,
     )
@@ -254,7 +254,7 @@ def test_text_embedding_with_dimension(instrument, span_exporter):
     """Test text embedding with dimension parameter."""
 
     response = TextEmbedding.call(
-        model="text-embedding-v1",
+        model="text-embedding-v4",
         input="What is machine learning?",
         dimension=512,
     )
@@ -280,7 +280,7 @@ def test_text_embedding_with_dimension(instrument, span_exporter):
     # Assert all span attributes including dimension_count
     _assert_embedding_span_attributes(
         span,
-        request_model="text-embedding-v1",
+        request_model="text-embedding-v4",
         response=response,
         input_tokens=input_tokens,
         dimension_count=512,  # Should be captured from request

instrumentation-loongsuite/loongsuite-instrumentation-google-adk/src/opentelemetry/instrumentation/google_adk/internal/_plugin.py

@@ -119,6 +119,7 @@ async def before_run_callback(
             invocation = InvokeAgentInvocation(
                 provider="google_adk",
                 agent_name=invocation_context.app_name,
+                agent_id=invocation_context.app_name,
             )
 
             # Set conversation_id if available
@@ -275,6 +276,7 @@ async def before_agent_callback(
             invocation = InvokeAgentInvocation(
                 provider="google_adk",
                 agent_name=agent.name,
+                agent_id=agent.name,
             )
 
             # Set agent attributes
@@ -508,6 +510,7 @@ async def before_tool_callback(
             # Create invocation object
             invocation = ExecuteToolInvocation(
                 tool_name=tool.name,
+                tool_type="function",
                 provider="google_adk",
             )