{"repo": "alibaba/loongsuite-python", "files": {"instrumentation-loongsuite/loongsuite-instrumentation-agentscope/src/opentelemetry/instrumentation/agentscope/utils.py": {"0d88889b2d45e96ad8bd14cf517bc109be900bde": {"t": 1785745990.1402972, "n": 22548, "h": "dd4203458d5271d3", "c": "# Copyright The OpenTelemetry Authors\n#\n# Licensed under the Apache License, Version 2.0 (the \"License\");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n#     http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an \"AS IS\" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n\n# -*- coding: utf-8 -*-\n\"\"\"Attributes processor for span attributes.\"\"\"\n\nfrom __future__ import annotations\n\nimport datetime\nimport enum\nimport inspect\nimport json\nimport logging\nimport mimetypes\nfrom dataclasses import is_dataclass\nfrom enum import Enum\nfrom typing import Any, Dict, List, Optional, Tuple\nfrom urllib.parse import urlparse\n\nfrom agentscope import _config\nfrom agentscope.agent import AgentBase\nfrom agentscope.embedding import EmbeddingModelBase\nfrom agentscope.message import Msg\nfrom agentscope.model import ChatModelBase, ChatResponse\nfrom pydantic import BaseModel\n\nfrom opentelemetry import baggage\nfrom opentelemetry.semconv._incubating.attributes import (\n    gen_ai_attributes as GenAIAttributes,\n)\nfrom opentelemetry.util.genai.extended_semconv.gen_ai_extended_attributes import (\n    GEN_AI_SESSION_ID,\n    GEN_AI_USER_ID,\n)\nfrom opentelemetry.util.genai.extended_types import (\n    EmbeddingInvocation,\n    InvokeAgentInvocation,\n)\nfrom opentelemetry.util.genai.types import (\n    Base64Blob,\n    InputMessage,\n    LLMInvocation,\n    OutputMessage,\n    Reasoning,\n    Text,\n    ToolCall,\n    ToolCallResponse,\n    Uri,\n)\n\nfrom .message_converter import get_message_converter\n\nlogger = logging.getLogger(__name__)\n\n\nclass AgentScopeGenAiProviderName(str, Enum):\n    \"\"\"Extended provider names not in standard OpenTelemetry semantic conventions.\"\"\"\n\n    OLLAMA = \"ollama\"\n    DASHSCOPE = \"dashscope\"\n    MOONSHOT = \"moonshot\"\n\n\n# Provider name mapping based on class names\n_PROVIDER_NAME_MAP = {\n    \"openai\": GenAIAttributes.GenAiProviderNameValues.OPENAI.value,\n    \"gemini\": GenAIAttributes.GenAiProviderNameValues.GCP_GEMINI.value,\n    \"anthropic\": GenAIAttributes.GenAiProviderNameValues.ANTHROPIC.value,\n    \"dashscope\": AgentScopeGenAiProviderName.DASHSCOPE.value,\n    \"ollama\": AgentScopeGenAiProviderName.OLLAMA.value,\n}\n\n# Base URL to provider mapping for OpenAI-compatible APIs\n_BASE_URL_PROVIDER_MAP = [\n    (\"openai.com\", GenAIAttributes.GenAiProviderNameValues.OPENAI.value),\n    (\n        \"api.deepseek.com\",\n        GenAIAttributes.GenAiProviderNameValues.DEEPSEEK.value,\n    ),\n    (\"dashscope.aliyuncs.com\", AgentScopeGenAiProviderName.DASHSCOPE.value),\n]\n\n\ndef _current_baggage_value(key: str) -> str | None:\n    try:\n        value = baggage.get_baggage(key)\n    except Exception:\n        return None\n    if value is None:\n        return None\n    text = str(value).strip()\n    return text or None\n\n\ndef entry_baggage_identity_attributes() -> dict[str, str]:\n    \"\"\"Return entry-level identity from current OpenTelemetry Baggage.\n\n    QwenPaw opens an Entry span before AgentScope runs and writes\n    ``gen_ai.session.id`` / ``gen_ai.user.id`` into Baggage. AgentScope has\n    its own ``run_id``; when both instrumentations are active, entry baggage is\n    the request-level identity and should color downstream spans.\n    \"\"\"\n    attributes: dict[str, str] = {}\n    session_id = _current_baggage_value(GEN_AI_SESSION_ID)\n    user_id = _current_baggage_value(GEN_AI_USER_ID)\n    if session_id:\n        attributes[GEN_AI_SESSION_ID] = session_id\n    if user_id:\n        attributes[GEN_AI_USER_ID] = user_id\n    return attributes\n\n\ndef apply_entry_baggage_identity(invocation: Any) -> str | None:\n    \"\"\"Copy entry-level identity baggage onto a GenAI invocation.\"\"\"\n    attributes = entry_baggage_identity_attributes()\n    for key, value in attributes.items():\n        invocation.attributes.setdefault(key, value)\n    return attributes.get(GEN_AI_SESSION_ID)\n\n\ndef get_provider_name(chat_model: ChatModelBase) -> str:\n    \"\"\"Parse chat model provider name\"\"\"\n    classname = chat_model.__class__.__name__\n    prefix = classname.removesuffix(\"ChatModel\").lower()\n\n    # Special handling for DashScopeChatModel with base_http_api_url\n    if (\n        prefix == \"dashscope\"\n        and hasattr(chat_model, \"base_http_api_url\")\n        and chat_model.base_http_api_url\n    ):\n        base_url = chat_model.base_http_api_url\n        for url_fragment, provider in _BASE_URL_PROVIDER_MAP:\n            if url_fragment in base_url:\n                return provider\n\n    return _PROVIDER_NAME_MAP.get(prefix, \"unknown\")\n\n\ndef get_embedding_provider_name(embedding_model: EmbeddingModelBase) -> str:\n    \"\"\"Parse embedding model provider name\"\"\"\n    class_name = embedding_model.__class__.__name__\n\n    if \"DashScope\" in class_name:\n        return AgentScopeGenAiProviderName.DASHSCOPE.value\n    elif \"OpenAI\" in class_name:\n        return GenAIAttributes.GenAiProviderNameValues.OPENAI.value\n    elif \"Gemini\" in class_name:\n        return GenAIAttributes.GenAiProviderNameValues.GCP_GEMINI.value\n    elif \"Ollama\" in class_name:\n        return AgentScopeGenAiProviderName.OLLAMA.value\n    else:\n        model_name = getattr(embedding_model, \"model_name\", \"\")\n        if \"text-embedding\" in model_name:\n            if \"qwen\" in model_name or \"dashscope\" in model_name:\n                return AgentScopeGenAiProviderName.DASHSCOPE.value\n            else:\n                return GenAIAttributes.GenAiProviderNameValues.OPENAI.value\n        return \"agentscope\"\n\n\ndef _get_tool_definitions(\n    tools: list[dict[str, Any]] | None,\n    tool_choice: str | None,\n) -> str | None:\n    \"\"\"Extract and serialize tool definitions for tracing.\"\"\"\n    if tools is None or not isinstance(tools, list) or len(tools) == 0:\n        return None\n\n    if tool_choice == \"none\":\n        return None\n\n    try:\n        flat_tools = []\n        for tool in tools:\n            if not isinstance(tool, dict) or \"function\" not in tool:\n                continue\n\n            func_def = tool[\"function\"]\n            flat_tool = {\n                \"type\": tool.get(\"type\", \"function\"),\n                \"name\": func_def.get(\"name\"),\n                \"description\": func_def.get(\"description\"),\n                \"parameters\": func_def.get(\"parameters\"),\n            }\n            flat_tool = {k: v for k, v in flat_tool.items() if v is not None}\n            flat_tools.append(flat_tool)\n\n        if flat_tools:\n            return _serialize_to_str(flat_tools)\n        return None\n\n    except Exception:\n        return None\n\n\ndef _to_serializable(obj: Any) -> Any:\n    \"\"\"Convert an object to a JSON serializable type.\"\"\"\n    if isinstance(obj, (str, int, bool, float, type(None))):\n        return obj\n    elif isinstance(obj, (list, tuple, set, frozenset)):\n        return [_to_serializable(x) for x in obj]\n    elif isinstance(obj, dict):\n        return {str(key): _to_serializable(val) for (key, val) in obj.items()}\n    elif isinstance(obj, (Msg, BaseModel)) or is_dataclass(obj):\n        return repr(obj)\n    elif inspect.isclass(obj) and issubclass(obj, BaseModel):\n        return repr(obj)\n    elif isinstance(obj, (datetime.date, datetime.datetime, datetime.time)):\n        return obj.isoformat()\n    elif isinstance(obj, datetime.timedelta):\n        return obj.total_seconds()\n    elif isinstance(obj, enum.Enum):\n        return _to_serializable(obj.value)\n    else:\n        return str(obj)\n\n\ndef _serialize_to_str(value: Any) -> str:\n    \"\"\"Serialize value to JSON string\"\"\"\n    try:\n        return json.dumps(value, ensure_ascii=False)\n    except TypeError:\n        return json.dumps(_to_serializable(value), ensure_ascii=False)\n\n\ndef _convert_block_to_part(block: Dict[str, Any]) -> Dict[str, Any] | None:\n    \"\"\"Convert content block to standardized part format.\"\"\"\n    block_type = block.get(\"type\")\n\n    if block_type == \"text\":\n        return {\"type\": \"text\", \"content\": block.get(\"text\", \"\")}\n\n    elif block_type == \"thinking\":\n        return {\n            \"type\": \"text\",\n            \"content\": f\"[Thinking] {block.get('thinking', '')}\",\n        }\n\n    elif block_type == \"tool_use\":\n        return {\n            \"type\": \"tool_call\",\n            \"id\": block.get(\"id\", \"\"),\n            \"name\": block.get(\"name\", \"\"),\n            \"arguments\": block.get(\"input\", {}),\n        }\n\n    elif block_type == \"tool_result\":\n        output = block.get(\"output\", \"\")\n        result = (\n            _serialize_to_str(output)\n            if isinstance(output, (list, dict))\n            else str(output)\n        )\n        return {\n            \"type\": \"tool_call_response\",\n            \"id\": block.get(\"id\", \"\"),\n            \"result\": result,\n        }\n\n    elif block_type in [\"image\", \"audio\", \"video\"]:\n        source = block.get(\"source\", {})\n        source_type = source.get(\"type\")\n        media_type = source.get(\"media_type\")\n\n        if source_type == \"url\":\n            url = source.get(\"url\", \"\")\n            # Infer mime_type from URL extension if not provided\n            if not media_type:\n                parsed_url = urlparse(url)\n                media_type, _ = mimetypes.guess_type(parsed_url.path)\n            return {\n                \"type\": \"uri\",\n                \"uri\": url,\n                \"modality\": block_type,\n                \"mime_type\": media_type,\n            }\n        elif source_type == \"base64\":\n            if not media_type:\n                default_media_types = {\n                    \"image\": \"image/jpeg\",\n                    \"audio\": \"audio/wav\",\n                    \"video\": \"video/mp4\",\n                }\n                media_type = default_media_types.get(block_type, \"unknown\")\n            return {\n                \"type\": \"blob\",\n                \"content\": source.get(\"data\", \"\"),\n                \"media_type\": media_type,\n                \"modality\": block_type,\n            }\n\n    return None\n\n\ndef _format_msg_to_parts(msg: Msg) -> dict[str, Any]:\n    \"\"\"Convert Msg to standard format (parts structure)\"\"\"\n    try:\n        parts = []\n        for block in msg.get_content_blocks():\n            part = _convert_block_to_part(block)\n            if part:\n                parts.append(part)\n\n        formatted_msg = {\"role\": msg.role, \"parts\": parts}\n        if msg.name:\n            formatted_msg[\"name\"] = msg.name\n\n        return formatted_msg\n\n    except Exception as e:\n        logger.debug(f\"Error formatting message: {e}\")\n        return {\n            \"role\": msg.role,\n            \"parts\": [\n                {\n                    \"type\": \"text\",\n                    \"content\": str(msg.content) if msg.content else \"\",\n                }\n            ],\n        }\n\n\ndef create_llm_invocation(\n    call_instance: ChatModelBase,\n    call_args: Tuple[Any, ...],\n    call_kwargs: Dict[str, Any],\n) -> LLMInvocation:\n    \"\"\"Create LLM invocation from call context.\"\"\"\n    provider_name = get_provider_name(call_instance)\n    request_model = getattr(call_instance, \"model_name\", \"unknown_model\")\n\n    # Convert input messages\n    input_messages = []\n    messages = call_args[0] if call_args else call_kwargs.get(\"messages\")\n    if messages:\n        try:\n            converted_messages = get_message_converter(provider_name)(messages)\n            input_messages = convert_agentscope_messages_to_genai_format(\n                converted_messages, provider_name=provider_name\n            )\n        except Exception as e:\n            logger.warning(\n                f\"Failed to convert input messages: {e}\", exc_info=True\n            )\n\n    invocation = LLMInvocation(\n        request_model=request_model,\n        provider=provider_name,\n        input_messages=input_messages,\n    )\n    entry_session_id = apply_entry_baggage_identity(invocation)\n    if entry_session_id and invocation.conversation_id is None:\n        invocation.conversation_id = entry_session_id\n\n    # Set optional request parameters if present\n    if call_kwargs.get(\"max_tokens\"):\n        invocation.max_tokens = call_kwargs[\"max_tokens\"]\n    if call_kwargs.get(\"temperature\"):\n        invocation.temperature = call_kwargs[\"temperature\"]\n    if call_kwargs.get(\"top_p\"):\n        invocation.top_p = call_kwargs[\"top_p\"]\n\n    # Set tool definitions if applicable\n    if not call_kwargs.get(\"structured_model\", False):\n        tool_definitions = _get_tool_definitions(\n            call_kwargs.get(\"tools\"), call_kwargs.get(\"tool_choice\")\n        )\n        if tool_definitions:\n            invocation.attributes[\"gen_ai.request.tool_definitions\"] = (\n                tool_definitions\n            )\n\n    return invocation\n\n\ndef create_embedding_invocation(\n    call_instance: EmbeddingModelBase,\n    call_args: Tuple[Any, ...],\n    call_kwargs: Dict[str, Any],\n) -> EmbeddingInvocation:\n    \"\"\"Create Embedding invocation from call context.\"\"\"\n    provider_name = get_embedding_provider_name(call_instance)\n    request_model = getattr(call_instance, \"model_name\", \"unknown_model\")\n\n    invocation = EmbeddingInvocation(\n        request_model=request_model,\n        provider=provider_name,\n    )\n    apply_entry_baggage_identity(invocation)\n\n    # Set encoding formats if present\n    if call_kwargs.get(\"encoding_formats\"):\n        invocation.encoding_formats = call_kwargs[\"encoding_formats\"]\n\n    return invocation\n\n\ndef create_agent_invocation(\n    reply_instance: AgentBase,\n    reply_args: Tuple[Any, ...],\n    reply_kwargs: Dict[str, Any],\n) -> InvokeAgentInvocation:\n    \"\"\"Create Agent invocation from reply call context.\"\"\"\n    # Get provider and model info\n    provider_name = None\n    request_model = None\n    if hasattr(reply_instance, \"model\") and reply_instance.model:\n        model = reply_instance.model\n        request_model = getattr(model, \"model_name\", None)\n        if isinstance(model, ChatModelBase):\n            provider_name = get_provider_name(model)\n\n    # Convert input messages\n    input_messages = []\n    msg = reply_args[0] if reply_args else reply_kwargs.get(\"msg\")\n    if msg:\n        try:\n            if isinstance(msg, Msg):\n                input_messages = convert_agentscope_messages_to_genai_format(\n                    [msg]\n                )\n            elif isinstance(msg, list):\n                input_messages = convert_agentscope_messages_to_genai_format(\n                    msg\n                )\n        except Exception as e:\n            logger.debug(f\"Error converting agent input messages: {e}\")\n\n    entry_session_id = _current_baggage_value(GEN_AI_SESSION_ID)\n    invocation = InvokeAgentInvocation(\n        provider=provider_name,\n        agent_name=getattr(reply_instance, \"name\", \"unknown_agent\"),\n        agent_id=getattr(reply_instance, \"id\", \"unknown\"),\n        agent_description=inspect.getdoc(reply_instance.__class__)\n        or \"No description available\",\n        conversation_id=entry_session_id or _config.run_id,\n        request_model=request_model,\n        input_messages=input_messages,\n    )\n    apply_entry_baggage_identity(invocation)\n\n    # Set system instruction if available\n    if hasattr(reply_instance, \"sys_prompt\") and reply_instance.sys_prompt:\n        sys_prompt = reply_instance.sys_prompt\n        if isinstance(sys_prompt, str):\n            invocation.system_instruction = [Text(content=sys_prompt)]\n        elif isinstance(sys_prompt, list):\n            invocation.system_instruction = sys_prompt\n\n    return invocation\n\n\ndef get_chatmodel_output_messages(chat_response: Any) -> List[Dict[str, Any]]:\n    \"\"\"Convert ChatResponse to OpenTelemetry standard output message format.\"\"\"\n    try:\n        if not isinstance(chat_response, ChatResponse):\n            return []\n\n        parts = []\n        for block in chat_response.content:\n            part = _convert_block_to_part(block)\n            if part:\n                parts.append(part)\n\n        if not parts:\n            parts.append({\"type\": \"text\", \"content\": \"\"})\n\n        return [\n            {\n                \"role\": \"assistant\",\n                \"parts\": parts,\n                \"finish_reason\": \"stop\",\n            }\n        ]\n\n    except Exception as e:\n        logger.warning(\n            f\"Error processing ChatResponse to output messages: {e}\"\n        )\n        return [\n            {\n                \"role\": \"assistant\",\n                \"parts\": [\n                    {\"type\": \"text\", \"content\": \"<error processing response>\"}\n                ],\n                \"finish_reason\": \"error\",\n            }\n        ]\n\n\ndef convert_agentscope_messages_to_genai_format(\n    messages: Any,\n    provider_name: Optional[str] = None,\n) -> List[InputMessage]:\n    \"\"\"Convert AgentScope messages to opentelemetry-util-genai InputMessage format.\n       parse _convert_block_to_part result from agentscope/tracing/_converter.py\n\n    This function is used by ExtendedTelemetryHandler which requires InputMessage objects.\n    \"\"\"\n    if not messages:\n        return []\n\n    if not isinstance(messages, list):\n        messages = [messages]\n\n    input_messages = []\n    for msg in messages:\n        if isinstance(msg, Msg):\n            msg_dict = _format_msg_to_parts(msg)\n        elif isinstance(msg, dict):\n            msg_dict = msg\n        else:\n            continue\n\n        role = msg_dict.get(\"role\", \"user\")\n        parts = msg_dict.get(\"parts\", [])\n\n        converted_parts = []\n        for part in parts:\n            part_type = part.get(\"type\")\n            if part_type == \"text\":\n                converted_parts.append(Text(content=part.get(\"content\", \"\")))\n            elif part_type == \"tool_call\":\n                converted_parts.append(\n                    ToolCall(\n                        id=part.get(\"id\"),\n                        name=part.get(\"name\", \"\"),\n                        arguments=part.get(\"arguments\", {}),\n                    )\n                )\n            elif part_type == \"tool_call_response\":\n                converted_parts.append(\n                    ToolCallResponse(\n                        id=part.get(\"id\"),\n                        response=part[\"response\"]\n                        if \"response\" in part\n                        else part.get(\"result\", \"\"),\n                    )\n                )\n            elif part_type == \"reasoning\":\n                converted_parts.append(\n                    Reasoning(content=part.get(\"content\", \"\"))\n                )\n            elif part_type == \"uri\":\n                converted_parts.append(\n                    Uri(\n                        uri=part.get(\"uri\"),\n                        modality=part.get(\"modality\"),\n                        mime_type=part.get(\"mime_type\"),\n                    )\n                )\n            elif part_type == \"blob\":\n                converted_parts.append(\n                    Base64Blob(\n                        content=part.get(\"content\", \"\"),\n                        modality=part.get(\"modality\"),\n                        mime_type=part.get(\"media_type\")\n                        if \"media_type\" in part\n                        else part.get(\"mime_type\"),\n                    )\n                )\n            else:\n                # Keep other types as-is\n                converted_parts.append(part)\n\n        input_messages.append(InputMessage(role=role, parts=converted_parts))\n\n    return input_messages\n\n\ndef convert_chatresponse_to_output_messages(\n    chat_response: Any,\n) -> List[OutputMessage]:\n    \"\"\"Convert ChatResponse to opentelemetry-util-genai OutputMessage format.\n\n    This function is used by ExtendedTelemetryHandler which requires OutputMessage objects.\n    \"\"\"\n    output_dicts = get_chatmodel_output_messages(chat_response)\n    if not output_dicts:\n        return []\n\n    output_messages = []\n    for msg_dict in output_dicts:\n        role = msg_dict.get(\"role\", \"assistant\")\n        parts_dicts = msg_dict.get(\"parts\", [])\n        finish_reason = msg_dict.get(\"finish_reason\", \"stop\")\n\n        converted_parts = []\n        for part in parts_dicts:\n            part_type = part.get(\"type\")\n            if part_type == \"text\":\n                converted_parts.append(Text(content=part.get(\"content\", \"\")))\n            elif part_type == \"tool_call\":\n                converted_parts.append(\n                    ToolCall(\n                        id=part.get(\"id\"),\n                        name=part.get(\"name\", \"\"),\n                        arguments=part.get(\"arguments\", {}),\n                    )\n                )\n            else:\n                converted_parts.append(part)\n\n        output_messages.append(\n            OutputMessage(\n                role=role,\n                parts=converted_parts,\n                finish_reason=finish_reason,\n            )\n        )\n\n    return output_messages\n\n\ndef convert_agent_response_to_output_messages(\n    agent_response: Any,\n) -> List[OutputMessage]:\n    \"\"\"Convert agent Msg response to opentelemetry-util-genai OutputMessage format.\n\n    This function is used by ExtendedTelemetryHandler which requires OutputMessage objects.\n    \"\"\"\n    if not hasattr(agent_response, \"content\"):\n        return []\n\n    try:\n        # Use _format_msg_to_parts to convert Msg to dict, then convert to OutputMessage\n        msg_dict = _format_msg_to_parts(agent_response)\n        parts_dicts = msg_dict.get(\"parts\", [])\n\n        converted_parts = []\n        for part in parts_dicts:\n            part_type = part.get(\"type\")\n            if part_type == \"text\":\n                converted_parts.append(Text(content=part.get(\"content\", \"\")))\n            elif part_type == \"reasoning\":\n                converted_parts.append(\n                    Reasoning(content=part.get(\"content\", \"\"))\n                )\n            elif part_type == \"tool_call\":\n                converted_parts.append(\n                    ToolCall(\n                        id=part.get(\"id\"),\n                        name=part.get(\"name\", \"\"),\n                        arguments=part.get(\"arguments\", {}),\n                    )\n                )\n            elif part_type in (\"uri\", \"blob\"):\n                converted_parts.append(part)\n            else:\n                converted_parts.append(Text(content=str(part)))\n\n        if not converted_parts:\n            converted_parts.append(Text(content=\"\"))\n\n        return [\n            OutputMessage(\n                role=\"assistant\",\n                parts=converted_parts,\n                finish_reason=\"stop\",\n            )\n        ]\n    except Exception as e:\n        logger.debug(f\"Failed to convert agent response: {e}\")\n        return []\n"}, "a55f07a839529b2d7155d7080acc8b4ac21fab25": {"t": 1785758580.6980226, "n": 22548, "h": "dd4203458d5271d3", "c": "# Copyright The OpenTelemetry Authors\n#\n# Licensed under the Apache License, Version 2.0 (the \"License\");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n#     http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an \"AS IS\" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n\n# -*- coding: utf-8 -*-\n\"\"\"Attributes processor for span attributes.\"\"\"\n\nfrom __future__ import annotations\n\nimport datetime\nimport enum\nimport inspect\nimport json\nimport logging\nimport mimetypes\nfrom dataclasses import is_dataclass\nfrom enum import Enum\nfrom typing import Any, Dict, List, Optional, Tuple\nfrom urllib.parse import urlparse\n\nfrom agentscope import _config\nfrom agentscope.agent import AgentBase\nfrom agentscope.embedding import EmbeddingModelBase\nfrom agentscope.message import Msg\nfrom agentscope.model import ChatModelBase, ChatResponse\nfrom pydantic import BaseModel\n\nfrom opentelemetry import baggage\nfrom opentelemetry.semconv._incubating.attributes import (\n    gen_ai_attributes as GenAIAttributes,\n)\nfrom opentelemetry.util.genai.extended_semconv.gen_ai_extended_attributes import (\n    GEN_AI_SESSION_ID,\n    GEN_AI_USER_ID,\n)\nfrom opentelemetry.util.genai.extended_types import (\n    EmbeddingInvocation,\n    InvokeAgentInvocation,\n)\nfrom opentelemetry.util.genai.types import (\n    Base64Blob,\n    InputMessage,\n    LLMInvocation,\n    OutputMessage,\n    Reasoning,\n    Text,\n    ToolCall,\n    ToolCallResponse,\n    Uri,\n)\n\nfrom .message_converter import get_message_converter\n\nlogger = logging.getLogger(__name__)\n\n\nclass AgentScopeGenAiProviderName(str, Enum):\n    \"\"\"Extended provider names not in standard OpenTelemetry semantic conventions.