feat(genai): align util telemetry with upstream semantic conventions

instrumentation-loongsuite/loongsuite-instrumentation-agentscope/CHANGELOG.md

@@ -7,6 +7,11 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## Unreleased
 
+### Changed
+
+- Align streaming timing, reasoning-token usage, provider names, and agent
+  version attributes with the current upstream GenAI conventions.
+
 ### Fixed
 
 - Start AgentScope v2 streaming LLM spans before invoking the underlying model

instrumentation-loongsuite/loongsuite-instrumentation-agentscope/src/opentelemetry/instrumentation/agentscope/_v2_middleware.py

@@ -20,7 +20,13 @@
 import json
 import logging
 import timeit
-from collections.abc import AsyncGenerator, Awaitable, Callable, Sequence
+from collections.abc import (
+    AsyncGenerator,
+    Awaitable,
+    Callable,
+    Mapping,
+    Sequence,
+)
 from contextvars import ContextVar
 from dataclasses import asdict, is_dataclass
 from typing import Any
@@ -179,6 +185,7 @@ async def on_model_call(
         try:
             result = await next_handler(**input_kwargs)
             if inspect.isasyncgen(result):
+                invocation.stream = True
                 return self._wrap_model_stream(
                     result,
                     invocation,
@@ -201,14 +208,18 @@ async def _wrap_model_stream(
         invocation: LLMInvocation,
         handler: ExtendedTelemetryHandler,
     ) -> AsyncGenerator[ChatResponse, None]:
-        first_token_seen = False
         last_chunk = None
         closed = False
         try:
             async for chunk in result:
-                if not first_token_seen:
-                    invocation.monotonic_first_token_s = timeit.default_timer()
-                    first_token_seen = True
+                now = timeit.default_timer()
+                if invocation.monotonic_first_chunk_s is None:
+                    invocation.monotonic_first_chunk_s = now
+                if (
+                    invocation.monotonic_first_token_s is None
+                    and _chat_response_has_token(chunk)
+                ):
+                    invocation.monotonic_first_token_s = now
                 last_chunk = chunk
                 yield chunk
         except BaseException as exc:
@@ -324,6 +335,7 @@ def _create_llm_invocation(
         provider=_get_provider_name(model),
         input_messages=_messages_to_inputs(input_kwargs.get("messages")),
         tool_definitions=_tool_definitions(input_kwargs.get("tools")),
+        stream=True if _is_streaming_model(model, input_kwargs) else None,
     )
     parameters = getattr(model, "parameters", None)
     for source in (parameters, input_kwargs):
@@ -461,6 +473,30 @@ def _is_first_token_event(item: Any) -> bool:
     return event_type in _FIRST_TOKEN_EVENT_TYPES
 
 
+def _chat_response_has_token(response: Any) -> bool:
+    content = _safe_get(response, "content")
+    if isinstance(content, str):
+        return bool(content)
+    for block in content or []:
+        block_type = _safe_get(block, "type")
+        if block_type == "text" and _safe_get(block, "text"):
+            return True
+        if block_type == "thinking" and _safe_get(block, "thinking"):
+            return True
+        if block_type == "tool_call":
+            return True
+    return False
+
+
+def _safe_get(value: Any, name: str) -> Any:
+    if isinstance(value, Mapping):
+        return value.get(name)
+    try:
+        return getattr(value, name, None)
+    except (AttributeError, KeyError):
+        return None
+
+
 def _middleware_arg_position() -> int | None:
     try:
         parameters = list(inspect.signature(Agent.__init__).parameters)

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

@@ -60,6 +60,7 @@
 import threading
 import timeit
 import weakref
+from collections.abc import Mapping
 from dataclasses import dataclass, field
 from functools import wraps
 from typing import Any, AsyncGenerator, MutableMapping
@@ -80,6 +81,31 @@
 
 logger = logging.getLogger(__name__)
 
+
+def _safe_get(value: Any, name: str) -> Any:
+    if isinstance(value, Mapping):
+        return value.get(name)
+    try:
+        return getattr(value, name, None)
+    except (AttributeError, KeyError):
+        return None
+
+
+def _stream_chunk_has_token(chunk: Any) -> bool:
+    content = _safe_get(chunk, "content")
+    if isinstance(content, str):
+        return bool(content)
+    for block in content or []:
+        block_type = _safe_get(block, "type")
+        if block_type == "text" and _safe_get(block, "text"):
+            return True
+        if block_type == "thinking" and _safe_get(block, "thinking"):
+            return True
+        if block_type == "tool_call":
+            return True
+    return bool(_safe_get(chunk, "tool_calls"))
+
+
 # Fixed hook name shared by all concurrent invocations on a given agent.
 # Using a stable name (rather than a per-call uuid) lets AgentScope's own
 # registry naturally deduplicate, so a single set of callbacks fires for
@@ -392,12 +418,16 @@ async def _wrap_streaming_response(
         """Wrap streaming response to update invocation when done."""
         try:
             last_chunk = None
-            first_token_received = False
             async for chunk in generator:
-                # Record time when first token is received
-                if not first_token_received:
-                    first_token_received = True
-                    invocation.monotonic_first_token_s = timeit.default_timer()
+                now = timeit.default_timer()
+                invocation.stream = True
+                if invocation.monotonic_first_chunk_s is None:
+                    invocation.monotonic_first_chunk_s = now
+                if (
+                    invocation.monotonic_first_token_s is None
+                    and _stream_chunk_has_token(chunk)
+                ):
+                    invocation.monotonic_first_token_s = now
 