\"\"\"\n\n    OLLAMA = \"ollama\"\n    DASHSCOPE = \"dashscope\"\n    MOONSHOT = \"moonshot\"\n\n\n# Provider name mapping based on class names\n_PROVIDER_NAME_MAP = {\n    \"openai\": GenAIAttributes.GenAiProviderNameValues.OPENAI.value,\n    \"gemini\": GenAIAttributes.GenAiProviderNameValues.GCP_GEMINI.value,\n    \"anthropic\": GenAIAttributes.GenAiProviderNameValues.ANTHROPIC.value,\n    \"dashscope\": AgentScopeGenAiProviderName.DASHSCOPE.value,\n    \"ollama\": AgentScopeGenAiProviderName.OLLAMA.value,\n}\n\n# Base URL to provider mapping for OpenAI-compatible APIs\n_BASE_URL_PROVIDER_MAP = [\n    (\"openai.com\", GenAIAttributes.GenAiProviderNameValues.OPENAI.value),\n    (\n        \"api.deepseek.com\",\n        GenAIAttributes.GenAiProviderNameValues.DEEPSEEK.value,\n    ),\n    (\"dashscope.aliyuncs.com\", AgentScopeGenAiProviderName.DASHSCOPE.value),\n]\n\n\ndef _current_baggage_value(key: str) -> str | None:\n    try:\n        value = baggage.get_baggage(key)\n    except Exception:\n        return None\n    if value is None:\n        return None\n    text = str(value).strip()\n    return text or None\n\n\ndef entry_baggage_identity_attributes() -> dict[str, str]:\n    \"\"\"Return entry-level identity from current OpenTelemetry Baggage.\n\n    QwenPaw opens an Entry span before AgentScope runs and writes\n    ``gen_ai.session.id`` / ``gen_ai.user.id`` into Baggage. AgentScope has\n    its own ``run_id``; when both instrumentations are active, entry baggage is\n    the request-level identity and should color downstream spans.\n    \"\"\"\n    attributes: dict[str, str] = {}\n    session_id = _current_baggage_value(GEN_AI_SESSION_ID)\n    user_id = _current_baggage_value(GEN_AI_USER_ID)\n    if session_id:\n        attributes[GEN_AI_SESSION_ID] = session_id\n    if user_id:\n        attributes[GEN_AI_USER_ID] = user_id\n    return attributes\n\n\ndef apply_entry_baggage_identity(invocation: Any) -> str | None:\n    \"\"\"Copy entry-level identity baggage onto a GenAI invocation.\"\"\"\n    attributes = entry_baggage_identity_attributes()\n    for key, value in attributes.items():\n        invocation.attributes.setdefault(key, value)\n    return attributes.get(GEN_AI_SESSION_ID)\n\n\ndef get_provider_name(chat_model: ChatModelBase) -> str:\n    \"\"\"Parse chat model provider name\"\"\"\n    classname = chat_model.__class__.__name__\n    prefix = classname.removesuffix(\"ChatModel\").lower()\n\n    # Special handling for DashScopeChatModel with base_http_api_url\n    if (\n        prefix == \"dashscope\"\n        and hasattr(chat_model, \"base_http_api_url\")\n        and chat_model.base_http_api_url\n    ):\n        base_url = chat_model.base_http_api_url\n        for url_fragment, provider in _BASE_URL_PROVIDER_MAP:\n            if url_fragment in base_url:\n                return provider\n\n    return _PROVIDER_NAME_MAP.get(prefix, \"unknown\")\n\n\ndef get_embedding_provider_name(embedding_model: EmbeddingModelBase) -> str:\n    \"\"\"Parse embedding model provider name\"\"\"\n    class_name = embedding_model.__class__.__name__\n\n    if \"DashScope\" in class_name:\n        return AgentScopeGenAiProviderName.DASHSCOPE.value\n    elif \"OpenAI\" in class_name:\n        return GenAIAttributes.GenAiProviderNameValues.OPENAI.value\n    elif \"Gemini\" in class_name:\n        return GenAIAttributes.GenAiProviderNameValues.GCP_GEMINI.value\n    elif \"Ollama\" in class_name:\n        return AgentScopeGenAiProviderName.OLLAMA.value\n    else:\n        model_name = getattr(embedding_model, \"model_name\", \"\")\n        if \"text-embedding\" in model_name:\n            if \"qwen\" in model_name or \"dashscope\" in model_name:\n                return AgentScopeGenAiProviderName.DASHSCOPE.value\n            else:\n                return GenAIAttributes.GenAiProviderNameValues.OPENAI.value\n        return \"agentscope\"\n\n\ndef _get_tool_definitions(\n    tools: list[dict[str, Any]] | None,\n    tool_choice: str | None,\n) -> str | None:\n    \"\"\"Extract and serialize tool definitions for tracing.\"\"\"\n    if tools is None or not isinstance(tools, list) or len(tools) == 0:\n        return None\n\n    if tool_choice == \"none\":\n        return None\n\n    try:\n        flat_tools = []\n        for tool in tools:\n            if not isinstance(tool, dict) or \"function\" not in tool:\n                continue\n\n            func_def = tool[\"function\"]\n            flat_tool = {\n                \"type\": tool.get(\"type\", \"function\"),\n                \"name\": func_def.get(\"name\"),\n                \"description\": func_def.get(\"description\"),\n                \"parameters\": func_def.get(\"parameters\"),\n            }\n            flat_tool = {k: v for k, v in flat_tool.items() if v is not None}\n            flat_tools.append(flat_tool)\n\n        if flat_tools:\n            return _serialize_to_str(flat_tools)\n        return None\n\n    except Exception:\n        return None\n\n\ndef _to_serializable(obj: Any) -> Any:\n    \"\"\"Convert an object to a JSON serializable type.\"\"\"\n    if isinstance(obj, (str, int, bool, float, type(None))):\n        return obj\n    elif isinstance(obj, (list, tuple, set, frozenset)):\n        return [_to_serializable(x) for x in obj]\n    elif isinstance(obj, dict):\n        return {str(key): _to_serializable(val) for (key, val) in obj.items()}\n    elif isinstance(obj, (Msg, BaseModel)) or is_dataclass(obj):\n        return repr(obj)\n    elif inspect.isclass(obj) and issubclass(obj, BaseModel):\n        return repr(obj)\n    elif isinstance(obj, (datetime.date, datetime.datetime, datetime.time)):\n        return obj.isoformat()\n    elif isinstance(obj, datetime.timedelta):\n        return obj.total_seconds()\n    elif isinstance(obj, enum.Enum):\n        return _to_serializable(obj.value)\n    else:\n        return str(obj)\n\n\ndef _serialize_to_str(value: Any) -> str:\n    \"\"\"Serialize value to JSON string\"\"\"\n    try:\n        return json.dumps(value, ensure_ascii=False)\n    except TypeError:\n        return json.dumps(_to_serializable(value), ensure_ascii=False)\n\n\ndef _convert_block_to_part(block: Dict[str, Any]) -> Dict[str, Any] | None:\n    \"\"\"Convert content block to standardized part format.\"\"\"\n    block_type = block.get(\"type\")\n\n    if block_type == \"text\":\n        return {\"type\": \"text\", \"content\": block.get(\"text\", \"\")}\n\n    elif block_type == \"thinking\":\n        return {\n            \"type\": \"text\",\n            \"content\": f\"[Thinking] {block.get('thinking', '')}\",\n        }\n\n    elif block_type == \"tool_use\":\n        return {\n            \"type\": \"tool_call\",\n            \"id\": block.get(\"id\", \"\"),\n            \"name\": block.get(\"name\", \"\"),\n            \"arguments\": block.get(\"input\", {}),\n        }\n\n    elif block_type == \"tool_result\":\n        output = block.get(\"output\", \"\")\n        result = (\n            _serialize_to_str(output)\n            if isinstance(output, (list, dict))\n            else str(output)\n        )\n        return {\n            \"type\": \"tool_call_response\",\n            \"id\": block.get(\"id\", \"\"),\n            \"result\": result,\n        }\n\n    elif block_type in [\"image\", \"audio\", \"video\"]:\n        source = block.get(\"source\", {})\n        source_type = source.get(\"type\")\n        media_type = source.get(\"media_type\")\n\n        if source_type == \"url\":\n            url = source.get(\"url\", \"\")\n            # Infer mime_type from URL extension if not provided\n            if not media_type:\n                parsed_url = urlparse(url)\n                media_type, _ = mimetypes.guess_type(parsed_url.path)\n            return {\n                \"type\": \"uri\",\n                \"uri\": url,\n                \"modality\": block_type,\n                \"mime_type\": media_type,\n            }\n        elif source_type == \"base64\":\n            if not media_type:\n                default_media_types = {\n                    \"image\": \"image/jpeg\",\n                    \"audio\": \"audio/wav\",\n                    \"video\": \"video/mp4\",\n                }\n                media_type = default_media_types.get(block_type, \"unknown\")\n            return {\n                \"type\": \"blob\",\n                \"content\": source.get(\"data\", \"\"),\n                \"media_type\": media_type,\n                \"modality\": block_type,\n            }\n\n    return None\n\n\ndef _format_msg_to_parts(msg: Msg) -> dict[str, Any]:\n    \"\"\"Convert Msg to standard format (parts structure)\"\"\"\n    try:\n        parts = []\n        for block in msg.get_content_blocks():\n            part = _convert_block_to_part(block)\n            if part:\n                parts.append(part)\n\n        formatted_msg = {\"role\": msg.role, \"parts\": parts}\n        if msg.name:\n            formatted_msg[\"name\"] = msg.name\n\n        return formatted_msg\n\n    except Exception as e:\n        logger.debug(f\"Error formatting message: {e}\")\n        return {\n            \"role\": msg.role,\n            \"parts\": [\n                {\n                    \"type\": \"text\",\n                    \"content\": str(msg.content) if msg.content else \"\",\n                }\n            ],\n        }\n\n\ndef create_llm_invocation(\n    call_instance: ChatModelBase,\n    call_args: Tuple[Any, ...],\n    call_kwargs: Dict[str, Any],\n) -> LLMInvocation:\n    \"\"\"Create LLM invocation from call context.\"\"\"\n    provider_name = get_provider_name(call_instance)\n    request_model = getattr(call_instance, \"model_name\", \"unknown_model\")\n\n    # Convert input messages\n    input_messages = []\n    messages = call_args[0] if call_args else call_kwargs.get(\"messages\")\n    if messages:\n        try:\n            converted_messages = get_message_converter(provider_name)(messages)\n            input_messages = convert_agentscope_messages_to_genai_format(\n                converted_messages, provider_name=provider_name\n            )\n        except Exception as e:\n            logger.warning(\n                f\"Failed to convert input messages: {e}\", exc_info=True\n            )\n\n    invocation = LLMInvocation(\n        request_model=request_model,\n        provider=provider_name,\n        input_messages=input_messages,\n    )\n    entry_session_id = apply_entry_baggage_identity(invocation)\n    if entry_session_id and invocation.conversation_id is None:\n        invocation.conversation_id = entry_session_id\n\n    # Set optional request parameters if present\n    if call_kwargs.get(\"max_tokens\"):\n        invocation.max_tokens = call_kwargs[\"max_tokens\"]\n    if call_kwargs.get(\"temperature\"):\n        invocation.temperature = call_kwargs[\"temperature\"]\n    if call_kwargs.get(\"top_p\"):\n        invocation.top_p = call_kwargs[\"top_p\"]\n\n    # Set tool definitions if applicable\n    if not call_kwargs.get(\"structured_model\", False):\n        tool_definitions = _get_tool_definitions(\n            call_kwargs.get(\"tools\"), call_kwargs.get(\"tool_choice\")\n        )\n        if tool_definitions:\n            invocation.attributes[\"gen_ai.request.tool_definitions\"] = (\n                tool_definitions\n            )\n\n    return invocation\n\n\ndef create_embedding_invocation(\n    call_instance: EmbeddingModelBase,\n    call_args: Tuple[Any, ...],\n    call_kwargs: Dict[str, Any],\n) -> EmbeddingInvocation:\n    \"\"\"Create Embedding invocation from call context.\"\"\"\n    provider_name = get_embedding_provider_name(call_instance)\n    request_model = getattr(call_instance, \"model_name\", \"unknown_model\")\n\n    invocation = EmbeddingInvocation(\n        request_model=request_model,\n        provider=provider_name,\n    )\n    apply_entry_baggage_identity(invocation)\n\n    # Set encoding formats if present\n    if call_kwargs.get(\"encoding_formats\"):\n        invocation.encoding_formats = call_kwargs[\"encoding_formats\"]\n\n    return invocation\n\n\ndef create_agent_invocation(\n    reply_instance: AgentBase,\n    reply_args: Tuple[Any, ...],\n    reply_kwargs: Dict[str, Any],\n) -> InvokeAgentInvocation:\n    \"\"\"Create Agent invocation from reply call context.\"\"\"\n    # Get provider and model info\n    provider_name = None\n    request_model = None\n    if hasattr(reply_instance, \"model\") and reply_instance.model:\n        model = reply_instance.model\n        request_model = getattr(model, \"model_name\", None)\n        if isinstance(model, ChatModelBase):\n            provider_name = get_provider_name(model)\n\n    # Convert input messages\n    input_messages = []\n    msg = reply_args[0] if reply_args else reply_kwargs.get(\"msg\")\n    if msg:\n        try:\n            if isinstance(msg, Msg):\n                input_messages = convert_agentscope_messages_to_genai_format(\n                    [msg]\n                )\n            elif isinstance(msg, list):\n                input_messages = convert_agentscope_messages_to_genai_format(\n                    msg\n                )\n        except Exception as e:\n            logger.debug(f\"Error converting agent input messages: {e}\")\n\n    entry_session_id = _current_baggage_value(GEN_AI_SESSION_ID)\n    invocation = InvokeAgentInvocation(\n        provider=provider_name,\n        agent_name=getattr(reply_instance, \"name\", \"unknown_agent\"),\n        agent_id=getattr(reply_instance, \"id\", \"unknown\"),\n        agent_description=inspect.getdoc(reply_instance.__class__)\n        or \"No description available\",\n        conversation_id=entry_session_id or _config.run_id,\n        request_model=request_model,\n        input_messages=input_messages,\n    )\n    apply_entry_baggage_identity(invocation)\n\n    # Set system instruction if available\n    if hasattr(reply_instance, \"sys_prompt\") and reply_instance.sys_prompt:\n        sys_prompt = reply_instance.sys_prompt\n        if isinstance(sys_prompt, str):\n            invocation.system_instruction = [Text(content=sys_prompt)]\n        elif isinstance(sys_prompt, list):\n            invocation.system_instruction = sys_prompt\n\n    return invocation\n\n\ndef get_chatmodel_output_messages(chat_response: Any) -> List[Dict[str, Any]]:\n    \"\"\"Convert ChatResponse to OpenTelemetry standard output message format.\"\"\"\n    try:\n        if not isinstance(chat_response, ChatResponse):\n            return []\n\n        parts = []\n        for block in chat_response.content:\n            part = _convert_block_to_part(block)\n            if part:\n                parts.append(part)\n\n        if not parts:\n            parts.append({\"type\": \"text\", \"content\": \"\"})\n\n        return [\n            {\n                \"role\": \"assistant\",\n                \"parts\": parts,\n                \"finish_reason\": \"stop\",\n            }\n        ]\n\n    except Exception as e:\n        logger.warning(\n            f\"Error processing ChatResponse to output messages: {e}\"\n        )\n        return [\n            {\n                \"role\": \"assistant\",\n                \"parts\": [\n                    {\"type\": \"text\", \"content\": \"<error processing response>\"}\n                ],\n                \"finish_reason\": \"error\",\n            }\n        ]\n\n\ndef convert_agentscope_messages_to_genai_format(\n    messages: Any,\n    provider_name: Optional[str] = None,\n) -> List[InputMessage]:\n    \"\"\"Convert AgentScope messages to opentelemetry-util-genai InputMessage format.\n       parse _convert_block_to_part result from agentscope/tracing/_converter.py\n\n    This function is used by ExtendedTelemetryHandler which requires InputMessage objects.\n    \"\"\"\n    if not messages:\n        return []\n\n    if not isinstance(messages, list):\n        messages = [messages]\n\n    input_messages = []\n    for msg in messages:\n        if isinstance(msg, Msg):\n            msg_dict = _format_msg_to_parts(msg)\n        elif isinstance(msg, dict):\n            msg_dict = msg\n        else:\n            continue\n\n        role = msg_dict.get(\"role\", \"user\")\n        parts = msg_dict.get(\"parts\", [])\n\n        converted_parts = []\n        for part in parts:\n            part_type = part.get(\"type\")\n            if part_type == \"text\":\n                converted_parts.append(Text(content=part.get(\"content\", \"\")))\n            elif part_type == \"tool_call\":\n                converted_parts.append(\n                    ToolCall(\n                        id=part.get(\"id\"),\n                        name=part.get(\"name\", \"\"),\n                        arguments=part.get(\"arguments\", {}),\n                    )\n                )\n            elif part_type == \"tool_call_response\":\n                converted_parts.append(\n                    ToolCallResponse(\n                        id=part.get(\"id\"),\n                        response=part[\"response\"]\n                        if \"response\" in part\n                        else part.get(\"result\", \"\"),\n                    )\n                )\n            elif part_type == \"reasoning\":\n                converted_parts.append(\n                    Reasoning(content=part.get(\"content\", \"\"))\n                )\n            elif part_type == \"uri\":\n                converted_parts.append(\n                    Uri(\n                        uri=part.get(\"uri\"),\n                        modality=part.get(\"modality\"),\n                        mime_type=part.get(\"mime_type\"),\n                    )\n                )\n            elif part_type == \"blob\":\n                converted_parts.append(\n                    Base64Blob(\n                        content=part.get(\"content\", \"\"),\n                        modality=part.get(\"modality\"),\n                        mime_type=part.get(\"media_type\")\n                        if \"media_type\" in part\n                        else part.get(\"mime_type\"),\n                    )\n                )\n            else:\n                # Keep other types as-is\n                converted_parts.append(part)\n\n        input_messages.append(InputMessage(role=role, parts=converted_parts))\n\n    return input_messages\n\n\ndef convert_chatresponse_to_output_messages(\n    chat_response: Any,\n) -> List[OutputMessage]:\n    \"\"\"Convert ChatResponse to opentelemetry-util-genai OutputMessage format.\n\n    This function is used by ExtendedTelemetryHandler which requires OutputMessage objects.\n    \"\"\"\n    output_dicts = get_chatmodel_output_messages(chat_response)\n    if not output_dicts:\n        return []\n\n    output_messages = []\n    for msg_dict in output_dicts:\n        role = msg_dict.get(\"role\", \"assistant\")\n        parts_dicts = msg_dict.get(\"parts\", [])\n        finish_reason = msg_dict.get(\"finish_reason\", \"stop\")\n\n        converted_parts = []\n        for part in parts_dicts:\n            part_type = part.get(\"type\")\n            if part_type == \"text\":\n                converted_parts.append(Text(content=part.get(\"content\", \"\")))\n            elif part_type == \"tool_call\":\n                converted_parts.append(\n                    ToolCall(\n                        id=part.get(\"id\"),\n                        name=part.get(\"name\", \"\"),\n                        arguments=part.get(\"arguments\", {}),\n                    )\n                )\n            else:\n                converted_parts.append(part)\n\n        output_messages.append(\n            OutputMessage(\n                role=role,\n                parts=converted_parts,\n                finish_reason=finish_reason,\n            )\n        )\n\n    return output_messages\n\n\ndef convert_agent_response_to_output_messages(\n    agent_response: Any,\n) -> List[OutputMessage]:\n    \"\"\"Convert agent Msg response to opentelemetry-util-genai OutputMessage format.\n\n    This function is used by ExtendedTelemetryHandler which requires OutputMessage objects.\n    \"\"\"\n    if not hasattr(agent_response, \"content\"):\n        return []\n\n    try:\n        # Use _format_msg_to_parts to convert Msg to dict, then convert to OutputMessage\n        msg_dict = _format_msg_to_parts(agent_response)\n        parts_dicts = msg_dict.get(\"parts\", [])\n\n        converted_parts = []\n        for part in parts_dicts:\n            part_type = part.get(\"type\")\n            if part_type == \"text\":\n                converted_parts.append(Text(content=part.get(\"content\", \"\")))\n            elif part_type == \"reasoning\":\n                converted_parts.append(\n                    Reasoning(content=part.get(\"content\", \"\"))\n                )\n            elif part_type == \"tool_call\":\n                converted_parts.append(\n                    ToolCall(\n                        id=part.get(\"id\"),\n                        name=part.get(\"name\", \"\"),\n                        arguments=part.get(\"arguments\", {}),\n                    )\n                )\n            elif part_type in (\"uri\", \"blob\"):\n                converted_parts.append(part)\n            else:\n                converted_parts.append(Text(content=str(part)))\n\n        if not converted_parts:\n            converted_parts.append(Text(content=\"\"))\n\n        return [\n            OutputMessage(\n                role=\"assistant\",\n                parts=converted_parts,\n                finish_reason=\"stop\",\n            )\n        ]\n    except Exception as e:\n        logger.debug(f\"Failed to convert agent response: {e}\")\n        return []\n"}, "ebedcb11047d6017694dedfe72a859e241dfeca8": {"t": 1785758582.1956472, "n": 20728, "h": "a89398588bc18094", "c": "# Copyright The OpenTelemetry Authors\n#\n# Licensed under the Apache License, Version 2.0 (the \"License\");\n# you may not use this file except in compliance with the License.\n# You may obtain a copy of the License at\n#\n#     http://www.apache.org/licenses/LICENSE-2.0\n#\n# Unless required by applicable law or agreed to in writing, software\n# distributed under the License is distributed on an \"AS IS\" BASIS,\n# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n# See the License for the specific language governing permissions and\n# limitations under the License.\n\n# -*- coding: utf-8 -*-\n\"\"\"Attributes processor for span attributes.\"\"\"\n\nfrom __future__ import annotations\n\nimport datetime\nimport enum\nimport inspect\nimport json\nimport logging\nimport mimetypes\nfrom dataclasses import is_dataclass\nfrom enum import Enum\nfrom typing import Any, Dict, List, Optional, Tuple\nfrom urllib.parse import urlparse\n\nfrom agentscope import _config\nfrom agentscope.agent import AgentBase\nfrom agentscope.embedding import EmbeddingModelBase\nfrom agentscope.message import Msg\nfrom agentscope.model import ChatModelBase, ChatResponse\nfrom pydantic import BaseModel\n\nfrom opentelemetry.semconv._incubating.attributes import (\n    gen_ai_attributes as GenAIAttributes,\n)\nfrom opentelemetry.util.genai.extended_types import (\n    EmbeddingInvocation,\n    InvokeAgentInvocation,\n)\nfrom opentelemetry.util.genai.types import (\n    Base64Blob,\n    InputMessage,\n    LLMInvocation,\n    OutputMessage,\n    Reasoning,\n    Text,\n    ToolCall,\n    ToolCallResponse,\n    Uri,\n)\n\nfrom .message_converter import get_message_converter\n\nlogger = logging.getLogger(__name__)\n\n\nclass AgentScopeGenAiProviderName(str, Enum):\n    \"\"\"Extended provider names not in standard OpenTelemetry semantic conventions.\"\"\"\n\n    OLLAMA = \"ollama\"\n    DASHSCOPE = \"dashscope\"\n    MOONSHOT = \"moonshot\"\n\n\n# Provider name mapping based on class names\n_PROVIDER_NAME_MAP = {\n    \"openai\": GenAIAttributes.GenAiProviderNameValues.OPENAI.value,\n    \"gemini\": GenAIAttributes.GenAiProviderNameValues.GCP_GEMINI.value,\n    \"anthropic\": GenAIAttributes.GenAiProviderNameValues.ANTHROPIC.value,\n    \"dashscope\": AgentScopeGenAiProviderName.DASHSCOPE.value,\n    \"ollama\": AgentScopeGenAiProviderName.OLLAMA.value,\n}\n\n# Base URL to provider mapping for OpenAI-compatible APIs\n_BASE_URL_PROVIDER_MAP = [\n    (\"openai.com\", GenAIAttributes.GenAiProviderNameValues.OPENAI.value),\n    (\n        \"api.deepseek.com\",\n        GenAIAttributes.GenAiProviderNameValues.DEEPSEEK.value,\n    ),\n    (\"dashscope.aliyuncs.com\", AgentScopeGenAiProviderName.DASHSCOPE.value),\n]\n\n\ndef get_provider_name(chat_model: ChatModelBase) -> str:\n    \"\"\"Parse chat model provider name\"\"\"\n    classname = chat_model.__class__.__name__\n    prefix = classname.removesuffix(\"ChatModel\").lower()\n\n    # Special handling for DashScopeChatModel with base_http_api_url\n    if (\n        prefix == \"dashscope\"\n        and hasattr(chat_model, \"base_http_api_url\")\n        and chat_model.base_http_api_url\n    ):\n        base_url = chat_model.base_http_api_url\n        for url_fragment, provider in _BASE_URL_PROVIDER_MAP:\n            if url_fragment in base_url:\n                return provider\n\n    return _PROVIDER_NAME_MAP.get(prefix, \"unknown\")\n\n\ndef get_embedding_provider_name(embedding_model: EmbeddingModelBase) -> str:\n    \"\"\"Parse embedding model provider name\"\"\"\n    class_name = embedding_model.__class__.__name__\n\n    if \"DashScope\" in class_name:\n        return AgentScopeGenAiProviderName.DASHSCOPE.value\n    elif \"OpenAI\" in class_name:\n        return GenAIAttributes.GenAiProviderNameValues.OPENAI.value\n    elif \"Gemini\" in class_name:\n        return GenAIAttributes.GenAiProviderNameValues.GCP_GEMINI.value\n    elif \"Ollama\" in class_name:\n        return AgentScopeGenAiProviderName.OLLAMA.value\n    else:\n        model_name = getattr(embedding_model, \"model_name\", \"\")\n        if \"text-embedding\" in model_name:\n            if \"qwen\" in model_name or \"dashscope\" in model_name:\n                return AgentScopeGenAiProviderName.DASHSCOPE.value\n            else:\n                return GenAIAttributes.GenAiProviderNameValues.OPENAI.value\n        return \"agentscope\"\n\n\ndef _get_tool_definitions(\n    tools: list[dict[str, Any]] | None,\n    tool_choice: str | None,\n) -> str | None:\n    \"\"\"Extract and serialize tool definitions for tracing.\"\"\"\n    if tools is None or not isinstance(tools, list) or len(tools) == 0:\n        return None\n\n    if tool_choice == \"none\":\n        return None\n\n    try:\n        flat_tools = []\n        for tool in tools:\n            if not isinstance(tool, dict) or \"function\" not in tool:\n                continue\n\n            func_def = tool[\"function\"]\n            flat_tool = {\n                \"type\": tool.get(\"type\", \"function\"),\n                \"name\": func_def.get(\"name\"),\n                \"description\": func_def.get(\"description\"),\n                \"parameters\": func_def.get(\"parameters\"),\n            }\n            flat_tool = {k: v for k, v in flat_tool.items() if v is not None}\n            flat_tools.append(flat_tool)\n\n        if flat_tools:\n            return _serialize_to_str(flat_tools)\n        return None\n\n    except Exception:\n        return None\n\n\ndef _to_serializable(obj: Any) -> Any:\n    \"\"\"Convert an object to a JSON serializable type.\"\"\"\n    if isinstance(obj, (str, int, bool, float, type(None))):\n        return obj\n    elif isinstance(obj, (list, tuple, set, frozenset)):\n        return [_to_serializable(x) for x in obj]\n    elif isinstance(obj, dict):\n        return {str(key): _to_serializable(val) for (key, val) in obj.items()}\n    elif isinstance(obj, (Msg, BaseModel)) or is_dataclass(obj):\n        return repr(obj)\n    elif inspect.isclass(obj) and issubclass(obj, BaseModel):\n        return repr(obj)\n    elif isinstance(obj, (datetime.date, datetime.datetime, datetime.time)):\n        return obj.isoformat()\n    elif isinstance(obj, datetime.timedelta):\n        return obj.total_seconds()\n    elif isinstance(obj, enum.Enum):\n        return _to_serializable(obj.value)\n    else:\n        return str(obj)\n\n\ndef _serialize_to_str(value: Any) -> str:\n    \"\"\"Serialize value to JSON string\"\"\"\n    try:\n        return json.dumps(value, ensure_ascii=False)\n    except TypeError:\n        return json.dumps(_to_serializable(value), ensure_ascii=False)\n\n\ndef _convert_block_to_part(block: Dict[str, Any]) -> Dict[str, Any] | None:\n    \"\"\"Convert content block to standardized part format.\"\"\"\n    block_type = block.get(\"type\")\n\n    if block_type == \"text\":\n        return {\"type\": \"text\", \"content\": block.get(\"text\", \"\")}\n\n    elif block_type == \"thinking\":\n        return {\n            \"type\": \"text\",\n            \"content\": f\"[Thinking] {block.get('thinking', '')}\",\n        }\n\n    elif block_type == \"tool_use\":\n        return {\n            \"type\": \"tool_call\",\n            \"id\": block.get(\"id\", \"\"),\n            \"name\": block.get(\"name\", \"\"),\n            \"arguments\": block.get(\"input\", {}),\n        }\n\n    elif block_type == \"tool_result\":\n        output = block.get(\"output\", \"\")\n        result = (\n            _serialize_to_str(output)\n            if isinstance(output, (list, dict))\n            else str(output)\n        )\n        return {\n            \"type\": \"tool_call_response\",\n            \"id\": block.get(\"id\", \"\"),\n            \"result\": result,\n        }\n\n    elif block_type in [\"image\", \"audio\", \"video\"]:\n        source = block.get(\"source\", {})\n        source_type = source.get(\"type\")\n        media_type = source.get(\"media_type\")\n\n        if source_type == \"url\":\n            url = source.get(\"url\", \"\")\n            # Infer mime_type from URL extension if not provided\n            if not media_type:\n                parsed_url = urlparse(url)\n                media_type, _ = mimetypes.guess_type(parsed_url.path)\n            return {\n                \"type\": \"uri\",\n                \"uri\": url,\n                \"modality\": block_type,\n                \"mime_type\": media_type,\n            }\n        elif source_type == \"base64\":\n            if not media_type:\n                default_media_types = {\n                    \"image\": \"image/jpeg\",\n                    \"audio\": \"audio/wav\",\n                    \"video\": \"video/mp4\",\n                }\n                media_type = default_media_types.get(block_type, \"unknown\")\n            return {\n                \"type\": \"blob\",\n                \"content\": source.get(\"data\", \"\"),\n                \"media_type\": media_type,\n                \"modality\": block_type,\n            }\n\n    return None\n\n\ndef _format_msg_to_parts(msg: Msg) -> dict[str, Any]:\n    \"\"\"Convert Msg to standard format (parts structure)\"\"\"\n    try:\n        parts = []\n        for block in msg.get_content_blocks():\n            part = _convert_block_to_part(block)\n            if part:\n                parts.append(part)\n\n        formatted_msg = {\"role\": msg.role, \"parts\": parts}\n        if msg.name:\n            formatted_msg[\"name\"] = msg.name\n\n        return formatted_msg\n\n    except Exception as e:\n        logger.debug(f\"Error formatting message: {e}\")\n        return {\n            \"role\": msg.role,\n            \"parts\": [\n                {\n                    \"type\": \"text\",\n                    \"content\": str(msg.content) if msg.content else \"\",\n                }\n            ],\n        }\n\n\ndef create_llm_invocation(\n    call_instance: ChatModelBase,\n    call_args: Tuple[Any, ...],\n    call_kwargs: Dict[str, Any],\n) -> LLMInvocation:\n    \"\"\"Create LLM invocation from call context.\"\"\"\n    provider_name = get_provider_name(call_instance)\n    request_model = getattr(call_instance, \"model_name\", \"unknown_model\")\n\n    # Convert input messages\n    input_messages = []\n    messages = call_args[0] if call_args else call_kwargs.get(\"messages\")\n    if messages:\n        try:\n            converted_messages = get_message_converter(provider_name)(messages)\n            input_messages = convert_agentscope_messages_to_genai_format(\n                converted_messages, provider_name=provider_name\n            )\n        except Exception as e:\n            logger.warning(\n                f\"Failed to convert input messages: {e}\", exc_info=True\n            )\n\n    invocation = LLMInvocation(\n        request_model=request_model,\n        provider=provider_name,\n        input_messages=input_messages,\n    )\n\n    # Set optional request parameters if present\n    if call_kwargs.get(\"max_tokens\"):\n        invocation.max_tokens = call_kwargs[\"max_tokens\"]\n    if call_kwargs.get(\"temperature\"):\n        invocation.temperature = call_kwargs[\"temperature\"]\n    if call_kwargs.get(\"top_p\"):\n        invocation.top_p = call_kwargs[\"top_p\"]\n\n    # Set tool definitions if applicable\n    if not call_kwargs.get(\"structured_model\", False):\n        tool_definitions = _get_tool_definitions(\n            call_kwargs.get(\"tools\"), call_kwargs.get(\"tool_choice\")\n        )\n        if tool_definitions:\n            invocation.attributes[\"gen_ai.request.tool_definitions\"] = (\n                tool_definitions\n            )\n\n    return invocation\n\n\ndef create_embedding_invocation(\n    call_instance: EmbeddingModelBase,\n    call_args: Tuple[Any, ...],\n    call_kwargs: Dict[str, Any],\n) -> EmbeddingInvocation:\n    \"\"\"Create Embedding invocation from call context.\"\"\"\n    provider_name = get_embedding_provider_name(call_instance)\n    request_model = getattr(call_instance, \"model_name\", \"unknown_model\")\n\n    invocation = EmbeddingInvocation(\n        request_model=request_model,\n        provider=provider_name,\n    )\n\n    # Set encoding formats if present\n    if call_kwargs.get(\"encoding_formats\"):\n        invocation.encoding_formats = call_kwargs[\"encoding_formats\"]\n\n    return invocation\n\n\ndef create_agent_invocation(\n    reply_instance: AgentBase,\n    reply_args: Tuple[Any, ...],\n    reply_kwargs: Dict[str, Any],\n) -> InvokeAgentInvocation:\n    \"\"\"Create Agent invocation from reply call context.\"\"\"\n    # Get provider and model info\n    provider_name = None\n    request_model = None\n    if hasattr(reply_instance, \"model\") and reply_instance.model:\n        model = reply_instance.model\n        request_model = getattr(model, \"model_name\", None)\n        if isinstance(model, ChatModelBase):\n            provider_name = get_provider_name(model)\n\n    # Convert input messages\n    input_messages = []\n    msg = reply_args[0] if reply_args else reply_kwargs.get(\"msg\")\n    if msg:\n        try:\n            if isinstance(msg, Msg):\n                input_messages = convert_agentscope_messages_to_genai_format(\n                    [msg]\n                )\n            elif isinstance(msg, list):\n                input_messages = convert_agentscope_messages_to_genai_format(\n                    msg\n                )\n        except Exception as e:\n            logger.debug(f\"Error converting agent input messages: {e}\")\n\n    invocation = InvokeAgentInvocation(\n        provider=provider_name,\n        agent_name=getattr(reply_instance, \"name\", \"unknown_agent\"),\n        agent_id=getattr(reply_instance, \"id\", \"unknown\"),\n        agent_description=inspect.getdoc(reply_instance.__class__)\n        or \"No description available\",\n        conversation_id=_config.run_id,\n        request_model=request_model,\n        input_messages=input_messages,\n    )\n\n    # Set system instruction if available\n    if hasattr(reply_instance, \"sys_prompt\") and reply_instance.sys_prompt:\n        sys_prompt = reply_instance.sys_prompt\n        if isinstance(sys_prompt, str):\n            invocation.system_instruction = [Text(content=sys_prompt)]\n        elif isinstance(sys_prompt, list):\n            invocation.system_instruction = sys_prompt\n\n    return invocation\n\n\ndef get_chatmodel_output_messages(chat_response: Any) -> List[Dict[str, Any]]:\n    \"\"\"Convert ChatResponse to OpenTelemetry standard output message format.\"\"\"\n    try:\n        if not isinstance(chat_response, ChatResponse):\n            return []\n\n        parts = []\n        for block in chat_response.content:\n            part = _convert_block_to_part(block)\n            if part:\n                parts.append(part)\n\n        if not parts:\n            parts.append({\"type\": \"text\", \"content\": \"\"})\n\n        return [\n            {\n                \"role\": \"assistant\",\n                \"parts\": parts,\n                \"finish_reason\": \"stop\",\n            }\n        ]\n\n    except Exception as e:\n        logger.warning(\n            f\"Error processing ChatResponse to output messages: {e}\"\n        )\n        return [\n            {\n                \"role\": \"assistant\",\n                \"parts\": [\n                    {\"type\": \"text\", \"content\": \"<error processing response>\"}\n                ],\n                \"finish_reason\": \"error\",\n            }\n        ]\n\n\ndef convert_agentscope_messages_to_genai_format(\n    messages: Any,\n    provider_name: Optional[str] = None,\n) -> List[InputMessage]:\n    \"\"\"Convert AgentScope messages to opentelemetry-util-genai InputMessage format.\n       parse _convert_block_to_part result from agentscope/tracing/_converter.py\n\n    This function is used by ExtendedTelemetryHandler which requires InputMessage objects.\n    \"\"\"\n    if not messages:\n        return []\n\n    if not isinstance(messages, list):\n        messages = [messages]\n\n    input_messages = []\n    for msg in messages:\n        if isinstance(msg, Msg):\n            msg_dict = _format_msg_to_parts(msg)\n        elif isinstance(msg, dict):\n            msg_dict = msg\n        else:\n            continue\n\n        role = msg_dict.get(\"role\", \"user\")\n        parts = msg_dict.get(\"parts\", [])\n\n        converted_parts = []\n        for part in parts:\n            part_type = part.get(\"type\")\n            if part_type == \"text\":\n                converted_parts.append(Text(content=part.get(\"content\", \"\")))\n            elif part_type == \"tool_call\":\n                converted_parts.append(\n                    ToolCall(\n                        id=part.get(\"id\"),\n                        name=part.get(\"name\", \"\"),\n                        arguments=part.get(\"arguments\", {}),\n                    )\n                )\n            elif part_type == \"tool_call_response\":\n                converted_parts.append(\n                    ToolCallResponse(\n                        id=part.get(\"id\"),\n                        response=part[\"response\"]\n                        if \"response\" in part\n                        else part.get(\"result\", \"\"),\n                    )\n                )\n            elif part_type == \"reasoning\":\n                converted_parts.append(\n                    Reasoning(content=part.get(\"content\", \"\"))\n                )\n            elif part_type == \"uri\":\n                converted_parts.append(\n                    Uri(\n                        uri=part.get(\"uri\"),\n                        modality=part.get(\"modality\"),\n                        mime_type=part.get(\"mime_type\"),\n                    )\n                )\n            elif part_type == \"blob\":\n                converted_parts.append(\n                    Base64Blob(\n                        content=part.get(\"content\", \"\"),\n                        modality=part.get(\"modality\"),\n                        mime_type=part.get(\"media_type\")\n                        if \"media_type\" in part\n                        else part.get(\"mime_type\"),\n                    )\n                )\n            else:\n                # Keep other types as-is\n                converted_parts.append(part)\n\n        input_messages.append(InputMessage(role=role, parts=converted_parts))\n\n    return input_messages\n\n\ndef convert_chatresponse_to_output_messages(\n    chat_response: Any,\n) -> List[OutputMessage]:\n    \"\"\"Convert ChatResponse to opentelemetry-util-genai OutputMessage format.\n\n    This function is used by ExtendedTelemetryHandler which requires OutputMessage objects.\n    \"\"\"\n    output_dicts = get_chatmodel_output_messages(chat_response)\n    if not output_dicts:\n        return []\n\n    output_messages = []\n    for msg_dict in output_dicts:\n        role = msg_dict.get(\"role\", \"assistant\")\n        parts_dicts = msg_dict.get(\"parts\", [])\n        finish_reason = msg_dict.get(\"finish_reason\", \"stop\")\n\n        converted_parts = []\n        for part in parts_dicts:\n            part_type = part.get(\"type\")\n            if part_type == \"text\":\n                converted_parts.append(Text(content=part.get(\"content\", \"\")))\n            elif part_type == \"tool_call\":\n                converted_parts.append(\n                    ToolCall(\n                        id=part.get(\"id\"),\n                        name=part.get(\"name\", \"\"),\n                        arguments=part.get(\"arguments\", {}),\n                    )\n                )\n            else:\n                converted_parts.append(part)\n\n        output_messages.append(\n            OutputMessage(\n                role=role,\n                parts=converted_parts,\n                finish_reason=finish_reason,\n            )\n        )\n\n    return output_messages\n\n\ndef convert_agent_response_to_output_messages(\n    agent_response: Any,\n) -> List[OutputMessage]:\n    \"\"\"Convert agent Msg response to opentelemetry-util-genai OutputMessage format.\n\n    This function is used by ExtendedTelemetryHandler which requires OutputMessage objects.\n    \"\"\"\n    if not hasattr(agent_response, \"content\"):\n        return []\n\n    try:\n        # Use _format_msg_to_parts to convert Msg to dict, then convert to OutputMessage\n        msg_dict = _format_msg_to_parts(agent_response)\n        parts_dicts = msg_dict.get(\"parts\", [])\n\n        converted_parts = []\n        for part in parts_dicts:\n            part_type = part.get(\"type\")\n            if part_type == \"text\":\n                converted_parts.append(Text(content=part.get(\"content\", \"\")))\n            elif part_type == \"reasoning\":\n                converted_parts.append(\n                    Reasoning(content=part.get(\"content\", \"\"))\n                )\n            elif part_type == \"tool_call\":\n                converted_parts.append(\n                    ToolCall(\n                        id=part.get(\"id\"),\n                        name=part.get(\"name\", \"\"),\n                        arguments=part.get(\"arguments\", {}),\n                    )\n                )\n            elif part_type in (\"uri\", \"blob\"):\n                converted_parts.append(part)\n            else:\n                converted_parts.append(Text(content=str(part)))\n\n        if not converted_parts:\n            converted_parts.append(Text(content=\"\"))\n\n        return [\n            OutputMessage(\n                role=\"assistant\",\n                parts=converted_parts,\n                finish_reason=\"stop\",\n            )\n        ]\n    except Exception as e:\n        logger.debug(f\"Failed to convert agent response: {e}\")\n        return []\n"}, "9fb379443103742c16735c603968cf02eb416be4": {"t": 1785758583.4098816, "n": 20143, "h": "188e1504c8753dcd", "c": "# -*- coding: utf-8 -*-\n\"\"\"Attributes processor for span attributes.\"\"\"\n\nfrom __future__ import annotations\n\nimport datetime\nimport enum\nimport inspect\nimport json\nimport logging\nimport mimetypes\nfrom dataclasses import is_dataclass\nfrom enum import Enum\nfrom typing import Any, Dict, List, Optional, Tuple\nfrom urllib.parse import urlparse\n\nfrom agentscope import _config\nfrom agentscope.agent import AgentBase\nfrom agentscope.embedding import EmbeddingModelBase\nfrom agentscope.message import Msg\nfrom agentscope.model import ChatModelBase, ChatResponse\nfrom pydantic import BaseModel\n\nfrom opentelemetry.semconv._incubating.attributes import (\n    gen_ai_attributes as GenAIAttributes,\n)\nfrom opentelemetry.util.genai.extended_types import (\n    EmbeddingInvocation,\n    InvokeAgentInvocation,\n)\nfrom opentelemetry.util.genai.types import (\n    Base64Blob,\n    InputMessage,\n    LLMInvocation,\n    OutputMessage,\n    Reasoning,\n    Text,\n    ToolCall,\n    ToolCallResponse,\n    Uri,\n)\n\nfrom .message_converter import get_message_converter\n\nlogger = logging.getLogger(__name__)\n\n\nclass AgentScopeGenAiProviderName(str, Enum):\n    \"\"\"Extended provider names not in standard OpenTelemetry semantic conventions.\"\"\"\n\n    OLLAMA = \"ollama\"\n    DASHSCOPE = \"dashscope\"\n    MOONSHOT = \"moonshot\"\n\n\n# Provider name mapping based on class names\n_PROVIDER_NAME_MAP = {\n    \"openai\": GenAIAttributes.GenAiProviderNameValues.OPENAI.value,\n    \"gemini\": GenAIAttributes.GenAiProviderNameValues.GCP_GEMINI.value,\n    \"anthropic\": GenAIAttributes.GenAiProviderNameValues.ANTHROPIC.value,\n    \"dashscope\": AgentScopeGenAiProviderName.DASHSCOPE.value,\n    \"ollama\": AgentScopeGenAiProviderName.OLLAMA.value,\n}\n\n# Base URL to provider mapping for OpenAI-compatible APIs\n_BASE_URL_PROVIDER_MAP = [\n    (\"openai.com\", GenAIAttributes.GenAiProviderNameValues.OPENAI.value),\n    (\n        \"api.deepseek.com\",\n        GenAIAttributes.GenAiProviderNameValues.DEEPSEEK.value,\n    ),\n    (\"dashscope.aliyuncs.com\", AgentScopeGenAiProviderName.DASHSCOPE.value),\n]\n\n\ndef get_provider_name(chat_model: ChatModelBase) -> str:\n    \"\"\"Parse chat model provider name\"\"\"\n    classname = chat_model.__class__.__name__\n    prefix = classname.removesuffix(\"ChatModel\").lower()\n\n    # Special handling for DashScopeChatModel with base_http_api_url\n    if (\n        prefix == \"dashscope\"\n        and hasattr(chat_model, \"base_http_api_url\")\n        and chat_model.base_http_api_url\n    ):\n        base_url = chat_model.base_http_api_url\n        for url_fragment, provider in _BASE_URL_PROVIDER_MAP:\n            if url_fragment in base_url:\n                return provider\n\n    return _PROVIDER_NAME_MAP.get(prefix, \"unknown\")\n\n\ndef get_embedding_provider_name(embedding_model: EmbeddingModelBase) -> str:\n    \"\"\"Parse embedding model provider name\"\"\"\n    class_name = embedding_model.__class__.__name__\n\n    if \"DashScope\" in class_name:\n        return AgentScopeGenAiProviderName.DASHSCOPE.value\n    elif \"OpenAI\" in class_name:\n        return GenAIAttributes.GenAiProviderNameValues.OPENAI.value\n    elif \"Gemini\" in class_name:\n        return GenAIAttributes.GenAiProviderNameValues.GCP_GEMINI.value\n    elif \"Ollama\" in class_name:\n        return AgentScopeGenAiProviderName.OLLAMA.value\n    else:\n        model_name = getattr(embedding_model, \"model_name\", \"\")\n        if \"text-embedding\" in model_name:\n            if \"qwen\" in model_name or \"dashscope\" in model_name:\n                return AgentScopeGenAiProviderName.DASHSCOPE.value\n            else:\n                return GenAIAttributes.GenAiProviderNameValues.OPENAI.value\n        return \"agentscope\"\n\n\ndef _get_tool_definitions(\n    tools: list[dict[str, Any]] | None,\n    tool_choice: str | None,\n) -> str | None:\n    \"\"\"Extract and serialize tool definitions for tracing.\"\"\"\n    if tools is None or not isinstance(tools, list) or len(tools) == 0:\n        return None\n\n    if tool_choice == \"none\":\n        return None\n\n    try:\n        flat_tools = []\n        for tool in tools:\n            if not isinstance(tool, dict) or \"function\" not in tool:\n                continue\n\n            func_def = tool[\"function\"]\n            flat_tool = {\n                \"type\": tool.get(\"type\", \"function\"),\n                \"name\": func_def.get(\"name\"),\n                \"description\": func_def.get(\"description\"),\n                \"parameters\": func_def.get(\"parameters\"),\n            }\n            flat_tool = {k: v for k, v in flat_tool.items() if v is not None}\n            flat_tools.append(flat_tool)\n\n        if flat_tools:\n            return _serialize_to_str(flat_tools)\n        return None\n\n    except Exception:\n        return None\n\n\ndef _to_serializable(obj: Any) -> Any:\n    \"\"\"Convert an object to a JSON serializable type.\"\"\"\n    if isinstance(obj, (str, int, bool, float, type(None))):\n        return obj\n    elif isinstance(obj, (list, tuple, set, frozenset)):\n        return [_to_serializable(x) for x in obj]\n    elif isinstance(obj, dict):\n        return {str(key): _to_serializable(val) for (key, val) in obj.items()}\n    elif isinstance(obj, (Msg, BaseModel)) or is_dataclass(obj):\n        return repr(obj)\n    elif inspect.isclass(obj) and issubclass(obj, BaseModel):\n        return repr(obj)\n    elif isinstance(obj, (datetime.date, datetime.datetime, datetime.time)):\n        return obj.isoformat()\n    elif isinstance(obj, datetime.timedelta):\n        return obj.total_seconds()\n    elif isinstance(obj, enum.Enum):\n        return _to_serializable(obj.value)\n    else:\n        return str(obj)\n\n\ndef _serialize_to_str(value: Any) -> str:\n    \"\"\"Serialize value to JSON string\"\"\"\n    try:\n        return json.dumps(value, ensure_ascii=False)\n    except TypeError:\n        return json.dumps(_to_serializable(value), ensure_ascii=False)\n\n\ndef _convert_block_to_part(block: Dict[str, Any]) -> Dict[str, Any] | None:\n    \"\"\"Convert content block to standardized part format.