                 last_chunk = chunk
                 yield chunk

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

@@ -674,7 +674,7 @@ def get_message_converter(
     elif provide_name == "ollama":  # AgentScopeGenAiProviderName.OLLAMA.value
         parser = OllamaMessageParser()
     elif (
-        provide_name == "moonshot"
+        provide_name == "moonshot_ai"
     ):  # AgentScopeGenAiProviderName.MOONSHOT.value
         # Moonshot uses OpenAI-compatible API format
         parser = OpenAIMessageParser()

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

@@ -69,7 +69,7 @@ class AgentScopeGenAiProviderName(str, Enum):
 
     OLLAMA = "ollama"
     DASHSCOPE = "dashscope"
-    MOONSHOT = "moonshot"
+    MOONSHOT = "moonshot_ai"
 
 
 # Provider name mapping based on class names
@@ -79,6 +79,7 @@ class AgentScopeGenAiProviderName(str, Enum):
     "anthropic": GenAIAttributes.GenAiProviderNameValues.ANTHROPIC.value,
     "dashscope": AgentScopeGenAiProviderName.DASHSCOPE.value,
     "ollama": AgentScopeGenAiProviderName.OLLAMA.value,
+    "moonshot": AgentScopeGenAiProviderName.MOONSHOT.value,
 }
 
 # Base URL to provider mapping for OpenAI-compatible APIs
@@ -89,6 +90,7 @@ class AgentScopeGenAiProviderName(str, Enum):
         GenAIAttributes.GenAiProviderNameValues.DEEPSEEK.value,
     ),
     ("dashscope.aliyuncs.com", AgentScopeGenAiProviderName.DASHSCOPE.value),
+    ("api.moonshot.cn", AgentScopeGenAiProviderName.MOONSHOT.value),
 ]
 
 

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

@@ -37,6 +37,7 @@
     create_embedding_invocation,
     create_llm_invocation,
     entry_baggage_identity_attributes,
+    get_provider_name,
 )
 from opentelemetry.util.genai.extended_semconv.gen_ai_extended_attributes import (
     GEN_AI_SESSION_ID,
@@ -47,6 +48,18 @@
 
 
 class TestUtils:
+    def test_moonshot_provider_uses_upstream_value(self):
+        class MoonshotChatModel:
+            pass
+
+        assert get_provider_name(MoonshotChatModel()) == "moonshot_ai"
+
+    def test_moonshot_compatible_base_url_uses_upstream_value(self):
+        class DashScopeChatModel:
+            base_http_api_url = "https://api.moonshot.cn/v1"
+
+        assert get_provider_name(DashScopeChatModel()) == "moonshot_ai"
+
     def test_convert_msg_with_tool_result(self):
         """Test conversion of AgentScope Msg with ToolResultBlock (End-to-End)"""
         # Construct a Msg object simulating a tool execution result

instrumentation-loongsuite/loongsuite-instrumentation-agno/CHANGELOG.md

@@ -9,6 +9,8 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ### Changed
 
+- Align streaming timing and response extraction with the current upstream
+  GenAI conventions.
 - Split Agno 2.x tool-call loops into ReAct step spans with one LLM span per
   provider request, and populate streaming agent token usage from child LLM
   spans when Agno run events do not include metrics.

instrumentation-loongsuite/loongsuite-instrumentation-agno/src/opentelemetry/instrumentation/agno/_wrapper.py

@@ -81,6 +81,37 @@ def _is_stream_close(exc: BaseException) -> bool:
     return isinstance(exc, (GeneratorExit, StopIteration, StopAsyncIteration))
 