\"\"\"\n    block_type = block.get(\"type\")\n\n    if block_type == \"text\":\n        return {\"type\": \"text\", \"content\": block.get(\"text\", \"\")}\n\n    elif block_type == \"thinking\":\n        return {\n            \"type\": \"text\",\n            \"content\": f\"[Thinking] {block.get('thinking', '')}\",\n        }\n\n    elif block_type == \"tool_use\":\n        return {\n            \"type\": \"tool_call\",\n            \"id\": block.get(\"id\", \"\"),\n            \"name\": block.get(\"name\", \"\"),\n            \"arguments\": block.get(\"input\", {}),\n        }\n\n    elif block_type == \"tool_result\":\n        output = block.get(\"output\", \"\")\n        result = (\n            _serialize_to_str(output)\n            if isinstance(output, (list, dict))\n            else str(output)\n        )\n        return {\n            \"type\": \"tool_call_response\",\n            \"id\": block.get(\"id\", \"\"),\n            \"result\": result,\n        }\n\n    elif block_type in [\"image\", \"audio\", \"video\"]:\n        source = block.get(\"source\", {})\n        source_type = source.get(\"type\")\n        media_type = source.get(\"media_type\")\n\n        if source_type == \"url\":\n            url = source.get(\"url\", \"\")\n            # Infer mime_type from URL extension if not provided\n            if not media_type:\n                parsed_url = urlparse(url)\n                media_type, _ = mimetypes.guess_type(parsed_url.path)\n            return {\n                \"type\": \"uri\",\n                \"uri\": url,\n                \"modality\": block_type,\n                \"mime_type\": media_type,\n            }\n        elif source_type == \"base64\":\n            if not media_type:\n                default_media_types = {\n                    \"image\": \"image/jpeg\",\n                    \"audio\": \"audio/wav\",\n                    \"video\": \"video/mp4\",\n                }\n                media_type = default_media_types.get(block_type, \"unknown\")\n            return {\n                \"type\": \"blob\",\n                \"content\": source.get(\"data\", \"\"),\n                \"media_type\": media_type,\n                \"modality\": block_type,\n            }\n\n    return None\n\n\ndef _format_msg_to_parts(msg: Msg) -> dict[str, Any]:\n    \"\"\"Convert Msg to standard format (parts structure)\"\"\"\n    try:\n        parts = []\n        for block in msg.get_content_blocks():\n            part = _convert_block_to_part(block)\n            if part:\n                parts.append(part)\n\n        formatted_msg = {\"role\": msg.role, \"parts\": parts}\n        if msg.name:\n            formatted_msg[\"name\"] = msg.name\n\n        return formatted_msg\n\n    except Exception as e:\n        logger.debug(f\"Error formatting message: {e}\")\n        return {\n            \"role\": msg.role,\n            \"parts\": [\n                {\n                    \"type\": \"text\",\n                    \"content\": str(msg.content) if msg.content else \"\",\n                }\n            ],\n        }\n\n\ndef create_llm_invocation(\n    call_instance: ChatModelBase,\n    call_args: Tuple[Any, ...],\n    call_kwargs: Dict[str, Any],\n) -> LLMInvocation:\n    \"\"\"Create LLM invocation from call context.\"\"\"\n    provider_name = get_provider_name(call_instance)\n    request_model = getattr(call_instance, \"model_name\", \"unknown_model\")\n\n    # Convert input messages\n    input_messages = []\n    messages = call_args[0] if call_args else call_kwargs.get(\"messages\")\n    if messages:\n        try:\n            converted_messages = get_message_converter(provider_name)(messages)\n            input_messages = convert_agentscope_messages_to_genai_format(\n                converted_messages, provider_name=provider_name\n            )\n        except Exception as e:\n            logger.warning(\n                f\"Failed to convert input messages: {e}\", exc_info=True\n            )\n\n    invocation = LLMInvocation(\n        request_model=request_model,\n        provider=provider_name,\n        input_messages=input_messages,\n    )\n\n    # Set optional request parameters if present\n    if call_kwargs.get(\"max_tokens\"):\n        invocation.max_tokens = call_kwargs[\"max_tokens\"]\n    if call_kwargs.get(\"temperature\"):\n        invocation.temperature = call_kwargs[\"temperature\"]\n    if call_kwargs.get(\"top_p\"):\n        invocation.top_p = call_kwargs[\"top_p\"]\n\n    # Set tool definitions if applicable\n    if not call_kwargs.get(\"structured_model\", False):\n        tool_definitions = _get_tool_definitions(\n            call_kwargs.get(\"tools\"), call_kwargs.get(\"tool_choice\")\n        )\n        if tool_definitions:\n            invocation.attributes[\"gen_ai.request.tool_definitions\"] = (\n                tool_definitions\n            )\n\n    return invocation\n\n\ndef create_embedding_invocation(\n    call_instance: EmbeddingModelBase,\n    call_args: Tuple[Any, ...],\n    call_kwargs: Dict[str, Any],\n) -> EmbeddingInvocation:\n    \"\"\"Create Embedding invocation from call context.\"\"\"\n    provider_name = get_embedding_provider_name(call_instance)\n    request_model = getattr(call_instance, \"model_name\", \"unknown_model\")\n\n    invocation = EmbeddingInvocation(\n        request_model=request_model,\n        provider=provider_name,\n    )\n\n    # Set encoding formats if present\n    if call_kwargs.get(\"encoding_formats\"):\n        invocation.encoding_formats = call_kwargs[\"encoding_formats\"]\n\n    return invocation\n\n\ndef create_agent_invocation(\n    reply_instance: AgentBase,\n    reply_args: Tuple[Any, ...],\n    reply_kwargs: Dict[str, Any],\n) -> InvokeAgentInvocation:\n    \"\"\"Create Agent invocation from reply call context.\"\"\"\n    # Get provider and model info\n    provider_name = None\n    request_model = None\n    if hasattr(reply_instance, \"model\") and reply_instance.model:\n        model = reply_instance.model\n        request_model = getattr(model, \"model_name\", None)\n        if isinstance(model, ChatModelBase):\n            provider_name = get_provider_name(model)\n\n    # Convert input messages\n    input_messages = []\n    msg = reply_args[0] if reply_args else reply_kwargs.get(\"msg\")\n    if msg:\n        try:\n            if isinstance(msg, Msg):\n                input_messages = convert_agentscope_messages_to_genai_format(\n                    [msg]\n                )\n            elif isinstance(msg, list):\n                input_messages = convert_agentscope_messages_to_genai_format(\n                    msg\n                )\n        except Exception as e:\n            logger.debug(f\"Error converting agent input messages: {e}\")\n\n    invocation = InvokeAgentInvocation(\n        provider=provider_name,\n        agent_name=getattr(reply_instance, \"name\", \"unknown_agent\"),\n        agent_id=getattr(reply_instance, \"id\", \"unknown\"),\n        agent_description=inspect.getdoc(reply_instance.__class__)\n        or \"No description available\",\n        conversation_id=_config.run_id,\n        request_model=request_model,\n        input_messages=input_messages,\n    )\n\n    # Set system instruction if available\n    if hasattr(reply_instance, \"sys_prompt\") and reply_instance.sys_prompt:\n        sys_prompt = reply_instance.sys_prompt\n        if isinstance(sys_prompt, str):\n            invocation.system_instruction = [Text(content=sys_prompt)]\n        elif isinstance(sys_prompt, list):\n            invocation.system_instruction = sys_prompt\n\n    return invocation\n\n\ndef get_chatmodel_output_messages(chat_response: Any) -> List[Dict[str, Any]]:\n    \"\"\"Convert ChatResponse to OpenTelemetry standard output message format.\"\"\"\n    try:\n        if not isinstance(chat_response, ChatResponse):\n            return []\n\n        parts = []\n        for block in chat_response.content:\n            part = _convert_block_to_part(block)\n            if part:\n                parts.append(part)\n\n        if not parts:\n            parts.append({\"type\": \"text\", \"content\": \"\"})\n\n        return [\n            {\n                \"role\": \"assistant\",\n                \"parts\": parts,\n                \"finish_reason\": \"stop\",\n            }\n        ]\n\n    except Exception as e:\n        logger.warning(\n            f\"Error processing ChatResponse to output messages: {e}\"\n        )\n        return [\n            {\n                \"role\": \"assistant\",\n                \"parts\": [\n                    {\"type\": \"text\", \"content\": \"<error processing response>\"}\n                ],\n                \"finish_reason\": \"error\",\n            }\n        ]\n\n\ndef convert_agentscope_messages_to_genai_format(\n    messages: Any,\n    provider_name: Optional[str] = None,\n) -> List[InputMessage]:\n    \"\"\"Convert AgentScope messages to opentelemetry-util-genai InputMessage format.\n       parse _convert_block_to_part result from agentscope/tracing/_converter.py\n\n    This function is used by ExtendedTelemetryHandler which requires InputMessage objects.\n    \"\"\"\n    if not messages:\n        return []\n\n    if not isinstance(messages, list):\n        messages = [messages]\n\n    input_messages = []\n    for msg in messages:\n        if isinstance(msg, Msg):\n            msg_dict = _format_msg_to_parts(msg)\n        elif isinstance(msg, dict):\n            msg_dict = msg\n        else:\n            continue\n\n        role = msg_dict.get(\"role\", \"user\")\n        parts = msg_dict.get(\"parts\", [])\n\n        converted_parts = []\n        for part in parts:\n            part_type = part.get(\"type\")\n            if part_type == \"text\":\n                converted_parts.append(Text(content=part.get(\"content\", \"\")))\n            elif part_type == \"tool_call\":\n                converted_parts.append(\n                    ToolCall(\n                        id=part.get(\"id\"),\n                        name=part.get(\"name\", \"\"),\n                        arguments=part.get(\"arguments\", {}),\n                    )\n                )\n            elif part_type == \"tool_call_response\":\n                converted_parts.append(\n                    ToolCallResponse(\n                        id=part.get(\"id\"),\n                        response=part[\"response\"]\n                        if \"response\" in part\n                        else part.get(\"result\", \"\"),\n                    )\n                )\n            elif part_type == \"reasoning\":\n                converted_parts.append(\n                    Reasoning(content=part.get(\"content\", \"\"))\n                )\n            elif part_type == \"uri\":\n                converted_parts.append(\n                    Uri(\n                        uri=part.get(\"uri\"),\n                        modality=part.get(\"modality\"),\n                        mime_type=part.get(\"mime_type\"),\n                    )\n                )\n            elif part_type == \"blob\":\n                converted_parts.append(\n                    Base64Blob(\n                        content=part.get(\"content\", \"\"),\n                        modality=part.get(\"modality\"),\n                        mime_type=part.get(\"media_type\")\n                        if \"media_type\" in part\n                        else part.get(\"mime_type\"),\n                    )\n                )\n            else:\n                # Keep other types as-is\n                converted_parts.append(part)\n\n        input_messages.append(InputMessage(role=role, parts=converted_parts))\n\n    return input_messages\n\n\ndef convert_chatresponse_to_output_messages(\n    chat_response: Any,\n) -> List[OutputMessage]:\n    \"\"\"Convert ChatResponse to opentelemetry-util-genai OutputMessage format.\n\n    This function is used by ExtendedTelemetryHandler which requires OutputMessage objects.\n    \"\"\"\n    output_dicts = get_chatmodel_output_messages(chat_response)\n    if not output_dicts:\n        return []\n\n    output_messages = []\n    for msg_dict in output_dicts:\n        role = msg_dict.get(\"role\", \"assistant\")\n        parts_dicts = msg_dict.get(\"parts\", [])\n        finish_reason = msg_dict.get(\"finish_reason\", \"stop\")\n\n        converted_parts = []\n        for part in parts_dicts:\n            part_type = part.get(\"type\")\n            if part_type == \"text\":\n                converted_parts.append(Text(content=part.get(\"content\", \"\")))\n            elif part_type == \"tool_call\":\n                converted_parts.append(\n                    ToolCall(\n                        id=part.get(\"id\"),\n                        name=part.get(\"name\", \"\"),\n                        arguments=part.get(\"arguments\", {}),\n                    )\n                )\n            else:\n                converted_parts.append(part)\n\n        output_messages.append(\n            OutputMessage(\n                role=role,\n                parts=converted_parts,\n                finish_reason=finish_reason,\n            )\n        )\n\n    return output_messages\n\n\ndef convert_agent_response_to_output_messages(\n    agent_response: Any,\n) -> List[OutputMessage]:\n    \"\"\"Convert agent Msg response to opentelemetry-util-genai OutputMessage format.\n\n    This function is used by ExtendedTelemetryHandler which requires OutputMessage objects.\n    \"\"\"\n    if not hasattr(agent_response, \"content\"):\n        return []\n\n    try:\n        # Use _format_msg_to_parts to convert Msg to dict, then convert to OutputMessage\n        msg_dict = _format_msg_to_parts(agent_response)\n        parts_dicts = msg_dict.get(\"parts\", [])\n\n        converted_parts = []\n        for part in parts_dicts:\n            part_type = part.get(\"type\")\n            if part_type == \"text\":\n                converted_parts.append(Text(content=part.get(\"content\", \"\")))\n            elif part_type == \"reasoning\":\n                converted_parts.append(\n                    Reasoning(content=part.get(\"content\", \"\"))\n                )\n            elif part_type == \"tool_call\":\n                converted_parts.append(\n                    ToolCall(\n                        id=part.get(\"id\"),\n                        name=part.get(\"name\", \"\"),\n                        arguments=part.get(\"arguments\", {}),\n                    )\n                )\n            elif part_type in (\"uri\", \"blob\"):\n                converted_parts.append(part)\n            else:\n                converted_parts.append(Text(content=str(part)))\n\n        if not converted_parts:\n            converted_parts.append(Text(content=\"\"))\n\n        return [\n            OutputMessage(\n                role=\"assistant\",\n                parts=converted_parts,\n                finish_reason=\"stop\",\n            )\n        ]\n    except Exception as e:\n        logger.debug(f\"Failed to convert agent response: {e}\")\n        return []\n"}, "e33feb8bb40d4a1cae610b6a95fcc19f5648ccdf": {"t": 1785758584.986063, "n": 19319, "h": "f3dca8c0250a0cc2", "c": "# -*- coding: utf-8 -*-\n\"\"\"Attributes processor for span attributes.\"\"\"\n\nfrom __future__ import annotations\n\nimport datetime\nimport enum\nimport inspect\nimport json\nimport logging\nimport mimetypes\nfrom dataclasses import is_dataclass\nfrom enum import Enum\nfrom typing import Any, Dict, List, Optional, Tuple\nfrom urllib.parse import urlparse\n\nfrom agentscope import _config\nfrom agentscope.agent import AgentBase\nfrom agentscope.embedding import EmbeddingModelBase\nfrom agentscope.message import Msg\nfrom agentscope.model import ChatModelBase, ChatResponse\nfrom pydantic import BaseModel\n\nfrom opentelemetry.semconv._incubating.attributes import (\n    gen_ai_attributes as GenAIAttributes,\n)\nfrom opentelemetry.util.genai.extended_types import (\n    EmbeddingInvocation,\n    InvokeAgentInvocation,\n)\nfrom opentelemetry.util.genai.types import (\n    InputMessage,\n    LLMInvocation,\n    OutputMessage,\n    Reasoning,\n    Text,\n    ToolCall,\n    ToolCallResponse,\n)\n\nfrom .message_converter import get_message_converter\n\nlogger = logging.getLogger(__name__)\n\n\nclass AgentScopeGenAiProviderName(str, Enum):\n    \"\"\"Extended provider names not in standard OpenTelemetry semantic conventions.\"\"\"\n\n    OLLAMA = \"ollama\"\n    DASHSCOPE = \"dashscope\"\n    MOONSHOT = \"moonshot\"\n\n\n# Provider name mapping based on class names\n_PROVIDER_NAME_MAP = {\n    \"openai\": GenAIAttributes.GenAiProviderNameValues.OPENAI.value,\n    \"gemini\": GenAIAttributes.GenAiProviderNameValues.GCP_GEMINI.value,\n    \"anthropic\": GenAIAttributes.GenAiProviderNameValues.ANTHROPIC.value,\n    \"dashscope\": AgentScopeGenAiProviderName.DASHSCOPE.value,\n    \"ollama\": AgentScopeGenAiProviderName.OLLAMA.value,\n}\n\n# Base URL to provider mapping for OpenAI-compatible APIs\n_BASE_URL_PROVIDER_MAP = [\n    (\"openai.com\", GenAIAttributes.GenAiProviderNameValues.OPENAI.value),\n    (\n        \"api.deepseek.com\",\n        GenAIAttributes.GenAiProviderNameValues.DEEPSEEK.value,\n    ),\n    (\"dashscope.aliyuncs.com\", AgentScopeGenAiProviderName.DASHSCOPE.value),\n]\n\n\ndef get_provider_name(chat_model: ChatModelBase) -> str:\n    \"\"\"Parse chat model provider name\"\"\"\n    classname = chat_model.__class__.__name__\n    prefix = classname.removesuffix(\"ChatModel\").lower()\n\n    # Special handling for DashScopeChatModel with base_http_api_url\n    if (\n        prefix == \"dashscope\"\n        and hasattr(chat_model, \"base_http_api_url\")\n        and chat_model.base_http_api_url\n    ):\n        base_url = chat_model.base_http_api_url\n        for url_fragment, provider in _BASE_URL_PROVIDER_MAP:\n            if url_fragment in base_url:\n                return provider\n\n    return _PROVIDER_NAME_MAP.get(prefix, \"unknown\")\n\n\ndef get_embedding_provider_name(embedding_model: EmbeddingModelBase) -> str:\n    \"\"\"Parse embedding model provider name\"\"\"\n    class_name = embedding_model.__class__.__name__\n\n    if \"DashScope\" in class_name:\n        return AgentScopeGenAiProviderName.DASHSCOPE.value\n    elif \"OpenAI\" in class_name:\n        return GenAIAttributes.GenAiProviderNameValues.OPENAI.value\n    elif \"Gemini\" in class_name:\n        return GenAIAttributes.GenAiProviderNameValues.GCP_GEMINI.value\n    elif \"Ollama\" in class_name:\n        return AgentScopeGenAiProviderName.OLLAMA.value\n    else:\n        model_name = getattr(embedding_model, \"model_name\", \"\")\n        if \"text-embedding\" in model_name:\n            if \"qwen\" in model_name or \"dashscope\" in model_name:\n                return AgentScopeGenAiProviderName.DASHSCOPE.value\n            else:\n                return GenAIAttributes.GenAiProviderNameValues.OPENAI.value\n        return \"agentscope\"\n\n\ndef _get_tool_definitions(\n    tools: list[dict[str, Any]] | None,\n    tool_choice: str | None,\n) -> str | None:\n    \"\"\"Extract and serialize tool definitions for tracing.\"\"\"\n    if tools is None or not isinstance(tools, list) or len(tools) == 0:\n        return None\n\n    if tool_choice == \"none\":\n        return None\n\n    try:\n        flat_tools = []\n        for tool in tools:\n            if not isinstance(tool, dict) or \"function\" not in tool:\n                continue\n\n            func_def = tool[\"function\"]\n            flat_tool = {\n                \"type\": tool.get(\"type\", \"function\"),\n                \"name\": func_def.get(\"name\"),\n                \"description\": func_def.get(\"description\"),\n                \"parameters\": func_def.get(\"parameters\"),\n            }\n            flat_tool = {k: v for k, v in flat_tool.items() if v is not None}\n            flat_tools.append(flat_tool)\n\n        if flat_tools:\n            return _serialize_to_str(flat_tools)\n        return None\n\n    except Exception:\n        return None\n\n\ndef _to_serializable(obj: Any) -> Any:\n    \"\"\"Convert an object to a JSON serializable type.\"\"\"\n    if isinstance(obj, (str, int, bool, float, type(None))):\n        return obj\n    elif isinstance(obj, (list, tuple, set, frozenset)):\n        return [_to_serializable(x) for x in obj]\n    elif isinstance(obj, dict):\n        return {str(key): _to_serializable(val) for (key, val) in obj.items()}\n    elif isinstance(obj, (Msg, BaseModel)) or is_dataclass(obj):\n        return repr(obj)\n    elif inspect.isclass(obj) and issubclass(obj, BaseModel):\n        return repr(obj)\n    elif isinstance(obj, (datetime.date, datetime.datetime, datetime.time)):\n        return obj.isoformat()\n    elif isinstance(obj, datetime.timedelta):\n        return obj.total_seconds()\n    elif isinstance(obj, enum.Enum):\n        return _to_serializable(obj.value)\n    else:\n        return str(obj)\n\n\ndef _serialize_to_str(value: Any) -> str:\n    \"\"\"Serialize value to JSON string\"\"\"\n    try:\n        return json.dumps(value, ensure_ascii=False)\n    except TypeError:\n        return json.dumps(_to_serializable(value), ensure_ascii=False)\n\n\ndef _convert_block_to_part(block: Dict[str, Any]) -> Dict[str, Any] | None:\n    \"\"\"Convert content block to standardized part format.\"\"\"\n    block_type = block.get(\"type\")\n\n    if block_type == \"text\":\n        return {\"type\": \"text\", \"content\": block.get(\"text\", \"\")}\n\n    elif block_type == \"thinking\":\n        return {\n            \"type\": \"text\",\n            \"content\": f\"[Thinking] {block.get('thinking', '')}\",\n        }\n\n    elif block_type == \"tool_use\":\n        return {\n            \"type\": \"tool_call\",\n            \"id\": block.get(\"id\", \"\"),\n            \"name\": block.get(\"name\", \"\"),\n            \"arguments\": block.get(\"input\", {}),\n        }\n\n    elif block_type == \"tool_result\":\n        output = block.get(\"output\", \"\")\n        result = (\n            _serialize_to_str(output)\n            if isinstance(output, (list, dict))\n            else str(output)\n        )\n        return {\n            \"type\": \"tool_call_response\",\n            \"id\": block.get(\"id\", \"\"),\n            \"result\": result,\n        }\n\n    elif block_type in [\"image\", \"audio\", \"video\"]:\n        source = block.get(\"source\", {})\n        source_type = source.get(\"type\")\n        media_type = source.get(\"media_type\")\n\n        if source_type == \"url\":\n            url = source.get(\"url\", \"\")\n            # Infer mime_type from URL extension if not provided\n            if not media_type:\n                parsed_url = urlparse(url)\n                media_type, _ = mimetypes.guess_type(parsed_url.path)\n            return {\n                \"type\": \"uri\",\n                \"uri\": url,\n                \"modality\": block_type,\n                \"mime_type\": media_type,\n            }\n        elif source_type == \"base64\":\n            if not media_type:\n                default_media_types = {\n                    \"image\": \"image/jpeg\",\n                    \"audio\": \"audio/wav\",\n                    \"video\": \"video/mp4\",\n                }\n                media_type = default_media_types.get(block_type, \"unknown\")\n            return {\n                \"type\": \"blob\",\n                \"content\": source.get(\"data\", \"\"),\n                \"media_type\": media_type,\n                \"modality\": block_type,\n            }\n\n    return None\n\n\ndef _format_msg_to_parts(msg: Msg) -> dict[str, Any]:\n    \"\"\"Convert Msg to standard format (parts structure)\"\"\"\n    try:\n        parts = []\n        for block in msg.get_content_blocks():\n            part = _convert_block_to_part(block)\n            if part:\n                parts.append(part)\n\n        formatted_msg = {\"role\": msg.role, \"parts\": parts}\n        if msg.name:\n            formatted_msg[\"name\"] = msg.name\n\n        return formatted_msg\n\n    except Exception as e:\n        logger.debug(f\"Error formatting message: {e}\")\n        return {\n            \"role\": msg.role,\n            \"parts\": [\n                {\n                    \"type\": \"text\",\n                    \"content\": str(msg.content) if msg.content else \"\",\n                }\n            ],\n        }\n\n\ndef create_llm_invocation(\n    call_instance: ChatModelBase,\n    call_args: Tuple[Any, ...],\n    call_kwargs: Dict[str, Any],\n) -> LLMInvocation:\n    \"\"\"Create LLM invocation from call context.\"\"\"\n    provider_name = get_provider_name(call_instance)\n    request_model = getattr(call_instance, \"model_name\", \"unknown_model\")\n\n    # Convert input messages\n    input_messages = []\n    messages = call_args[0] if call_args else call_kwargs.get(\"messages\")\n    if messages:\n        try:\n            converted_messages = get_message_converter(provider_name)(messages)\n            input_messages = convert_agentscope_messages_to_genai_format(\n                converted_messages, provider_name=provider_name\n            )\n        except Exception as e:\n            logger.warning(\n                f\"Failed to convert input messages: {e}\", exc_info=True\n            )\n\n    invocation = LLMInvocation(\n        request_model=request_model,\n        provider=provider_name,\n        input_messages=input_messages,\n    )\n\n    # Set optional request parameters if present\n    if call_kwargs.get(\"max_tokens\"):\n        invocation.max_tokens = call_kwargs[\"max_tokens\"]\n    if call_kwargs.get(\"temperature\"):\n        invocation.temperature = call_kwargs[\"temperature\"]\n    if call_kwargs.get(\"top_p\"):\n        invocation.top_p = call_kwargs[\"top_p\"]\n\n    # Set tool definitions if applicable\n    if not call_kwargs.get(\"structured_model\", False):\n        tool_definitions = _get_tool_definitions(\n            call_kwargs.get(\"tools\"), call_kwargs.get(\"tool_choice\")\n        )\n        if tool_definitions:\n            invocation.attributes[\"gen_ai.request.tool_definitions\"] = (\n                tool_definitions\n            )\n\n    return invocation\n\n\ndef create_embedding_invocation(\n    call_instance: EmbeddingModelBase,\n    call_args: Tuple[Any, ...],\n    call_kwargs: Dict[str, Any],\n) -> EmbeddingInvocation:\n    \"\"\"Create Embedding invocation from call context.\"\"\"\n    provider_name = get_embedding_provider_name(call_instance)\n    request_model = getattr(call_instance, \"model_name\", \"unknown_model\")\n\n    invocation = EmbeddingInvocation(\n        request_model=request_model,\n        provider=provider_name,\n    )\n\n    # Set encoding formats if present\n    if call_kwargs.get(\"encoding_formats\"):\n        invocation.encoding_formats = call_kwargs[\"encoding_formats\"]\n\n    return invocation\n\n\ndef create_agent_invocation(\n    reply_instance: AgentBase,\n    reply_args: Tuple[Any, ...],\n    reply_kwargs: Dict[str, Any],\n) -> InvokeAgentInvocation:\n    \"\"\"Create Agent invocation from reply call context.\"\"\"\n    # Get provider and model info\n    provider_name = None\n    request_model = None\n    if hasattr(reply_instance, \"model\") and reply_instance.model:\n        model = reply_instance.model\n        request_model = getattr(model, \"model_name\", None)\n        if isinstance(model, ChatModelBase):\n            provider_name = get_provider_name(model)\n\n    # Convert input messages\n    input_messages = []\n    msg = reply_args[0] if reply_args else reply_kwargs.get(\"msg\")\n    if msg:\n        try:\n            if isinstance(msg, Msg):\n                input_messages = convert_agentscope_messages_to_genai_format(\n                    [msg]\n                )\n            elif isinstance(msg, list):\n                input_messages = convert_agentscope_messages_to_genai_format(\n                    msg\n                )\n        except Exception as e:\n            logger.debug(f\"Error converting agent input messages: {e}\")\n\n    invocation = InvokeAgentInvocation(\n        provider=provider_name,\n        agent_name=getattr(reply_instance, \"name\", \"unknown_agent\"),\n        agent_id=getattr(reply_instance, \"id\", \"unknown\"),\n        agent_description=inspect.getdoc(reply_instance.__class__)\n        or \"No description available\",\n        conversation_id=_config.run_id,\n        request_model=request_model,\n        input_messages=input_messages,\n    )\n\n    # Set system instruction if available\n    if hasattr(reply_instance, \"sys_prompt\") and reply_instance.sys_prompt:\n        sys_prompt = reply_instance.sys_prompt\n        if isinstance(sys_prompt, str):\n            invocation.system_instruction = [Text(content=sys_prompt)]\n        elif isinstance(sys_prompt, list):\n            invocation.system_instruction = sys_prompt\n\n    return invocation\n\n\ndef get_chatmodel_output_messages(chat_response: Any) -> List[Dict[str, Any]]:\n    \"\"\"Convert ChatResponse to OpenTelemetry standard output message format.\"\"\"\n    try:\n        if not isinstance(chat_response, ChatResponse):\n            return []\n\n        parts = []\n        for block in chat_response.content:\n            part = _convert_block_to_part(block)\n            if part:\n                parts.append(part)\n\n        if not parts:\n            parts.append({\"type\": \"text\", \"content\": \"\"})\n\n        return [\n            {\n                \"role\": \"assistant\",\n                \"parts\": parts,\n                \"finish_reason\": \"stop\",\n            }\n        ]\n\n    except Exception as e:\n        logger.warning(\n            f\"Error processing ChatResponse to output messages: {e}\"\n        )\n        return [\n            {\n                \"role\": \"assistant\",\n                \"parts\": [\n                    {\"type\": \"text\", \"content\": \"<error processing response>\"}\n                ],\n                \"finish_reason\": \"error\",\n            }\n        ]\n\n\ndef convert_agentscope_messages_to_genai_format(\n    messages: Any,\n    provider_name: Optional[str] = None,\n) -> List[InputMessage]:\n    \"\"\"Convert AgentScope messages to opentelemetry-util-genai InputMessage format.\n\n    This function is used by ExtendedTelemetryHandler which requires InputMessage objects.\n    \"\"\"\n    if not messages:\n        return []\n\n    if not isinstance(messages, list):\n        messages = [messages]\n\n    input_messages = []\n    for msg in messages:\n        if isinstance(msg, Msg):\n            msg_dict = _format_msg_to_parts(msg)\n        elif isinstance(msg, dict):\n            msg_dict = msg\n        else:\n            continue\n\n        role = msg_dict.get(\"role\", \"user\")\n        parts = msg_dict.get(\"parts\", [])\n\n        converted_parts = []\n        for part in parts:\n            part_type = part.get(\"type\")\n            if part_type == \"text\":\n                converted_parts.append(Text(content=part.get(\"content\", \"\")))\n            elif part_type == \"tool_call\":\n                converted_parts.append(\n                    ToolCall(\n                        id=part.get(\"id\"),\n                        name=part.get(\"name\", \"\"),\n                        arguments=part.get(\"arguments\", {}),\n                    )\n                )\n            elif part_type == \"tool_call_response\":\n                converted_parts.append(\n                    ToolCallResponse(\n                        id=part.get(\"id\"),\n                        response=part.get(\"result\", \"\"),\n                    )\n                )\n            elif part_type == \"reasoning\":\n                converted_parts.append(\n                    Reasoning(content=part.get(\"content\", \"\"))\n                )\n            elif part_type in (\"uri\", \"blob\"):\n                converted_parts.append(part)\n            else:\n                # Keep other types as-is\n                converted_parts.append(part)\n\n        input_messages.append(InputMessage(role=role, parts=converted_parts))\n\n    return input_messages\n\n\ndef convert_chatresponse_to_output_messages(\n    chat_response: Any,\n) -> List[OutputMessage]:\n    \"\"\"Convert ChatResponse to opentelemetry-util-genai OutputMessage format.\n\n    This function is used by ExtendedTelemetryHandler which requires OutputMessage objects.\n    \"\"\"\n    output_dicts = get_chatmodel_output_messages(chat_response)\n    if not output_dicts:\n        return []\n\n    output_messages = []\n    for msg_dict in output_dicts:\n        role = msg_dict.get(\"role\", \"assistant\")\n        parts_dicts = msg_dict.get(\"parts\", [])\n        finish_reason = msg_dict.get(\"finish_reason\", \"stop\")\n\n        converted_parts = []\n        for part in parts_dicts:\n            part_type = part.get(\"type\")\n            if part_type == \"text\":\n                converted_parts.append(Text(content=part.get(\"content\", \"\")))\n            elif part_type == \"tool_call\":\n                converted_parts.append(\n                    ToolCall(\n                        id=part.get(\"id\"),\n                        name=part.get(\"name\", \"\"),\n                        arguments=part.get(\"arguments\", {}),\n                    )\n                )\n            else:\n                converted_parts.append(part)\n\n        output_messages.append(\n            OutputMessage(\n                role=role,\n                parts=converted_parts,\n                finish_reason=finish_reason,\n            )\n        )\n\n    return output_messages\n\n\ndef convert_agent_response_to_output_messages(\n    agent_response: Any,\n) -> List[OutputMessage]:\n    \"\"\"Convert agent Msg response to opentelemetry-util-genai OutputMessage format.\n\n    This function is used by ExtendedTelemetryHandler which requires OutputMessage objects.\n    \"\"\"\n    if not hasattr(agent_response, \"content\"):\n        return []\n\n    try:\n        # Use _format_msg_to_parts to convert Msg to dict, then convert to OutputMessage\n        msg_dict = _format_msg_to_parts(agent_response)\n        parts_dicts = msg_dict.get(\"parts\", [])\n\n        converted_parts = []\n        for part in parts_dicts:\n            part_type = part.get(\"type\")\n            if part_type == \"text\":\n                converted_parts.append(Text(content=part.get(\"content\", \"\")))\n            elif part_type == \"reasoning\":\n                converted_parts.append(\n                    Reasoning(content=part.get(\"content\", \"\"))\n                )\n            elif part_type == \"tool_call\":\n                converted_parts.append(\n                    ToolCall(\n                        id=part.get(\"id\"),\n                        name=part.get(\"name\", \"\"),\n                        arguments=part.get(\"arguments\", {}),\n                    )\n                )\n            elif part_type in (\"uri\", \"blob\"):\n                converted_parts.append(part)\n            else:\n                converted_parts.append(Text(content=str(part)))\n\n        if not converted_parts:\n            converted_parts.append(Text(content=\"\"))\n\n        return [\n            OutputMessage(\n                role=\"assistant\",\n                parts=converted_parts,\n                finish_reason=\"stop\",\n            )\n        ]\n    except Exception as e:\n        logger.debug(f\"Failed to convert agent response: {e}\")\n        return []\n"}, "31f28b475dfec73e0954a7596b86f0e48a66cd9f": {"t": 1785758586.554759, "n": 19230, "h": "2ff23aa05b7e967c", "c": "# -*- coding: utf-8 -*-\n\"\"\"Attributes processor for span attributes.\"\"\"\n\nfrom __future__ import annotations\n\nimport datetime\nimport enum\nimport inspect\nimport json\nimport logging\nimport mimetypes\nfrom dataclasses import is_dataclass\nfrom enum import Enum\nfrom typing import Any, Dict, List, Optional, Tuple\n\nfrom agentscope import _config\nfrom agentscope.agent import AgentBase\nfrom agentscope.embedding import EmbeddingModelBase\nfrom agentscope.message import Msg\nfrom agentscope.model import ChatModelBase, ChatResponse\nfrom pydantic import BaseModel\n\nfrom opentelemetry.semconv._incubating.attributes import (\n    gen_ai_attributes as GenAIAttributes,\n)\nfrom opentelemetry.util.genai.extended_types import (\n    EmbeddingInvocation,\n    InvokeAgentInvocation,\n)\nfrom opentelemetry.util.genai.types import (\n    InputMessage,\n    LLMInvocation,\n    OutputMessage,\n    Reasoning,\n    Text,\n    ToolCall,\n    ToolCallResponse,\n)\n\nfrom .message_converter import get_message_converter\n\nlogger = logging.getLogger(__name__)\n\n\nclass AgentScopeGenAiProviderName(str, Enum):\n    \"\"\"Extended provider names not in standard OpenTelemetry semantic conventions.\"\"\"\n\n    OLLAMA = \"ollama\"\n    DASHSCOPE = \"dashscope\"\n    MOONSHOT = \"moonshot\"\n\n\n# Provider name mapping based on class names\n_PROVIDER_NAME_MAP = {\n    \"openai\": GenAIAttributes.GenAiProviderNameValues.OPENAI.value,\n    \"gemini\": GenAIAttributes.GenAiProviderNameValues.GCP_GEMINI.value,\n    \"anthropic\": GenAIAttributes.GenAiProviderNameValues.ANTHROPIC.value,\n    \"dashscope\": AgentScopeGenAiProviderName.DASHSCOPE.value,\n    \"ollama\": AgentScopeGenAiProviderName.OLLAMA.value,\n}\n\n# Base URL to provider mapping for OpenAI-compatible APIs\n_BASE_URL_PROVIDER_MAP = [\n    (\"openai.com\", GenAIAttributes.GenAiProviderNameValues.OPENAI.value),\n    (\n        \"api.deepseek.com\",\n        GenAIAttributes.GenAiProviderNameValues.DEEPSEEK.value,\n    ),\n    (\"dashscope.aliyuncs.com\", AgentScopeGenAiProviderName.DASHSCOPE.value),\n]\n\n\ndef get_provider_name(chat_model: ChatModelBase) -> str:\n    \"\"\"Parse chat model provider name\"\"\"\n    classname = chat_model.__class__.__name__\n    prefix = classname.removesuffix(\"ChatModel\").lower()\n\n    # Special handling for DashScopeChatModel with base_http_api_url\n    if (\n        prefix == \"dashscope\"\n        and hasattr(chat_model, \"base_http_api_url\")\n        and chat_model.base_http_api_url\n    ):\n        base_url = chat_model.base_http_api_url\n        for url_fragment, provider in _BASE_URL_PROVIDER_MAP:\n            if url_fragment in base_url:\n                return provider\n\n    return _PROVIDER_NAME_MAP.get(prefix, \"unknown\")\n\n\ndef get_embedding_provider_name(embedding_model: EmbeddingModelBase) -> str:\n    \"\"\"Parse embedding model provider name\"\"\"\n    class_name = embedding_model.__class__.__name__\n\n    if \"DashScope\" in class_name:\n        return AgentScopeGenAiProviderName.DASHSCOPE.value\n    elif \"OpenAI\" in class_name:\n        return GenAIAttributes.GenAiProviderNameValues.OPENAI.value\n    elif \"Gemini\" in class_name:\n        return GenAIAttributes.GenAiProviderNameValues.GCP_GEMINI.value\n    elif \"Ollama\" in class_name:\n        return AgentScopeGenAiProviderName.OLLAMA.value\n    else:\n        model_name = getattr(embedding_model, \"model_name\", \"\")\n        if \"text-embedding\" in model_name:\n            if \"qwen\" in model_name or \"dashscope\" in model_name:\n                return AgentScopeGenAiProviderName.DASHSCOPE.value\n            else:\n                return GenAIAttributes.GenAiProviderNameValues.OPENAI.value\n        return \"agentscope\"\n\n\ndef _get_tool_definitions(\n    tools: list[dict[str, Any]] | None,\n    tool_choice: str | None,\n) -> str | None:\n    \"\"\"Extract and serialize tool definitions for tracing.\"\"\"\n    if tools is None or not isinstance(tools, list) or len(tools) == 0:\n        return None\n\n    if tool_choice == \"none\":\n        return None\n\n    try:\n        flat_tools = []\n        for tool in tools:\n            if not isinstance(tool, dict) or \"function\" not in tool:\n                continue\n\n            func_def = tool[\"function\"]\n            flat_tool = {\n                \"type\": tool.get(\"type\", \"function\"),\n                \"name\": func_def.get(\"name\"),\n                \"description\": func_def.get(\"description\"),\n                \"parameters\": func_def.get(\"parameters\"),\n            }\n            flat_tool = {k: v for k, v in flat_tool.items() if v is not None}\n            flat_tools.append(flat_tool)\n\n        if flat_tools:\n            return _serialize_to_str(flat_tools)\n        return None\n\n    except Exception:\n        return None\n\n\ndef _to_serializable(obj: Any) -> Any:\n    \"\"\"Convert an object to a JSON serializable type.\"\"\"\n    if isinstance(obj, (str, int, bool, float, type(None))):\n        return obj\n    elif isinstance(obj, (list, tuple, set, frozenset)):\n        return [_to_serializable(x) for x in obj]\n    elif isinstance(obj, dict):\n        return {str(key): _to_serializable(val) for (key, val) in obj.items()}\n    elif isinstance(obj, (Msg, BaseModel)) or is_dataclass(obj):\n        return repr(obj)\n    elif inspect.isclass(obj) and issubclass(obj, BaseModel):\n        return repr(obj)\n    elif isinstance(obj, (datetime.date, datetime.datetime, datetime.time)):\n        return obj.isoformat()\n    elif isinstance(obj, datetime.timedelta):\n        return obj.total_seconds()\n    elif isinstance(obj, enum.Enum):\n        return _to_serializable(obj.value)\n    else:\n        return str(obj)\n\n\ndef _serialize_to_str(value: Any) -> str:\n    \"\"\"Serialize value to JSON string\"\"\"\n    try:\n        return json.dumps(value, ensure_ascii=False)\n    except TypeError:\n        return json.dumps(_to_serializable(value), ensure_ascii=False)\n\n\ndef _convert_block_to_part(block: Dict[str, Any]) -> Dict[str, Any] | None:\n    \"\"\"Convert content block to standardized part format.\"\"\"\n    block_type = block.get(\"type\")\n\n    if block_type == \"text\":\n        return {\"type\": \"text\", \"content\": block.get(\"text\", \"\")}\n\n    elif block_type == \"thinking\":\n        return {\n            \"type\": \"text\",\n            \"content\": f\"[Thinking] {block.get('thinking', '')}\",\n        }\n\n    elif block_type == \"tool_use\":\n        return {\n            \"type\": \"tool_call\",\n            \"id\": block.get(\"id\", \"\"),\n            \"name\": block.get(\"name\", \"\"),\n            \"arguments\": block.get(\"input\", {}),\n        }\n\n    elif block_type == \"tool_result\":\n        output = block.get(\"output\", \"\")\n        result = (\n            _serialize_to_str(output)\n            if isinstance(output, (list, dict))\n            else str(output)\n        )\n        return {\n            \"type\": \"tool_call_response\",\n            \"id\": block.get(\"id\", \"\"),\n            \"result\": result,\n        }\n\n    elif block_type in [\"image\", \"audio\", \"video\"]:\n        source = block.get(\"source\", {})\n        source_type = source.get(\"type\")\n        media_type = source.get(\"media_type\")\n\n        if source_type == \"url\":\n            url = source.get(\"url\", \"\")\n            # Infer mime_type from URL extension if not provided\n            if not media_type:\n                media_type, _ = mimetypes.guess_type(url)\n            return {\n                \"type\": \"uri\",\n                \"uri\": url,\n                \"modality\": block_type,\n                \"mime_type\": media_type,\n            }\n        elif source_type == \"base64\":\n            if not media_type:\n                default_media_types = {\n                    \"image\": \"image/jpeg\",\n                    \"audio\": \"audio/wav\",\n                    \"video\": \"video/mp4\",\n                }\n                media_type = default_media_types.get(block_type, \"unknown\")\n            return {\n                \"type\": \"blob\",\n                \"content\": source.get(\"data\", \"\"),\n                \"media_type\": media_type,\n                \"modality\": block_type,\n            }\n\n    return None\n\n\ndef _format_msg_to_parts(msg: Msg) -> dict[str, Any]:\n    \"\"\"Convert Msg to standard format (parts structure)\"\"\"\n    try:\n        parts = []\n        for block in msg.get_content_blocks():\n            part = _convert_block_to_part(block)\n            if part:\n                parts.append(part)\n\n        formatted_msg = {\"role\": msg.role, \"parts\": parts}\n        if msg.name:\n            formatted_msg[\"name\"] = msg.name\n\n        return formatted_msg\n\n    except Exception as e:\n        logger.debug(f\"Error formatting message: {e}\")\n        return {\n            \"role\": msg.role,\n            \"parts\": [\n                {\n                    \"type\": \"text\",\n                    \"content\": str(msg.content) if msg.content else \"\",\n                }\n            ],\n        }\n\n\ndef create_llm_invocation(\n    call_instance: ChatModelBase,\n    call_args: Tuple[Any, ...],\n    call_kwargs: Dict[str, Any],\n) -> LLMInvocation:\n    \"\"\"Create LLM invocation from call context.\"\"\"\n    provider_name = get_provider_name(call_instance)\n    request_model = getattr(call_instance, \"model_name\", \"unknown_model\")\n\n    # Convert input messages\n    input_messages = []\n    messages = call_args[0] if call_args else call_kwargs.get(\"messages\")\n    if messages:\n        try:\n            converted_messages = get_message_converter(provider_name)(messages)\n            input_messages = convert_agentscope_messages_to_genai_format(\n                converted_messages, provider_name=provider_name\n            )\n        except Exception as e:\n            logger.warning(\n                f\"Failed to convert input messages: {e}\", exc_info=True\n            )\n\n    invocation = LLMInvocation(\n        request_model=request_model,\n        provider=provider_name,\n        input_messages=input_messages,\n    )\n\n    # Set optional request parameters if present\n    if call_kwargs.get(\"max_tokens\"):\n        invocation.max_tokens = call_kwargs[\"max_tokens\"]\n    if call_kwargs.get(\"temperature\"):\n        invocation.temperature = call_kwargs[\"temperature\"]\n    if call_kwargs.get(\"top_p\"):\n        invocation.top_p = call_kwargs[\"top_p\"]\n\n    # Set tool definitions if applicable\n    if not call_kwargs.get(\"structured_model\", False):\n        tool_definitions = _get_tool_definitions(\n            call_kwargs.get(\"tools\"), call_kwargs.get(\"tool_choice\")\n        )\n        if tool_definitions:\n            invocation.attributes[\"gen_ai.request.tool_definitions\"] = (\n                tool_definitions\n            )\n\n    return invocation\n\n\ndef create_embedding_invocation(\n    call_instance: EmbeddingModelBase,\n    call_args: Tuple[Any, ...],\n    call_kwargs: Dict[str, Any],\n) -> EmbeddingInvocation:\n    \"\"\"Create Embedding invocation from call context.\"\"\"\n    provider_name = get_embedding_provider_name(call_instance)\n    request_model = getattr(call_instance, \"model_name\", \"unknown_model\")\n\n    invocation = EmbeddingInvocation(\n        request_model=request_model,\n        provider=provider_name,\n    )\n\n    # Set encoding formats if present\n    if call_kwargs.get(\"encoding_formats\"):\n        invocation.encoding_formats = call_kwargs[\"encoding_formats\"]\n\n    return invocation\n\n\ndef create_agent_invocation(\n    reply_instance: AgentBase,\n    reply_args: Tuple[Any, ...],\n    reply_kwargs: Dict[str, Any],\n) -> InvokeAgentInvocation:\n    \"\"\"Create Agent invocation from reply call context.\"\"\"\n    # Get provider and model info\n    provider_name = None\n    request_model = None\n    if hasattr(reply_instance, \"model\") and reply_instance.model:\n        model = reply_instance.model\n        request_model = getattr(model, \"model_name\", None)\n        if isinstance(model, ChatModelBase):\n            provider_name = get_provider_name(model)\n\n    # Convert input messages\n    input_messages = []\n    msg = reply_args[0] if reply_args else reply_kwargs.get(\"msg\")\n    if msg:\n        try:\n            if isinstance(msg, Msg):\n                input_messages = convert_agentscope_messages_to_genai_format(\n                    [msg]\n                )\n            elif isinstance(msg, list):\n                input_messages = convert_agentscope_messages_to_genai_format(\n                    msg\n                )\n        except Exception as e:\n            logger.debug(f\"Error converting agent input messages: {e}\")\n\n    invocation = InvokeAgentInvocation(\n        provider=provider_name,\n        agent_name=getattr(reply_instance, \"name\", \"unknown_agent\"),\n        agent_id=getattr(reply_instance, \"id\", \"unknown\"),\n        agent_description=inspect.getdoc(reply_instance.__class__)\n        or \"No description available\",\n        conversation_id=_config.run_id,\n        request_model=request_model,\n        input_messages=input_messages,\n    )\n\n    # Set system instruction if available\n    if hasattr(reply_instance, \"sys_prompt\") and reply_instance.sys_prompt:\n        sys_prompt = reply_instance.sys_prompt\n        if isinstance(sys_prompt, str):\n            invocation.system_instruction = [Text(content=sys_prompt)]\n        elif isinstance(sys_prompt, list):\n            invocation.system_instruction = sys_prompt\n\n    return invocation\n\n\ndef get_chatmodel_output_messages(chat_response: Any) -> List[Dict[str, Any]]:\n    \"\"\"Convert ChatResponse to OpenTelemetry standard output message format.