 
+def _safe_get(value: Any, name: str) -> Any:
+    if isinstance(value, Mapping):
+        return value.get(name)
+    try:
+        return getattr(value, name, None)
+    except (AttributeError, KeyError):
+        return None
+
+
+def _response_has_token(response: Any) -> bool:
+    return any(
+        (
+            _safe_get(response, "content"),
+            _safe_get(response, "reasoning_content"),
+            _safe_get(response, "redacted_reasoning_content"),
+            _safe_get(response, "tool_calls"),
+        )
+    )
+
+
+def _mark_llm_stream_item(invocation: Any, response: Any) -> None:
+    now = timeit.default_timer()
+    invocation.stream = True
+    if invocation.monotonic_first_chunk_s is None:
+        invocation.monotonic_first_chunk_s = now
+    if invocation.monotonic_first_token_s is None and _response_has_token(
+        response
+    ):
+        invocation.monotonic_first_token_s = now
+
+
 @dataclass
 class _AgnoRunState:
     handler: ExtendedTelemetryHandler
@@ -698,8 +729,7 @@ def process_response_stream(
         try:
             stream = wrapped(*args, **kwargs)
             for response in stream:
-                if invocation.monotonic_first_token_s is None:
-                    invocation.monotonic_first_token_s = timeit.default_timer()
+                _mark_llm_stream_item(invocation, response)
                 responses.append(response)
                 yield response
             self._finish_llm_call(
@@ -744,8 +774,7 @@ async def aprocess_response_stream(
             if inspect.isawaitable(stream):
                 stream = await stream
             async for response in stream:
-                if invocation.monotonic_first_token_s is None:
-                    invocation.monotonic_first_token_s = timeit.default_timer()
+                _mark_llm_stream_item(invocation, response)
                 responses.append(response)
                 yield response
             self._finish_llm_call(
@@ -886,8 +915,7 @@ def response_stream(
         try:
             stream = wrapped(*args, **kwargs)
             for response in stream:
-                if invocation.monotonic_first_token_s is None:
-                    invocation.monotonic_first_token_s = timeit.default_timer()
+                _mark_llm_stream_item(invocation, response)
                 responses.append(response)
                 yield response
             if responses:
@@ -967,8 +995,7 @@ async def aresponse_stream(
             if inspect.isawaitable(stream):
                 stream = await stream
             async for response in stream:
-                if invocation.monotonic_first_token_s is None:
-                    invocation.monotonic_first_token_s = timeit.default_timer()
+                _mark_llm_stream_item(invocation, response)
                 responses.append(response)
                 yield response
             if responses:

instrumentation-loongsuite/loongsuite-instrumentation-autogen/CHANGELOG.md

@@ -9,6 +9,8 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ### Fixed
 
+- Capture time to first chunk independently from time to first token, and emit
+  the stream request indicator only when AutoGen enables model streaming.
 - Avoid enriching non-AutoGen spans that use overlapping GenAI operation names.
 
 ## Version 0.7.0 (2026-07-03)

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

@@ -464,15 +464,20 @@ def _direct_model_create_stream_wrapper(wrapped, instance, args, kwargs):  # typ
     async def _generator():  # type: ignore[no-untyped-def]
         handler = _get_handler()
         invocation = _make_direct_model_invocation(instance, args, kwargs)
+        invocation.stream = True
         handler.start_llm(invocation)
         _mark_autogen_live_span(invocation)
         try:
             async for item in wrapped(*args, **kwargs):
+                now = timeit.default_timer()
+                if invocation.monotonic_first_chunk_s is None:
+                    invocation.monotonic_first_chunk_s = now
                 if (
                     isinstance(item, str)
+                    and bool(item)
                     and invocation.monotonic_first_token_s is None
                 ):
-                    invocation.monotonic_first_token_s = timeit.default_timer()
+                    invocation.monotonic_first_token_s = now
                 elif type(item).__name__ == "CreateResult":
                     apply_create_result(invocation, item)
                 yield item
@@ -523,6 +528,11 @@ async def _generator():  # type: ignore[no-untyped-def]
             agent_name=agent_name,
             output_type=output_type,
         )
+        invocation.stream = (
+            True
+            if _arg_value(args, kwargs, "model_client_stream") is True
+            else None
+        )
         handler.start_llm(invocation)
         _mark_autogen_live_span(invocation)
         try:
@@ -531,11 +541,15 @@ async def _generator():  # type: ignore[no-untyped-def]
                 _suppress_direct_model_span,
                 True,
             ):
+                now = timeit.default_timer()
+                if invocation.monotonic_first_chunk_s is None:
+                    invocation.monotonic_first_chunk_s = now
                 if (
                     type(item).__name__ == "ModelClientStreamingChunkEvent"
+                    and bool(getattr(item, "content", None))
                     and invocation.monotonic_first_token_s is None
                 ):
-                    invocation.monotonic_first_token_s = timeit.default_timer()
+                    invocation.monotonic_first_token_s = now
                 elif type(item).__name__ == "CreateResult":
                     apply_create_result(invocation, item)
                 yield item