\"\"\"\n    try:\n        if not isinstance(chat_response, ChatResponse):\n            return []\n\n        parts = []\n        for block in chat_response.content:\n            part = _convert_block_to_part(block)\n            if part:\n                parts.append(part)\n\n        if not parts:\n            parts.append({\"type\": \"text\", \"content\": \"\"})\n\n        return [\n            {\n                \"role\": \"assistant\",\n                \"parts\": parts,\n                \"finish_reason\": \"stop\",\n            }\n        ]\n\n    except Exception as e:\n        logger.warning(\n            f\"Error processing ChatResponse to output messages: {e}\"\n        )\n        return [\n            {\n                \"role\": \"assistant\",\n                \"parts\": [\n                    {\"type\": \"text\", \"content\": \"<error processing response>\"}\n                ],\n                \"finish_reason\": \"error\",\n            }\n        ]\n\n\ndef convert_agentscope_messages_to_genai_format(\n    messages: Any,\n    provider_name: Optional[str] = None,\n) -> List[InputMessage]:\n    \"\"\"Convert AgentScope messages to opentelemetry-util-genai InputMessage format.\n\n    This function is used by ExtendedTelemetryHandler which requires InputMessage objects.\n    \"\"\"\n    if not messages:\n        return []\n\n    if not isinstance(messages, list):\n        messages = [messages]\n\n    input_messages = []\n    for msg in messages:\n        if isinstance(msg, Msg):\n            msg_dict = _format_msg_to_parts(msg)\n        elif isinstance(msg, dict):\n            msg_dict = msg\n        else:\n            continue\n\n        role = msg_dict.get(\"role\", \"user\")\n        parts = msg_dict.get(\"parts\", [])\n\n        converted_parts = []\n        for part in parts:\n            part_type = part.get(\"type\")\n            if part_type == \"text\":\n                converted_parts.append(Text(content=part.get(\"content\", \"\")))\n            elif part_type == \"tool_call\":\n                converted_parts.append(\n                    ToolCall(\n                        id=part.get(\"id\"),\n                        name=part.get(\"name\", \"\"),\n                        arguments=part.get(\"arguments\", {}),\n                    )\n                )\n            elif part_type == \"tool_call_response\":\n                converted_parts.append(\n                    ToolCallResponse(\n                        id=part.get(\"id\"),\n                        response=part.get(\"result\", \"\"),\n                    )\n                )\n            elif part_type == \"reasoning\":\n                converted_parts.append(\n                    Reasoning(content=part.get(\"content\", \"\"))\n                )\n            elif part_type in (\"uri\", \"blob\"):\n                converted_parts.append(part)\n            else:\n                # Keep other types as-is\n                converted_parts.append(part)\n\n        input_messages.append(InputMessage(role=role, parts=converted_parts))\n\n    return input_messages\n\n\ndef convert_chatresponse_to_output_messages(\n    chat_response: Any,\n) -> List[OutputMessage]:\n    \"\"\"Convert ChatResponse to opentelemetry-util-genai OutputMessage format.\n\n    This function is used by ExtendedTelemetryHandler which requires OutputMessage objects.\n    \"\"\"\n    output_dicts = get_chatmodel_output_messages(chat_response)\n    if not output_dicts:\n        return []\n\n    output_messages = []\n    for msg_dict in output_dicts:\n        role = msg_dict.get(\"role\", \"assistant\")\n        parts_dicts = msg_dict.get(\"parts\", [])\n        finish_reason = msg_dict.get(\"finish_reason\", \"stop\")\n\n        converted_parts = []\n        for part in parts_dicts:\n            part_type = part.get(\"type\")\n            if part_type == \"text\":\n                converted_parts.append(Text(content=part.get(\"content\", \"\")))\n            elif part_type == \"tool_call\":\n                converted_parts.append(\n                    ToolCall(\n                        id=part.get(\"id\"),\n                        name=part.get(\"name\", \"\"),\n                        arguments=part.get(\"arguments\", {}),\n                    )\n                )\n            else:\n                converted_parts.append(part)\n\n        output_messages.append(\n            OutputMessage(\n                role=role,\n                parts=converted_parts,\n                finish_reason=finish_reason,\n            )\n        )\n\n    return output_messages\n\n\ndef convert_agent_response_to_output_messages(\n    agent_response: Any,\n) -> List[OutputMessage]:\n    \"\"\"Convert agent Msg response to opentelemetry-util-genai OutputMessage format.\n\n    This function is used by ExtendedTelemetryHandler which requires OutputMessage objects.\n    \"\"\"\n    if not hasattr(agent_response, \"content\"):\n        return []\n\n    try:\n        # Use _format_msg_to_parts to convert Msg to dict, then convert to OutputMessage\n        msg_dict = _format_msg_to_parts(agent_response)\n        parts_dicts = msg_dict.get(\"parts\", [])\n\n        converted_parts = []\n        for part in parts_dicts:\n            part_type = part.get(\"type\")\n            if part_type == \"text\":\n                converted_parts.append(Text(content=part.get(\"content\", \"\")))\n            elif part_type == \"reasoning\":\n                converted_parts.append(\n                    Reasoning(content=part.get(\"content\", \"\"))\n                )\n            elif part_type == \"tool_call\":\n                converted_parts.append(\n                    ToolCall(\n                        id=part.get(\"id\"),\n                        name=part.get(\"name\", \"\"),\n                        arguments=part.get(\"arguments\", {}),\n                    )\n                )\n            elif part_type in (\"uri\", \"blob\"):\n                converted_parts.append(part)\n            else:\n                converted_parts.append(Text(content=str(part)))\n\n        if not converted_parts:\n            converted_parts.append(Text(content=\"\"))\n\n        return [\n            OutputMessage(\n                role=\"assistant\",\n                parts=converted_parts,\n                finish_reason=\"stop\",\n            )\n        ]\n    except Exception as e:\n        logger.debug(f\"Failed to convert agent response: {e}\")\n        return []\n"}, "232fc7a10beafe96c05b34664b803760754bb07a": {"t": 1785758588.038534, "n": 19037, "h": "2bd433a95ed8b283", "c": "# -*- coding: utf-8 -*-\n\"\"\"Attributes processor for span attributes.\"\"\"\n\nfrom __future__ import annotations\n\nimport datetime\nimport enum\nimport inspect\nimport json\nimport logging\nfrom dataclasses import is_dataclass\nfrom enum import Enum\nfrom typing import Any, Dict, List, Optional, Tuple\n\nfrom agentscope import _config\nfrom agentscope.agent import AgentBase\nfrom agentscope.embedding import EmbeddingModelBase\nfrom agentscope.message import Msg\nfrom agentscope.model import ChatModelBase, ChatResponse\nfrom pydantic import BaseModel\n\nfrom opentelemetry.semconv._incubating.attributes import (\n    gen_ai_attributes as GenAIAttributes,\n)\nfrom opentelemetry.util.genai.extended_types import (\n    EmbeddingInvocation,\n    InvokeAgentInvocation,\n)\nfrom opentelemetry.util.genai.types import (\n    InputMessage,\n    LLMInvocation,\n    OutputMessage,\n    Reasoning,\n    Text,\n    ToolCall,\n    ToolCallResponse,\n)\n\nfrom .message_converter import get_message_converter\n\nlogger = logging.getLogger(__name__)\n\n\nclass AgentScopeGenAiProviderName(str, Enum):\n    \"\"\"Extended provider names not in standard OpenTelemetry semantic conventions.\"\"\"\n\n    OLLAMA = \"ollama\"\n    DASHSCOPE = \"dashscope\"\n    MOONSHOT = \"moonshot\"\n\n\n# Provider name mapping based on class names\n_PROVIDER_NAME_MAP = {\n    \"openai\": GenAIAttributes.GenAiProviderNameValues.OPENAI.value,\n    \"gemini\": GenAIAttributes.GenAiProviderNameValues.GCP_GEMINI.value,\n    \"anthropic\": GenAIAttributes.GenAiProviderNameValues.ANTHROPIC.value,\n    \"dashscope\": AgentScopeGenAiProviderName.DASHSCOPE.value,\n    \"ollama\": AgentScopeGenAiProviderName.OLLAMA.value,\n}\n\n# Base URL to provider mapping for OpenAI-compatible APIs\n_BASE_URL_PROVIDER_MAP = [\n    (\"openai.com\", GenAIAttributes.GenAiProviderNameValues.OPENAI.value),\n    (\n        \"api.deepseek.com\",\n        GenAIAttributes.GenAiProviderNameValues.DEEPSEEK.value,\n    ),\n    (\"dashscope.aliyuncs.com\", AgentScopeGenAiProviderName.DASHSCOPE.value),\n]\n\n\ndef get_provider_name(chat_model: ChatModelBase) -> str:\n    \"\"\"Parse chat model provider name\"\"\"\n    classname = chat_model.__class__.__name__\n    prefix = classname.removesuffix(\"ChatModel\").lower()\n\n    # Special handling for DashScopeChatModel with base_http_api_url\n    if (\n        prefix == \"dashscope\"\n        and hasattr(chat_model, \"base_http_api_url\")\n        and chat_model.base_http_api_url\n    ):\n        base_url = chat_model.base_http_api_url\n        for url_fragment, provider in _BASE_URL_PROVIDER_MAP:\n            if url_fragment in base_url:\n                return provider\n\n    return _PROVIDER_NAME_MAP.get(prefix, \"unknown\")\n\n\ndef get_embedding_provider_name(embedding_model: EmbeddingModelBase) -> str:\n    \"\"\"Parse embedding model provider name\"\"\"\n    class_name = embedding_model.__class__.__name__\n\n    if \"DashScope\" in class_name:\n        return AgentScopeGenAiProviderName.DASHSCOPE.value\n    elif \"OpenAI\" in class_name:\n        return GenAIAttributes.GenAiProviderNameValues.OPENAI.value\n    elif \"Gemini\" in class_name:\n        return GenAIAttributes.GenAiProviderNameValues.GCP_GEMINI.value\n    elif \"Ollama\" in class_name:\n        return AgentScopeGenAiProviderName.OLLAMA.value\n    else:\n        model_name = getattr(embedding_model, \"model_name\", \"\")\n        if \"text-embedding\" in model_name:\n            if \"qwen\" in model_name or \"dashscope\" in model_name:\n                return AgentScopeGenAiProviderName.DASHSCOPE.value\n            else:\n                return GenAIAttributes.GenAiProviderNameValues.OPENAI.value\n        return \"agentscope\"\n\n\ndef _get_tool_definitions(\n    tools: list[dict[str, Any]] | None,\n    tool_choice: str | None,\n) -> str | None:\n    \"\"\"Extract and serialize tool definitions for tracing.\"\"\"\n    if tools is None or not isinstance(tools, list) or len(tools) == 0:\n        return None\n\n    if tool_choice == \"none\":\n        return None\n\n    try:\n        flat_tools = []\n        for tool in tools:\n            if not isinstance(tool, dict) or \"function\" not in tool:\n                continue\n\n            func_def = tool[\"function\"]\n            flat_tool = {\n                \"type\": tool.get(\"type\", \"function\"),\n                \"name\": func_def.get(\"name\"),\n                \"description\": func_def.get(\"description\"),\n                \"parameters\": func_def.get(\"parameters\"),\n            }\n            flat_tool = {k: v for k, v in flat_tool.items() if v is not None}\n            flat_tools.append(flat_tool)\n\n        if flat_tools:\n            return _serialize_to_str(flat_tools)\n        return None\n\n    except Exception:\n        return None\n\n\ndef _to_serializable(obj: Any) -> Any:\n    \"\"\"Convert an object to a JSON serializable type.\"\"\"\n    if isinstance(obj, (str, int, bool, float, type(None))):\n        return obj\n    elif isinstance(obj, (list, tuple, set, frozenset)):\n        return [_to_serializable(x) for x in obj]\n    elif isinstance(obj, dict):\n        return {str(key): _to_serializable(val) for (key, val) in obj.items()}\n    elif isinstance(obj, (Msg, BaseModel)) or is_dataclass(obj):\n        return repr(obj)\n    elif inspect.isclass(obj) and issubclass(obj, BaseModel):\n        return repr(obj)\n    elif isinstance(obj, (datetime.date, datetime.datetime, datetime.time)):\n        return obj.isoformat()\n    elif isinstance(obj, datetime.timedelta):\n        return obj.total_seconds()\n    elif isinstance(obj, enum.Enum):\n        return _to_serializable(obj.value)\n    else:\n        return str(obj)\n\n\ndef _serialize_to_str(value: Any) -> str:\n    \"\"\"Serialize value to JSON string\"\"\"\n    try:\n        return json.dumps(value, ensure_ascii=False)\n    except TypeError:\n        return json.dumps(_to_serializable(value), ensure_ascii=False)\n\n\ndef _convert_block_to_part(block: Dict[str, Any]) -> Dict[str, Any] | None:\n    \"\"\"Convert content block to standardized part format.\"\"\"\n    block_type = block.get(\"type\")\n\n    if block_type == \"text\":\n        return {\"type\": \"text\", \"content\": block.get(\"text\", \"\")}\n\n    elif block_type == \"thinking\":\n        return {\n            \"type\": \"text\",\n            \"content\": f\"[Thinking] {block.get('thinking', '')}\",\n        }\n\n    elif block_type == \"tool_use\":\n        return {\n            \"type\": \"tool_call\",\n            \"id\": block.get(\"id\", \"\"),\n            \"name\": block.get(\"name\", \"\"),\n            \"arguments\": block.get(\"input\", {}),\n        }\n\n    elif block_type == \"tool_result\":\n        output = block.get(\"output\", \"\")\n        result = (\n            _serialize_to_str(output)\n            if isinstance(output, (list, dict))\n            else str(output)\n        )\n        return {\n            \"type\": \"tool_call_response\",\n            \"id\": block.get(\"id\", \"\"),\n            \"result\": result,\n        }\n\n    elif block_type in [\"image\", \"audio\", \"video\"]:\n        source = block.get(\"source\", {})\n        source_type = source.get(\"type\")\n        media_type = source.get(\"media_type\")\n\n        if source_type == \"url\":\n            return {\n                \"type\": \"uri\",\n                \"uri\": source.get(\"url\", \"\"),\n                \"modality\": block_type,\n                \"mime_type\": media_type,\n            }\n        elif source_type == \"base64\":\n            if not media_type:\n                default_media_types = {\n                    \"image\": \"image/jpeg\",\n                    \"audio\": \"audio/wav\",\n                    \"video\": \"video/mp4\",\n                }\n                media_type = default_media_types.get(block_type, \"unknown\")\n            return {\n                \"type\": \"blob\",\n                \"content\": source.get(\"data\", \"\"),\n                \"media_type\": media_type,\n                \"modality\": block_type,\n            }\n\n    return None\n\n\ndef _format_msg_to_parts(msg: Msg) -> dict[str, Any]:\n    \"\"\"Convert Msg to standard format (parts structure)\"\"\"\n    try:\n        parts = []\n        for block in msg.get_content_blocks():\n            part = _convert_block_to_part(block)\n            if part:\n                parts.append(part)\n\n        formatted_msg = {\"role\": msg.role, \"parts\": parts}\n        if msg.name:\n            formatted_msg[\"name\"] = msg.name\n\n        return formatted_msg\n\n    except Exception as e:\n        logger.debug(f\"Error formatting message: {e}\")\n        return {\n            \"role\": msg.role,\n            \"parts\": [\n                {\n                    \"type\": \"text\",\n                    \"content\": str(msg.content) if msg.content else \"\",\n                }\n            ],\n        }\n\n\ndef create_llm_invocation(\n    call_instance: ChatModelBase,\n    call_args: Tuple[Any, ...],\n    call_kwargs: Dict[str, Any],\n) -> LLMInvocation:\n    \"\"\"Create LLM invocation from call context.\"\"\"\n    provider_name = get_provider_name(call_instance)\n    request_model = getattr(call_instance, \"model_name\", \"unknown_model\")\n\n    # Convert input messages\n    input_messages = []\n    messages = call_args[0] if call_args else call_kwargs.get(\"messages\")\n    if messages:\n        try:\n            converted_messages = get_message_converter(provider_name)(messages)\n            input_messages = convert_agentscope_messages_to_genai_format(\n                converted_messages, provider_name=provider_name\n            )\n        except Exception as e:\n            logger.warning(\n                f\"Failed to convert input messages: {e}\", exc_info=True\n            )\n\n    invocation = LLMInvocation(\n        request_model=request_model,\n        provider=provider_name,\n        input_messages=input_messages,\n    )\n\n    # Set optional request parameters if present\n    if call_kwargs.get(\"max_tokens\"):\n        invocation.max_tokens = call_kwargs[\"max_tokens\"]\n    if call_kwargs.get(\"temperature\"):\n        invocation.temperature = call_kwargs[\"temperature\"]\n    if call_kwargs.get(\"top_p\"):\n        invocation.top_p = call_kwargs[\"top_p\"]\n\n    # Set tool definitions if applicable\n    if not call_kwargs.get(\"structured_model\", False):\n        tool_definitions = _get_tool_definitions(\n            call_kwargs.get(\"tools\"), call_kwargs.get(\"tool_choice\")\n        )\n        if tool_definitions:\n            invocation.attributes[\"gen_ai.request.tool_definitions\"] = (\n                tool_definitions\n            )\n\n    return invocation\n\n\ndef create_embedding_invocation(\n    call_instance: EmbeddingModelBase,\n    call_args: Tuple[Any, ...],\n    call_kwargs: Dict[str, Any],\n) -> EmbeddingInvocation:\n    \"\"\"Create Embedding invocation from call context.\"\"\"\n    provider_name = get_embedding_provider_name(call_instance)\n    request_model = getattr(call_instance, \"model_name\", \"unknown_model\")\n\n    invocation = EmbeddingInvocation(\n        request_model=request_model,\n        provider=provider_name,\n    )\n\n    # Set encoding formats if present\n    if call_kwargs.get(\"encoding_formats\"):\n        invocation.encoding_formats = call_kwargs[\"encoding_formats\"]\n\n    return invocation\n\n\ndef create_agent_invocation(\n    reply_instance: AgentBase,\n    reply_args: Tuple[Any, ...],\n    reply_kwargs: Dict[str, Any],\n) -> InvokeAgentInvocation:\n    \"\"\"Create Agent invocation from reply call context.\"\"\"\n    # Get provider and model info\n    provider_name = None\n    request_model = None\n    if hasattr(reply_instance, \"model\") and reply_instance.model:\n        model = reply_instance.model\n        request_model = getattr(model, \"model_name\", None)\n        if isinstance(model, ChatModelBase):\n            provider_name = get_provider_name(model)\n\n    # Convert input messages\n    input_messages = []\n    msg = reply_args[0] if reply_args else reply_kwargs.get(\"msg\")\n    if msg:\n        try:\n            if isinstance(msg, Msg):\n                input_messages = convert_agentscope_messages_to_genai_format(\n                    [msg]\n                )\n            elif isinstance(msg, list):\n                input_messages = convert_agentscope_messages_to_genai_format(\n                    msg\n                )\n        except Exception as e:\n            logger.debug(f\"Error converting agent input messages: {e}\")\n\n    invocation = InvokeAgentInvocation(\n        provider=provider_name,\n        agent_name=getattr(reply_instance, \"name\", \"unknown_agent\"),\n        agent_id=getattr(reply_instance, \"id\", \"unknown\"),\n        agent_description=inspect.getdoc(reply_instance.__class__)\n        or \"No description available\",\n        conversation_id=_config.run_id,\n        request_model=request_model,\n        input_messages=input_messages,\n    )\n\n    # Set system instruction if available\n    if hasattr(reply_instance, \"sys_prompt\") and reply_instance.sys_prompt:\n        sys_prompt = reply_instance.sys_prompt\n        if isinstance(sys_prompt, str):\n            invocation.system_instruction = [Text(content=sys_prompt)]\n        elif isinstance(sys_prompt, list):\n            invocation.system_instruction = sys_prompt\n\n    return invocation\n\n\ndef get_chatmodel_output_messages(chat_response: Any) -> List[Dict[str, Any]]:\n    \"\"\"Convert ChatResponse to OpenTelemetry standard output message format.\"\"\"\n    try:\n        if not isinstance(chat_response, ChatResponse):\n            return []\n\n        parts = []\n        for block in chat_response.content:\n            part = _convert_block_to_part(block)\n            if part:\n                parts.append(part)\n\n        if not parts:\n            parts.append({\"type\": \"text\", \"content\": \"\"})\n\n        return [\n            {\n                \"role\": \"assistant\",\n                \"parts\": parts,\n                \"finish_reason\": \"stop\",\n            }\n        ]\n\n    except Exception as e:\n        logger.warning(\n            f\"Error processing ChatResponse to output messages: {e}\"\n        )\n        return [\n            {\n                \"role\": \"assistant\",\n                \"parts\": [\n                    {\"type\": \"text\", \"content\": \"<error processing response>\"}\n                ],\n                \"finish_reason\": \"error\",\n            }\n        ]\n\n\ndef convert_agentscope_messages_to_genai_format(\n    messages: Any,\n    provider_name: Optional[str] = None,\n) -> List[InputMessage]:\n    \"\"\"Convert AgentScope messages to opentelemetry-util-genai InputMessage format.\n\n    This function is used by ExtendedTelemetryHandler which requires InputMessage objects.\n    \"\"\"\n    if not messages:\n        return []\n\n    if not isinstance(messages, list):\n        messages = [messages]\n\n    input_messages = []\n    for msg in messages:\n        if isinstance(msg, Msg):\n            msg_dict = _format_msg_to_parts(msg)\n        elif isinstance(msg, dict):\n            msg_dict = msg\n        else:\n            continue\n\n        role = msg_dict.get(\"role\", \"user\")\n        parts = msg_dict.get(\"parts\", [])\n\n        converted_parts = []\n        for part in parts:\n            part_type = part.get(\"type\")\n            if part_type == \"text\":\n                converted_parts.append(Text(content=part.get(\"content\", \"\")))\n            elif part_type == \"tool_call\":\n                converted_parts.append(\n                    ToolCall(\n                        id=part.get(\"id\"),\n                        name=part.get(\"name\", \"\"),\n                        arguments=part.get(\"arguments\", {}),\n                    )\n                )\n            elif part_type == \"tool_call_response\":\n                converted_parts.append(\n                    ToolCallResponse(\n                        id=part.get(\"id\"),\n                        response=part.get(\"result\", \"\"),\n                    )\n                )\n            elif part_type == \"reasoning\":\n                converted_parts.append(\n                    Reasoning(content=part.get(\"content\", \"\"))\n                )\n            elif part_type in (\"uri\", \"blob\"):\n                converted_parts.append(part)\n            else:\n                # Keep other types as-is\n                converted_parts.append(part)\n\n        input_messages.append(InputMessage(role=role, parts=converted_parts))\n\n    return input_messages\n\n\ndef convert_chatresponse_to_output_messages(\n    chat_response: Any,\n) -> List[OutputMessage]:\n    \"\"\"Convert ChatResponse to opentelemetry-util-genai OutputMessage format.\n\n    This function is used by ExtendedTelemetryHandler which requires OutputMessage objects.\n    \"\"\"\n    output_dicts = get_chatmodel_output_messages(chat_response)\n    if not output_dicts:\n        return []\n\n    output_messages = []\n    for msg_dict in output_dicts:\n        role = msg_dict.get(\"role\", \"assistant\")\n        parts_dicts = msg_dict.get(\"parts\", [])\n        finish_reason = msg_dict.get(\"finish_reason\", \"stop\")\n\n        converted_parts = []\n        for part in parts_dicts:\n            part_type = part.get(\"type\")\n            if part_type == \"text\":\n                converted_parts.append(Text(content=part.get(\"content\", \"\")))\n            elif part_type == \"tool_call\":\n                converted_parts.append(\n                    ToolCall(\n                        id=part.get(\"id\"),\n                        name=part.get(\"name\", \"\"),\n                        arguments=part.get(\"arguments\", {}),\n                    )\n                )\n            else:\n                converted_parts.append(part)\n\n        output_messages.append(\n            OutputMessage(\n                role=role,\n                parts=converted_parts,\n                finish_reason=finish_reason,\n            )\n        )\n\n    return output_messages\n\n\ndef convert_agent_response_to_output_messages(\n    agent_response: Any,\n) -> List[OutputMessage]:\n    \"\"\"Convert agent Msg response to opentelemetry-util-genai OutputMessage format.\n\n    This function is used by ExtendedTelemetryHandler which requires OutputMessage objects.\n    \"\"\"\n    if not hasattr(agent_response, \"content\"):\n        return []\n\n    try:\n        # Use _format_msg_to_parts to convert Msg to dict, then convert to OutputMessage\n        msg_dict = _format_msg_to_parts(agent_response)\n        parts_dicts = msg_dict.get(\"parts\", [])\n\n        converted_parts = []\n        for part in parts_dicts:\n            part_type = part.get(\"type\")\n            if part_type == \"text\":\n                converted_parts.append(Text(content=part.get(\"content\", \"\")))\n            elif part_type == \"reasoning\":\n                converted_parts.append(\n                    Reasoning(content=part.get(\"content\", \"\"))\n                )\n            elif part_type == \"tool_call\":\n                converted_parts.append(\n                    ToolCall(\n                        id=part.get(\"id\"),\n                        name=part.get(\"name\", \"\"),\n                        arguments=part.get(\"arguments\", {}),\n                    )\n                )\n            elif part_type in (\"uri\", \"blob\"):\n                converted_parts.append(part)\n            else:\n                converted_parts.append(Text(content=str(part)))\n\n        if not converted_parts:\n            converted_parts.append(Text(content=\"\"))\n\n        return [\n            OutputMessage(\n                role=\"assistant\",\n                parts=converted_parts,\n                finish_reason=\"stop\",\n            )\n        ]\n    except Exception as e:\n        logger.debug(f\"Failed to convert agent response: {e}\")\n        return []\n"}, "68afc3c3efa238660673830d9a0f4e491fc7ab78": {"t": 1785758589.455682, "n": 18220, "h": "61e3fa037d847d9a", "c": "# -*- coding: utf-8 -*-\n\"\"\"Attributes processor for span attributes.\"\"\"\n\nfrom __future__ import annotations\n\nimport datetime\nimport enum\nimport inspect\nimport json\nimport logging\nfrom dataclasses import is_dataclass\nfrom enum import Enum\nfrom typing import Any, Dict, List, Optional, Tuple, Union\n\nfrom agentscope import _config\nfrom agentscope.agent import AgentBase\nfrom agentscope.embedding import EmbeddingModelBase\nfrom agentscope.message import Msg\nfrom agentscope.model import ChatModelBase, ChatResponse\nfrom pydantic import BaseModel\n\nfrom opentelemetry.semconv._incubating.attributes import (\n    gen_ai_attributes as GenAIAttributes,\n)\nfrom opentelemetry.util.genai.extended_types import (\n    EmbeddingInvocation,\n    InvokeAgentInvocation,\n)\nfrom opentelemetry.util.genai.types import (\n    InputMessage,\n    LLMInvocation,\n    OutputMessage,\n    Text,\n    ToolCall,\n    ToolCallResponse,\n)\n\nfrom .message_converter import get_message_converter\n\nlogger = logging.getLogger(__name__)\n\n\nclass AgentScopeGenAiProviderName(str, Enum):\n    \"\"\"Extended provider names not in standard OpenTelemetry semantic conventions.\"\"\"\n\n    OLLAMA = \"ollama\"\n    DASHSCOPE = \"dashscope\"\n    MOONSHOT = \"moonshot\"\n\n\n# Provider name mapping based on class names\n_PROVIDER_NAME_MAP = {\n    \"openai\": GenAIAttributes.GenAiProviderNameValues.OPENAI.value,\n    \"gemini\": GenAIAttributes.GenAiProviderNameValues.GCP_GEMINI.value,\n    \"anthropic\": GenAIAttributes.GenAiProviderNameValues.ANTHROPIC.value,\n    \"dashscope\": AgentScopeGenAiProviderName.DASHSCOPE.value,\n    \"ollama\": AgentScopeGenAiProviderName.OLLAMA.value,\n}\n\n# Base URL to provider mapping for OpenAI-compatible APIs\n_BASE_URL_PROVIDER_MAP = [\n    (\"openai.com\", GenAIAttributes.GenAiProviderNameValues.OPENAI.value),\n    (\n        \"api.deepseek.com\",\n        GenAIAttributes.GenAiProviderNameValues.DEEPSEEK.value,\n    ),\n    (\"dashscope.aliyuncs.com\", AgentScopeGenAiProviderName.DASHSCOPE.value),\n]\n\n\ndef get_provider_name(chat_model: ChatModelBase) -> str:\n    \"\"\"Parse chat model provider name\"\"\"\n    classname = chat_model.__class__.__name__\n    prefix = classname.removesuffix(\"ChatModel\").lower()\n\n    # Special handling for DashScopeChatModel with base_http_api_url\n    if (\n        prefix == \"dashscope\"\n        and hasattr(chat_model, \"base_http_api_url\")\n        and chat_model.base_http_api_url\n    ):\n        base_url = chat_model.base_http_api_url\n        for url_fragment, provider in _BASE_URL_PROVIDER_MAP:\n            if url_fragment in base_url:\n                return provider\n\n    return _PROVIDER_NAME_MAP.get(prefix, \"unknown\")\n\n\ndef get_embedding_provider_name(embedding_model: EmbeddingModelBase) -> str:\n    \"\"\"Parse embedding model provider name\"\"\"\n    class_name = embedding_model.__class__.__name__\n\n    if \"DashScope\" in class_name:\n        return AgentScopeGenAiProviderName.DASHSCOPE.value\n    elif \"OpenAI\" in class_name:\n        return GenAIAttributes.GenAiProviderNameValues.OPENAI.value\n    elif \"Gemini\" in class_name:\n        return GenAIAttributes.GenAiProviderNameValues.GCP_GEMINI.value\n    elif \"Ollama\" in class_name:\n        return AgentScopeGenAiProviderName.OLLAMA.value\n    else:\n        model_name = getattr(embedding_model, \"model_name\", \"\")\n        if \"text-embedding\" in model_name:\n            if \"qwen\" in model_name or \"dashscope\" in model_name:\n                return AgentScopeGenAiProviderName.DASHSCOPE.value\n            else:\n                return GenAIAttributes.GenAiProviderNameValues.OPENAI.value\n        return \"agentscope\"\n\n\ndef _get_tool_definitions(\n    tools: list[dict[str, Any]] | None,\n    tool_choice: str | None,\n) -> str | None:\n    \"\"\"Extract and serialize tool definitions for tracing.\"\"\"\n    if tools is None or not isinstance(tools, list) or len(tools) == 0:\n        return None\n\n    if tool_choice == \"none\":\n        return None\n\n    try:\n        flat_tools = []\n        for tool in tools:\n            if not isinstance(tool, dict) or \"function\" not in tool:\n                continue\n\n            func_def = tool[\"function\"]\n            flat_tool = {\n                \"type\": tool.get(\"type\", \"function\"),\n                \"name\": func_def.get(\"name\"),\n                \"description\": func_def.get(\"description\"),\n                \"parameters\": func_def.get(\"parameters\"),\n            }\n            flat_tool = {k: v for k, v in flat_tool.items() if v is not None}\n            flat_tools.append(flat_tool)\n\n        if flat_tools:\n            return _serialize_to_str(flat_tools)\n        return None\n\n    except Exception:\n        return None\n\n\ndef _to_serializable(obj: Any) -> Any:\n    \"\"\"Convert an object to a JSON serializable type.\"\"\"\n    if isinstance(obj, (str, int, bool, float, type(None))):\n        return obj\n    elif isinstance(obj, (list, tuple, set, frozenset)):\n        return [_to_serializable(x) for x in obj]\n    elif isinstance(obj, dict):\n        return {str(key): _to_serializable(val) for (key, val) in obj.items()}\n    elif isinstance(obj, (Msg, BaseModel)) or is_dataclass(obj):\n        return repr(obj)\n    elif inspect.isclass(obj) and issubclass(obj, BaseModel):\n        return repr(obj)\n    elif isinstance(obj, (datetime.date, datetime.datetime, datetime.time)):\n        return obj.isoformat()\n    elif isinstance(obj, datetime.timedelta):\n        return obj.total_seconds()\n    elif isinstance(obj, enum.Enum):\n        return _to_serializable(obj.value)\n    else:\n        return str(obj)\n\n\ndef _serialize_to_str(value: Any) -> str:\n    \"\"\"Serialize value to JSON string\"\"\"\n    try:\n        return json.dumps(value, ensure_ascii=False)\n    except TypeError:\n        return json.dumps(_to_serializable(value), ensure_ascii=False)\n\n\ndef _convert_block_to_part(block: Dict[str, Any]) -> Dict[str, Any] | None:\n    \"\"\"Convert content block to standardized part format.\"\"\"\n    block_type = block.get(\"type\")\n\n    if block_type == \"text\":\n        return {\"type\": \"text\", \"content\": block.get(\"text\", \"\")}\n\n    elif block_type == \"thinking\":\n        return {\n            \"type\": \"text\",\n            \"content\": f\"[Thinking] {block.get('thinking', '')}\",\n        }\n\n    elif block_type == \"tool_use\":\n        return {\n            \"type\": \"tool_call\",\n            \"id\": block.get(\"id\", \"\"),\n            \"name\": block.get(\"name\", \"\"),\n            \"arguments\": block.get(\"input\", {}),\n        }\n\n    elif block_type == \"tool_result\":\n        output = block.get(\"output\", \"\")\n        result = (\n            _serialize_to_str(output)\n            if isinstance(output, (list, dict))\n            else str(output)\n        )\n        return {\n            \"type\": \"tool_call_response\",\n            \"id\": block.get(\"id\", \"\"),\n            \"result\": result,\n        }\n\n    elif block_type == \"image\":\n        source = block.get(\"source\", {})\n        source_type = source.get(\"type\")\n        if source_type == \"url\":\n            return {\"type\": \"image\", \"url\": source.get(\"url\", \"\")}\n        elif source_type == \"base64\":\n            data = source.get(\"data\", \"\")\n            media_type = source.get(\"media_type\", \"image/jpeg\")\n            return {\"type\": \"image\", \"url\": f\"data:{media_type};base64,{data}\"}\n\n    elif block_type == \"audio\":\n        return {\"type\": \"audio\", \"source\": block.get(\"source\", {})}\n\n    elif block_type == \"video\":\n        return {\"type\": \"video\", \"source\": block.get(\"source\", {})}\n\n    return None\n\n\ndef _format_msg_to_parts(msg: Msg) -> dict[str, Any]:\n    \"\"\"Convert Msg to standard format (parts structure)\"\"\"\n    try:\n        parts = []\n        for block in msg.get_content_blocks():\n            part = _convert_block_to_part(block)\n            if part:\n                parts.append(part)\n\n        formatted_msg = {\"role\": msg.role, \"parts\": parts}\n        if msg.name:\n            formatted_msg[\"name\"] = msg.name\n\n        return formatted_msg\n\n    except Exception as e:\n        logger.debug(f\"Error formatting message: {e}\")\n        return {\n            \"role\": msg.role,\n            \"parts\": [\n                {\n                    \"type\": \"text\",\n                    \"content\": str(msg.content) if msg.content else \"\",\n                }\n            ],\n        }\n\n\ndef create_llm_invocation(\n    call_instance: ChatModelBase,\n    call_args: Tuple[Any, ...],\n    call_kwargs: Dict[str, Any],\n) -> LLMInvocation:\n    \"\"\"Create LLM invocation from call context.\"\"\"\n    provider_name = get_provider_name(call_instance)\n    request_model = getattr(call_instance, \"model_name\", \"unknown_model\")\n\n    # Convert input messages\n    input_messages = []\n    messages = call_args[0] if call_args else call_kwargs.get(\"messages\")\n    if messages:\n        try:\n            converted_messages = get_message_converter(provider_name)(messages)\n            input_messages = convert_agentscope_messages_to_genai_format(\n                converted_messages, provider_name=provider_name\n            )\n        except Exception as e:\n            logger.warning(\n                f\"Failed to convert input messages: {e}\", exc_info=True\n            )\n\n    invocation = LLMInvocation(\n        request_model=request_model,\n        provider=provider_name,\n        input_messages=input_messages,\n    )\n\n    # Set optional request parameters if present\n    if call_kwargs.get(\"max_tokens\"):\n        invocation.max_tokens = call_kwargs[\"max_tokens\"]\n    if call_kwargs.get(\"temperature\"):\n        invocation.temperature = call_kwargs[\"temperature\"]\n    if call_kwargs.get(\"top_p\"):\n        invocation.top_p = call_kwargs[\"top_p\"]\n\n    # Set tool definitions if applicable\n    if not call_kwargs.get(\"structured_model\", False):\n        tool_definitions = _get_tool_definitions(\n            call_kwargs.get(\"tools\"), call_kwargs.get(\"tool_choice\")\n        )\n        if tool_definitions:\n            invocation.attributes[\"gen_ai.request.tool_definitions\"] = (\n                tool_definitions\n            )\n\n    return invocation\n\n\ndef create_embedding_invocation(\n    call_instance: EmbeddingModelBase,\n    call_args: Tuple[Any, ...],\n    call_kwargs: Dict[str, Any],\n) -> EmbeddingInvocation:\n    \"\"\"Create Embedding invocation from call context.\"\"\"\n    provider_name = get_embedding_provider_name(call_instance)\n    request_model = getattr(call_instance, \"model_name\", \"unknown_model\")\n\n    invocation = EmbeddingInvocation(\n        request_model=request_model,\n        provider=provider_name,\n    )\n\n    # Set encoding formats if present\n    if call_kwargs.get(\"encoding_formats\"):\n        invocation.encoding_formats = call_kwargs[\"encoding_formats\"]\n\n    return invocation\n\n\ndef create_agent_invocation(\n    reply_instance: AgentBase,\n    reply_args: Tuple[Any, ...],\n    reply_kwargs: Dict[str, Any],\n) -> InvokeAgentInvocation:\n    \"\"\"Create Agent invocation from reply call context.\"\"\"\n    # Get provider and model info\n    provider_name = None\n    request_model = None\n    if hasattr(reply_instance, \"model\") and reply_instance.model:\n        model = reply_instance.model\n        request_model = getattr(model, \"model_name\", None)\n        if isinstance(model, ChatModelBase):\n            provider_name = get_provider_name(model)\n\n    # Convert input messages\n    input_messages = []\n    msg = reply_args[0] if reply_args else reply_kwargs.get(\"msg\")\n    if msg:\n        try:\n            if isinstance(msg, Msg):\n                input_messages = convert_agentscope_messages_to_genai_format(\n                    [msg]\n                )\n            elif isinstance(msg, list):\n                input_messages = convert_agentscope_messages_to_genai_format(\n                    msg\n                )\n        except Exception as e:\n            logger.debug(f\"Error converting agent input messages: {e}\")\n\n    invocation = InvokeAgentInvocation(\n        provider=provider_name,\n        agent_name=getattr(reply_instance, \"name\", \"unknown_agent\"),\n        agent_id=getattr(reply_instance, \"id\", \"unknown\"),\n        agent_description=inspect.getdoc(reply_instance.__class__)\n        or \"No description available\",\n        conversation_id=_config.run_id,\n        request_model=request_model,\n        input_messages=input_messages,\n    )\n\n    # Set system instruction if available\n    if hasattr(reply_instance, \"sys_prompt\") and reply_instance.sys_prompt:\n        sys_prompt = reply_instance.sys_prompt\n        if isinstance(sys_prompt, str):\n            invocation.system_instruction = [Text(content=sys_prompt)]\n        elif isinstance(sys_prompt, list):\n            invocation.system_instruction = sys_prompt\n\n    return invocation\n\n\ndef get_chatmodel_output_messages(chat_response: Any) -> List[Dict[str, Any]]:\n    \"\"\"Convert ChatResponse to OpenTelemetry standard output message format.\"\"\"\n    try:\n        if not isinstance(chat_response, ChatResponse):\n            return []\n\n        parts = []\n        for block in chat_response.content:\n            part = _convert_block_to_part(block)\n            if part:\n                parts.append(part)\n\n        if not parts:\n            parts.append({\"type\": \"text\", \"content\": \"\"})\n\n        return [\n            {\n                \"role\": \"assistant\",\n                \"parts\": parts,\n                \"finish_reason\": \"stop\",\n            }\n        ]\n\n    except Exception as e:\n        logger.warning(\n            f\"Error processing ChatResponse to output messages: {e}\"\n        )\n        return [\n            {\n                \"role\": \"assistant\",\n                \"parts\": [\n                    {\"type\": \"text\", \"content\": \"<error processing response>\"}\n                ],\n                \"finish_reason\": \"error\",\n            }\n        ]\n\n\ndef convert_agentscope_messages_to_genai_format(\n    messages: Any,\n    provider_name: Optional[str] = None,\n) -> List[InputMessage]:\n    \"\"\"Convert AgentScope messages to opentelemetry-util-genai InputMessage format.\n\n    This function is used by ExtendedTelemetryHandler which requires InputMessage objects.\n    \"\"\"\n    if not messages:\n        return []\n\n    if not isinstance(messages, list):\n        messages = [messages]\n\n    input_messages = []\n    for msg in messages:\n        if isinstance(msg, Msg):\n            msg_dict = _format_msg_to_parts(msg)\n        elif isinstance(msg, dict):\n            msg_dict = msg\n        else:\n            continue\n\n        role = msg_dict.get(\"role\", \"user\")\n        parts = msg_dict.get(\"parts\", [])\n\n        converted_parts = []\n        for part in parts:\n            part_type = part.get(\"type\")\n            if part_type == \"text\":\n                converted_parts.append(Text(content=part.get(\"content\", \"\")))\n            elif part_type == \"tool_call\":\n                converted_parts.append(\n                    ToolCall(\n                        id=part.get(\"id\"),\n                        name=part.get(\"name\", \"\"),\n                        arguments=part.get(\"arguments\", {}),\n                    )\n                )\n            elif part_type == \"tool_call_response\":\n                converted_parts.append(\n                    ToolCallResponse(\n                        id=part.get(\"id\"),\n                        response=part.get(\"result\", \"\"),\n                    )\n                )\n            else:\n                # Keep other types as-is\n                converted_parts.append(part)\n\n        input_messages.append(InputMessage(role=role, parts=converted_parts))\n\n    return input_messages\n\n\ndef convert_chatresponse_to_output_messages(\n    chat_response: Any,\n) -> List[OutputMessage]:\n    \"\"\"Convert ChatResponse to opentelemetry-util-genai OutputMessage format.\n\n    This function is used by ExtendedTelemetryHandler which requires OutputMessage objects.\n    \"\"\"\n    output_dicts = get_chatmodel_output_messages(chat_response)\n    if not output_dicts:\n        return []\n\n    output_messages = []\n    for msg_dict in output_dicts:\n        role = msg_dict.get(\"role\", \"assistant\")\n        parts_dicts = msg_dict.get(\"parts\", [])\n        finish_reason = msg_dict.get(\"finish_reason\", \"stop\")\n\n        converted_parts = []\n        for part in parts_dicts:\n            part_type = part.get(\"type\")\n            if part_type == \"text\":\n                converted_parts.append(Text(content=part.get(\"content\", \"\")))\n            elif part_type == \"tool_call\":\n                converted_parts.append(\n                    ToolCall(\n                        id=part.get(\"id\"),\n                        name=part.get(\"name\", \"\"),\n                        arguments=part.get(\"arguments\", {}),\n                    )\n                )\n            else:\n                converted_parts.append(part)\n\n        output_messages.append(\n            OutputMessage(\n                role=role,\n                parts=converted_parts,\n                finish_reason=finish_reason,\n            )\n        )\n\n    return output_messages\n\n\ndef convert_agent_response_to_output_messages(\n    agent_response: Any,\n) -> List[OutputMessage]:\n    \"\"\"Convert agent Msg response to opentelemetry-util-genai OutputMessage format.\n\n    This function is used by ExtendedTelemetryHandler which requires OutputMessage objects.\n    \"\"\"\n    if not hasattr(agent_response, \"content\"):\n        return []\n\n    try:\n        # Use _format_msg_to_parts to convert Msg to dict, then convert to OutputMessage\n        msg_dict = _format_msg_to_parts(agent_response)\n        parts_dicts = msg_dict.get(\"parts\", [])\n\n        converted_parts = []\n        for part in parts_dicts:\n            part_type = part.get(\"type\")\n            if part_type == \"text\":\n                converted_parts.append(Text(content=part.get(\"content\", \"\")))\n            elif part_type == \"tool_call\":\n                converted_parts.append(\n                    ToolCall(\n                        id=part.get(\"id\"),\n                        name=part.get(\"name\", \"\"),\n                        arguments=part.get(\"arguments\", {}),\n                    )\n                )\n            else:\n                converted_parts.append(Text(content=str(part)))\n\n        if not converted_parts:\n            converted_parts.append(Text(content=\"\"))\n\n        return [\n            OutputMessage(\n                role=\"assistant\",\n                parts=converted_parts,\n                finish_reason=\"stop\",\n            )\n        ]\n    except Exception as e:\n        logger.debug(f\"Failed to convert agent response: {e}\")\n        return []\n"}, "a73c05d3abf966ca71e01672712061f89085e965": {"t": 1785758590.8437877, "n": 18459, "h": "d295b5496b391277", "c": "# -*- coding: utf-8 -*-\n\"\"\"Attributes processor for span attributes.\"\"\"\n\nfrom __future__ import annotations\n\nimport datetime\nimport enum\nimport inspect\nimport json\nimport logging\nfrom dataclasses import is_dataclass\nfrom enum import Enum\nfrom typing import Any, Dict, List, Optional, Tuple\n\nfrom agentscope import _config\nfrom agentscope.agent import AgentBase\nfrom agentscope.embedding import EmbeddingModelBase\nfrom agentscope.message import Msg\nfrom agentscope.model import ChatModelBase, ChatResponse\nfrom pydantic import BaseModel\n\nfrom opentelemetry.semconv._incubating.attributes import (\n    gen_ai_attributes as GenAIAttributes,\n)\nfrom opentelemetry.util.genai.types import (\n    InputMessage,\n    OutputMessage,\n    Text,\n    ToolCall,\n    ToolCallResponse,\n)\n\nfrom .message_converter import get_message_converter\n\nlogger = logging.getLogger(__name__)\n\n\nclass AgentScopeGenAiProviderName(str, Enum):\n    \"\"\"Extended provider names not in standard OpenTelemetry semantic conventions.\"\"\"\n\n    OLLAMA = \"ollama\"\n    DASHSCOPE = \"dashscope\"\n    MOONSHOT = \"moonshot\"\n\n\n# Provider name mapping based on class names\n_PROVIDER_NAME_MAP = {\n    \"openai\": GenAIAttributes.GenAiProviderNameValues.OPENAI.value,\n    \"gemini\": GenAIAttributes.GenAiProviderNameValues.GCP_GEMINI.value,\n    \"anthropic\": GenAIAttributes.GenAiProviderNameValues.ANTHROPIC.value,\n    \"dashscope\": AgentScopeGenAiProviderName.DASHSCOPE.value,\n    \"ollama\": AgentScopeGenAiProviderName.OLLAMA.value,\n}\n\n# Base URL to provider mapping for OpenAI-compatible APIs\n_BASE_URL_PROVIDER_MAP = [\n    (\"openai.com\", GenAIAttributes.GenAiProviderNameValues.OPENAI.value),\n    (\n        \"api.deepseek.com\",\n        GenAIAttributes.GenAiProviderNameValues.DEEPSEEK.value,\n    ),\n    (\"dashscope.aliyuncs.com\", AgentScopeGenAiProviderName.DASHSCOPE.value),\n]\n\n\ndef get_provider_name(chat_model: ChatModelBase) -> str:\n    \"\"\"Parse chat model provider name\"\"\"\n    classname = chat_model.__class__.__name__\n    prefix = classname.removesuffix(\"ChatModel\").lower()\n\n    # Special handling for DashScopeChatModel with base_http_api_url\n    if (\n        prefix == \"dashscope\"\n        and hasattr(chat_model, \"base_http_api_url\")\n        and chat_model.base_http_api_url\n    ):\n        base_url = chat_model.base_http_api_url\n        for url_fragment, provider in _BASE_URL_PROVIDER_MAP:\n            if url_fragment in base_url:\n                return provider\n\n    return _PROVIDER_NAME_MAP.get(prefix, \"unknown\")\n\n\ndef get_embedding_provider_name(embedding_model: EmbeddingModelBase) -> str:\n    \"\"\"Parse embedding model provider name\"\"\"\n    class_name = embedding_model.__class__.__name__\n\n    if \"DashScope\" in class_name:\n        return AgentScopeGenAiProviderName.DASHSCOPE.value\n    elif \"OpenAI\" in class_name:\n        return GenAIAttributes.GenAiProviderNameValues.OPENAI.value\n    elif \"Gemini\" in class_name:\n        return GenAIAttributes.GenAiProviderNameValues.GCP_GEMINI.value\n    elif \"Ollama\" in class_name:\n        return AgentScopeGenAiProviderName.OLLAMA.value\n    else:\n        model_name = getattr(embedding_model, \"model_name\", \"\")\n        if \"text-embedding\" in model_name:\n            if \"qwen\" in model_name or \"dashscope\" in model_name:\n                return AgentScopeGenAiProviderName.DASHSCOPE.value\n            else:\n                return GenAIAttributes.GenAiProviderNameValues.OPENAI.value\n        return \"agentscope\"\n\n\ndef _get_tool_definitions(\n    tools: list[dict[str, Any]] | None,\n    tool_choice: str | None,\n) -> str | None:\n    \"\"\"Extract and serialize tool definitions for tracing.\"\"\"\n    if tools is None or not isinstance(tools, list) or len(tools) == 0:\n        return None\n\n    if tool_choice == \"none\":\n        return None\n\n    try:\n        flat_tools = []\n        for tool in tools:\n            if not isinstance(tool, dict) or \"function\" not in tool:\n                continue\n\n            func_def = tool[\"function\"]\n            flat_tool = {\n                \"type\": tool.get(\"type\", \"function\"),\n                \"name\": func_def.get(\"name\"),\n                \"description\": func_def.get(\"description\"),\n                \"parameters\": func_def.get(\"parameters\"),\n            }\n            flat_tool = {k: v for k, v in flat_tool.items() if v is not None}\n            flat_tools.append(flat_tool)\n\n        if flat_tools:\n            return _serialize_to_str(flat_tools)\n        return None\n\n    except Exception:\n        return None\n\n\ndef _to_serializable(obj: Any) -> Any:\n    \"\"\"Convert an object to a JSON serializable type.\"\"\"\n    if isinstance(obj, (str, int, bool, float, type(None))):\n        return obj\n    elif isinstance(obj, (list, tuple, set, frozenset)):\n        return [_to_serializable(x) for x in obj]\n    elif isinstance(obj, dict):\n        return {str(key): _to_serializable(val) for (key, val) in obj.items()}\n    elif isinstance(obj, (Msg, BaseModel)) or is_dataclass(obj):\n        return repr(obj)\n    elif inspect.isclass(obj) and issubclass(obj, BaseModel):\n        return repr(obj)\n    elif isinstance(obj, (datetime.date, datetime.datetime, datetime.time)):\n        return obj.isoformat()\n    elif isinstance(obj, datetime.timedelta):\n        return obj.total_seconds()\n    elif isinstance(obj, enum.Enum):\n        return _to_serializable(obj.value)\n    else:\n        return str(obj)\n\n\ndef _serialize_to_str(value: Any) -> str:\n    \"\"\"Serialize value to JSON string\"\"\"\n    try:\n        return json.dumps(value, ensure_ascii=False)\n    except TypeError:\n        return json.dumps(_to_serializable(value), ensure_ascii=False)\n\n\ndef _convert_block_to_part(block: Dict[str, Any]) -> Dict[str, Any] | None:\n    \"\"\"Convert content block to standardized part format.\"\"\"\n    block_type = block.get(\"type\")\n\n    if block_type == \"text\":\n        return {\"type\": \"text\", \"content\": block.get(\"text\", \"\")}\n\n    elif block_type == \"thinking\":\n        return {\n            \"type\": \"text\",\n            \"content\": f\"[Thinking] {block.get('thinking', '')}\",\n        }\n\n    elif block_type == \"tool_use\":\n        return {\n            \"type\": \"tool_call\",\n            \"id\": block.get(\"id\", \"\"),\n            \"name\": block.get(\"name\", \"\"),\n            \"arguments\": block.get(\"input\", {}),\n        }\n\n    elif block_type == \"tool_result\":\n        output = block.get(\"output\", \"\")\n        result = (\n            _serialize_to_str(output)\n            if isinstance(output, (list, dict))\n            else str(output)\n        )\n        return {\n            \"type\": \"tool_call_response\",\n            \"id\": block.get(\"id\", \"\"),\n            \"result\": result,\n        }\n\n    elif block_type == \"image\":\n        source = block.get(\"source\", {})\n        source_type = source.get(\"type\")\n        if source_type == \"url\":\n            return {\"type\": \"image\", \"url\": source.get(\"url\", \"\")}\n        elif source_type == \"base64\":\n            data = source.get(\"data\", \"\")\n            media_type = source.get(\"media_type\", \"image/jpeg\")\n            return {\"type\": \"image\", \"url\": f\"data:{media_type};base64,{data}\"}\n\n    elif block_type == \"audio\":\n        return {\"type\": \"audio\", \"source\": block.get(\"source\", {})}\n\n    elif block_type == \"video\":\n        return {\"type\": \"video\", \"source\": block.get(\"source\", {})}\n\n    return None\n\n\ndef _format_msg_to_parts(msg: Msg) -> dict[str, Any]:\n    \"\"\"Convert Msg to standard format (parts structure)\"\"\"\n    try:\n        parts = []\n        for block in msg.get_content_blocks():\n            part = _convert_block_to_part(block)\n            if part:\n                parts.append(part)\n\n        formatted_msg = {\"role\": msg.role, \"parts\": parts}\n        if msg.name:\n            formatted_msg[\"name\"] = msg.name\n\n        return formatted_msg\n\n    except Exception as e:\n        logger.debug(f\"Error formatting message: {e}\")\n        return {\n            \"role\": msg.role,\n            \"parts\": [\n                {\n                    \"type\": \"text\",\n                    \"content\": str(msg.content) if msg.content else \"\",\n                }\n            ],\n        }\n\n\ndef extract_llm_attributes(\n    call_instance: ChatModelBase,\n    call_args: Tuple[Any, ...],\n    call_kwargs: Dict[str, Any],\n) -> Dict[str, Any]:\n    \"\"\"Extract LLM call attributes\"\"\"\n    provider_name = get_provider_name(call_instance)\n\n    messages = call_args[0] if call_args else call_kwargs.get(\"messages\")\n\n    if messages:\n        try:\n            input_messages = get_message_converter(provider_name)(messages)\n        except Exception as e:\n            logger.warning(\n                f\"Failed to convert input messages: {e}\", exc_info=True\n            )\n            input_messages = {\"args\": call_args, \"kwargs\": call_kwargs}\n    else:\n        logger.debug(\"No messages provided for LLM call\")\n        input_messages = {\"args\": call_args, \"kwargs\": call_kwargs}\n\n    tool_definitions = None\n    if not call_kwargs.get(\"structured_model\", False):\n        tool_definitions = _get_tool_definitions(\n            call_kwargs.get(\"tools\"), call_kwargs.get(\"tool_choice\")\n        )\n\n    return {\n        \"operation_name\": GenAIAttributes.GenAiOperationNameValues.CHAT.value,\n        \"provider_name\": provider_name,\n        \"request_model\": getattr(call_instance, \"model_name\", \"unknown_model\"),\n        \"request_max_tokens\": call_kwargs.get(\"max_tokens\"),\n        \"request_temperature\": call_kwargs.get(\"temperature\"),\n        \"request_top_p\": call_kwargs.get(\"top_p\"),\n        \"request_top_k\": call_kwargs.get(\"top_k\"),\n        \"request_stop_sequences\": call_kwargs.get(\"stop_sequences\"),\n        \"request_tool_definitions\": tool_definitions,\n        \"input_messages\": _serialize_to_str(input_messages),\n    }\n\n\ndef extract_embedding_attributes(\n    call_instance: EmbeddingModelBase,\n    call_args: Tuple[Any, ...],\n    call_kwargs: Dict[str, Any],\n) -> Dict[str, Any]:\n    \"\"\"Extract Embedding call attributes\"\"\"\n    provider_name = get_embedding_provider_name(call_instance)\n\n    text_for_embedding = call_args[0] if call_args else call_kwargs.get(\"text\")\n\n    if text_for_embedding:\n        input_message = []\n        for text_item in text_for_embedding:\n            input_message.append(\n                {\n                    \"role\": \"user\",\n                    \"content\": [{\"type\": \"text\", \"text\": text_item}],\n                }\n            )\n        input_messages = input_message\n    else:\n        logger.debug(\"No text provided for embedding call\")\n        input_messages = {\"args\": call_args, \"kwargs\": call_kwargs}\n\n    return {\n        \"operation_name\": GenAIAttributes.GenAiOperationNameValues.EMBEDDINGS.value,\n        \"provider_name\": provider_name,\n        \"request_model\": getattr(call_instance, \"model_name\", \"unknown_model\"),\n        \"request_encoding_formats\": call_kwargs.get(\"encoding_formats\"),\n        \"input_messages\": _serialize_to_str(input_messages),\n    }\n\n\ndef extract_agent_attributes(\n    reply_instance: AgentBase,\n    reply_args: Tuple[Any, ...],\n    reply_kwargs: Dict[str, Any],\n) -> Dict[str, Any]:\n    \"\"\"Extract Agent reply call attributes\"\"\"\n    msg = reply_args[0] if reply_args else reply_kwargs.get(\"msg\")\n\n    if msg:\n        try:\n            if isinstance(msg, Msg):\n                input_messages = _format_msg_to_parts(msg)\n            elif isinstance(msg, list):\n                input_messages = [\n                    _format_msg_to_parts(m) for m in msg if isinstance(m, Msg)\n                ]\n            else:\n                input_messages = []\n        except Exception as e:\n            logger.debug(f\"Error formatting agent messages: {e}\")\n            input_messages = []\n    else:\n        logger.debug(\"No msg provided for agent reply\")\n        input_messages = {\"args\": reply_args, \"kwargs\": reply_kwargs}\n\n    request_model = None\n    provider_name = None\n    if hasattr(reply_instance, \"model\") and reply_instance.model:\n        model = reply_instance.model\n        # Only set request_model if model has a model_name attribute\n        request_model = getattr(model, \"model_name\", None)\n        if isinstance(model, ChatModelBase):\n            provider_name = get_provider_name(model)\n\n    return {\n        \"operation_name\": GenAIAttributes.GenAiOperationNameValues.INVOKE_AGENT.value,\n        \"agent_id\": getattr(reply_instance, \"id\", \"unknown\"),\n        \"agent_name\": getattr(reply_instance, \"name\", \"unknown_agent\"),\n        \"agent_description\": inspect.getdoc(reply_instance.__class__)\n        or \"No description available\",\n        \"system_instructions\": reply_instance.sys_prompt\n        if hasattr(reply_instance, \"sys_prompt\")\n        else None,\n        \"request_model\": request_model,\n        \"provider_name\": provider_name,\n        \"conversation_id\": _config.run_id,\n        \"input_messages\": _serialize_to_str(input_messages),\n    }\n\n\ndef get_chatmodel_output_messages(chat_response: Any) -> List[Dict[str, Any]]:\n    \"\"\"Convert ChatResponse to OpenTelemetry standard output message format.\"\"\"\n    try:\n        if not isinstance(chat_response, ChatResponse):\n            return []\n\n        parts = []\n        for block in chat_response.content:\n            part = _convert_block_to_part(block)\n            if part:\n                parts.append(part)\n\n        if not parts:\n            parts.append({\"type\": \"text\", \"content\": \"\"})\n\n        return [\n            {\n                \"role\": \"assistant\",\n                \"parts\": parts,\n                \"finish_reason\": \"stop\",\n            }\n        ]\n\n    except Exception as e:\n        logger.warning(\n            f\"Error processing ChatResponse to output messages: {e}\"\n        )\n        return [\n            {\n                \"role\": \"assistant\",\n                \"parts\": [\n                    {\"type\": \"text\", \"content\": \"<error processing response>\"}\n                ],\n                \"finish_reason\": \"error\",\n            }\n        ]\n\n\ndef convert_agentscope_messages_to_genai_format(\n    messages: Any,\n    provider_name: Optional[str] = None,\n) -> List[InputMessage]:\n    \"\"\"Convert AgentScope messages to opentelemetry-util-genai InputMessage format.\n\n    This function is used by ExtendedTelemetryHandler which requires InputMessage objects.\n    \"\"\"\n    if not messages:\n        return []\n\n    if not isinstance(messages, list):\n        messages = [messages]\n\n    input_messages = []\n    for msg in messages:\n        if isinstance(msg, Msg):\n            msg_dict = _format_msg_to_parts(msg)\n        elif isinstance(msg, dict):\n            msg_dict = msg\n        else:\n            continue\n\n        role = msg_dict.get(\"role\", \"user\")\n        parts = msg_dict.get(\"parts\", [])\n\n        converted_parts = []\n        for part in parts:\n            part_type = part.get(\"type\")\n            if part_type == \"text\":\n                converted_parts.append(Text(content=part.get(\"content\", \"\")))\n            elif part_type == \"tool_call\":\n                converted_parts.append(\n                    ToolCall(\n                        id=part.get(\"id\"),\n                        name=part.get(\"name\", \"\"),\n                        arguments=part.get(\"arguments\", {}),\n                    )\n                )\n            elif part_type == \"tool_call_response\":\n                converted_parts.append(\n                    ToolCallResponse(\n                        id=part.get(\"id\"),\n                        response=part.get(\"result\", \"\"),\n                    )\n                )\n            else:\n                # Keep other types as-is\n                converted_parts.append(part)\n\n        input_messages.append(InputMessage(role=role, parts=converted_parts))\n\n    return input_messages\n\n\ndef convert_chatresponse_to_output_messages(\n    chat_response: Any,\n) -> List[OutputMessage]:\n    \"\"\"Convert ChatResponse to opentelemetry-util-genai OutputMessage format.\n\n    This function is used by ExtendedTelemetryHandler which requires OutputMessage objects.\n    \"\"\"\n    output_dicts = get_chatmodel_output_messages(chat_response)\n    if not output_dicts:\n        return []\n\n    output_messages = []\n    for msg_dict in output_dicts:\n        role = msg_dict.get(\"role\", \"assistant\")\n        parts_dicts = msg_dict.get(\"parts\", [])\n        finish_reason = msg_dict.get(\"finish_reason\", \"stop\")\n\n        converted_parts = []\n        for part in parts_dicts:\n            part_type = part.get(\"type\")\n            if part_type == \"text\":\n                converted_parts.append(Text(content=part.get(\"content\", \"\")))\n            elif part_type == \"tool_call\":\n                converted_parts.append(\n                    ToolCall(\n                        id=part.get(\"id\"),\n                        name=part.get(\"name\", \"\"),\n                        arguments=part.get(\"arguments\", {}),\n                    )\n                )\n            else:\n                converted_parts.append(part)\n\n        output_messages.append(\n            OutputMessage(\n                role=role,\n                parts=converted_parts,\n                finish_reason=finish_reason,\n            )\n        )\n\n    return output_messages\n\n\ndef convert_agent_response_to_output_messages(\n    agent_response: Any,\n) -> List[OutputMessage]:\n    \"\"\"Convert agent Msg response to opentelemetry-util-genai OutputMessage format.\n\n    This function is used by ExtendedTelemetryHandler which requires OutputMessage objects.\n    \"\"\"\n    if not hasattr(agent_response, \"content\"):\n        return []\n\n    try:\n        # Use _format_msg_to_parts to convert Msg to dict, then convert to OutputMessage\n        msg_dict = _format_msg_to_parts(agent_response)\n        parts_dicts = msg_dict.get(\"parts\", [])\n\n        converted_parts = []\n        for part in parts_dicts:\n            part_type = part.get(\"type\")\n            if part_type == \"text\":\n                converted_parts.append(Text(content=part.get(\"content\", \"\")))\n            elif part_type == \"tool_call\":\n                converted_parts.append(\n                    ToolCall(\n                        id=part.get(\"id\"),\n                        name=part.get(\"name\", \"\"),\n                        arguments=part.get(\"arguments\", {}),\n                    )\n                )\n            else:\n                converted_parts.append(Text(content=str(part)))\n\n        if not converted_parts:\n            converted_parts.append(Text(content=\"\"))\n\n        return [\n            OutputMessage(\n                role=\"assistant\",\n                parts=converted_parts,\n                finish_reason=\"stop\",\n            )\n        ]\n    except Exception as e:\n        logger.debug(f\"Failed to convert agent response: {e}\")\n        return []\n"}, "8737f9105299b78439747b2669725df147a68025": {"t": 1785758592.2688181, "n": 18743, "h": "7f61c81b65d4bb69", "c": "# -*- coding: utf-8 -*-\n\"\"\"Attributes processor for span attributes.\"\"\"\n\nfrom __future__ import annotations\n\nimport datetime\nimport enum\nimport inspect\nimport json\nimport logging\nfrom dataclasses import is_dataclass\nfrom enum import Enum\nfrom typing import Any, Dict, List, Optional, Tuple\n\nfrom agentscope import _config\nfrom agentscope.agent import AgentBase\nfrom agentscope.embedding import EmbeddingModelBase\nfrom agentscope.message import Msg\nfrom agentscope.model import ChatModelBase, ChatResponse\nfrom pydantic import BaseModel\n\nfrom opentelemetry.semconv._incubating.attributes import (\n    gen_ai_attributes as GenAIAttributes,\n)\nfrom opentelemetry.util.genai.types import (\n    InputMessage,\n    OutputMessage,\n    Text,\n    ToolCall,\n    ToolCallResponse,\n)\n\nfrom .message_converter import get_message_converter\n\nlogger = logging.getLogger(__name__)\n\n\nclass AgentScopeGenAiProviderName(str, Enum):\n    \"\"\"Extended provider names not in standard OpenTelemetry semantic conventions.\"\"\"\n\n    OLLAMA = \"ollama\"\n    DASHSCOPE = \"dashscope\"\n    MOONSHOT = \"moonshot\"\n\n\n# Provider name mapping based on class names\n_PROVIDER_NAME_MAP = {\n    \"openai\": GenAIAttributes.GenAiProviderNameValues.OPENAI.value,\n    \"gemini\": GenAIAttributes.GenAiProviderNameValues.GCP_GEMINI.value,\n    \"anthropic\": GenAIAttributes.GenAiProviderNameValues.ANTHROPIC.value,\n    \"dashscope\": AgentScopeGenAiProviderName.DASHSCOPE.value,\n    \"ollama\": AgentScopeGenAiProviderName.OLLAMA.value,\n}\n\n# Base URL to provider mapping for OpenAI-compatible APIs\n_BASE_URL_PROVIDER_MAP = [\n    (\"openai.com\", GenAIAttributes.GenAiProviderNameValues.OPENAI.value),\n    (\n        \"api.deepseek.com\",\n        GenAIAttributes.GenAiProviderNameValues.DEEPSEEK.value,\n    ),\n    (\"dashscope.aliyuncs.com\", AgentScopeGenAiProviderName.DASHSCOPE.value),\n]\n\n\ndef get_provider_name(chat_model: ChatModelBase) -> str:\n    \"\"\"Parse chat model provider name\"\"\"\n    classname = chat_model.__class__.__name__\n    prefix = classname.removesuffix(\"ChatModel\").lower()\n\n    # Special handling for DashScopeChatModel with base_http_api_url\n    if (\n        prefix == \"dashscope\"\n        and hasattr(chat_model, \"base_http_api_url\")\n        and chat_model.base_http_api_url\n    ):\n        base_url = chat_model.base_http_api_url\n        for url_fragment, provider in _BASE_URL_PROVIDER_MAP:\n            if url_fragment in base_url:\n                return provider\n\n    return _PROVIDER_NAME_MAP.get(prefix, \"unknown\")\n\n\ndef get_embedding_provider_name(embedding_model: EmbeddingModelBase) -> str:\n    \"\"\"Parse embedding model provider name\"\"\"\n    class_name = embedding_model.__class__.__name__\n\n    if \"DashScope\" in class_name:\n        return AgentScopeGenAiProviderName.DASHSCOPE.value\n    elif \"OpenAI\" in class_name:\n        return GenAIAttributes.GenAiProviderNameValues.OPENAI.value\n    elif \"Gemini\" in class_name:\n        return GenAIAttributes.GenAiProviderNameValues.GCP_GEMINI.value\n    elif \"Ollama\" in class_name:\n        return AgentScopeGenAiProviderName.OLLAMA.value\n    else:\n        model_name = getattr(embedding_model, \"model_name\", \"\")\n        if \"text-embedding\" in model_name:\n            if \"qwen\" in model_name or \"dashscope\" in model_name:\n                return AgentScopeGenAiProviderName.DASHSCOPE.value\n            else:\n                return GenAIAttributes.GenAiProviderNameValues.OPENAI.value\n        return \"agentscope\"\n\n\ndef _get_tool_definitions(\n    tools: list[dict[str, Any]] | None,\n    tool_choice: str | None,\n) -> str | None:\n    \"\"\"Extract and serialize tool definitions for tracing.\"\"\"\n    if tools is None or not isinstance(tools, list) or len(tools) == 0:\n        return None\n\n    if tool_choice == \"none\":\n        return None\n\n    try:\n        flat_tools = []\n        for tool in tools:\n            if not isinstance(tool, dict) or \"function\" not in tool:\n                continue\n\n            func_def = tool[\"function\"]\n            flat_tool = {\n                \"type\": tool.get(\"type\", \"function\"),\n                \"name\": func_def.get(\"name\"),\n                \"description\": func_def.get(\"description\"),\n                \"parameters\": func_def.get(\"parameters\"),\n            }\n            flat_tool = {k: v for k, v in flat_tool.items() if v is not None}\n            flat_tools.append(flat_tool)\n\n        if flat_tools:\n            return _serialize_to_str(flat_tools)\n        return None\n\n    except Exception:\n        return None\n\n\ndef _to_serializable(obj: Any) -> Any:\n    \"\"\"Convert an object to a JSON serializable type.\"\"\"\n    if isinstance(obj, (str, int, bool, float, type(None))):\n        return obj\n    elif isinstance(obj, (list, tuple, set, frozenset)):\n        return [_to_serializable(x) for x in obj]\n    elif isinstance(obj, dict):\n        return {str(key): _to_serializable(val) for (key, val) in obj.items()}\n    elif isinstance(obj, (Msg, BaseModel)) or is_dataclass(obj):\n        return repr(obj)\n    elif inspect.isclass(obj) and issubclass(obj, BaseModel):\n        return repr(obj)\n    elif isinstance(obj, (datetime.date, datetime.datetime, datetime.time)):\n        return obj.isoformat()\n    elif isinstance(obj, datetime.timedelta):\n        return obj.total_seconds()\n    elif isinstance(obj, enum.Enum):\n        return _to_serializable(obj.value)\n    else:\n        return str(obj)\n\n\ndef _serialize_to_str(value: Any) -> str:\n    \"\"\"Serialize value to JSON string\"\"\"\n    try:\n        return json.dumps(value, ensure_ascii=False)\n    except TypeError:\n        return json.dumps(_to_serializable(value), ensure_ascii=False)\n\n\ndef _convert_block_to_part(block: Dict[str, Any]) -> Dict[str, Any] | None:\n    \"\"\"Convert content block to standardized part format.\"\"\"\n    block_type = block.get(\"type\")\n\n    if block_type == \"text\":\n        return {\"type\": \"text\", \"content\": block.get(\"text\", \"\")}\n\n    elif block_type == \"thinking\":\n        return {\n            \"type\": \"text\",\n            \"content\": f\"[Thinking] {block.get('thinking', '')}\",\n        }\n\n    elif block_type == \"tool_use\":\n        return {\n            \"type\": \"tool_call\",\n            \"id\": block.get(\"id\", \"\"),\n            \"name\": block.get(\"name\", \"\"),\n            \"arguments\": block.get(\"input\", {}),\n        }\n\n    elif block_type == \"tool_result\":\n        output = block.get(\"output\", \"\")\n        result = (\n            _serialize_to_str(output)\n            if isinstance(output, (list, dict))\n            else str(output)\n        )\n        return {\n            \"type\": \"tool_call_response\",\n            \"id\": block.get(\"id\", \"\"),\n            \"result\": result,\n        }\n\n    elif block_type == \"image\":\n        source = block.get(\"source\", {})\n        source_type = source.get(\"type\")\n        if source_type == \"url\":\n            return {\"type\": \"image\", \"url\": source.get(\"url\", \"\")}\n        elif source_type == \"base64\":\n            data = source.get(\"data\", \"\")\n            media_type = source.get(\"media_type\", \"image/jpeg\")\n            return {\"type\": \"image\", \"url\": f\"data:{media_type};base64,{data}\"}\n\n    elif block_type == \"audio\":\n        return {\"type\": \"audio\", \"source\": block.get(\"source\", {})}\n\n    elif block_type == \"video\":\n        return {\"type\": \"video\", \"source\": block.get(\"source\", {})}\n\n    return None\n\n\ndef _format_msg_to_parts(msg: Msg) -> dict[str, Any]:\n    \"\"\"Convert Msg to standard format (parts structure)\"\"\"\n    try:\n        parts = []\n        for block in msg.get_content_blocks():\n            part = _convert_block_to_part(block)\n            if part:\n                parts.append(part)\n\n        formatted_msg = {\"role\": msg.role, \"parts\": parts}\n        if msg.name:\n            formatted_msg[\"name\"] = msg.name\n\n        return formatted_msg\n\n    except Exception as e:\n        logger.debug(f\"Error formatting message: {e}\")\n        return {\n            \"role\": msg.role,\n            \"parts\": [\n                {\n                    \"type\": \"text\",\n                    \"content\": str(msg.content) if msg.content else \"\",\n                }\n            ],\n        }\n\n\ndef extract_llm_attributes(\n    call_instance: ChatModelBase,\n    call_args: Tuple[Any, ...],\n    call_kwargs: Dict[str, Any],\n) -> Dict[str, Any]:\n    \"\"\"Extract LLM call attributes\"\"\"\n    provider_name = get_provider_name(call_instance)\n\n    messages = call_args[0] if call_args else call_kwargs.get(\"messages\")\n\n    if messages:\n        try:\n            input_messages = get_message_converter(provider_name)(messages)\n        except Exception as e:\n            logger.warning(\n                f\"Failed to convert input messages: {e}\", exc_info=True\n            )\n            input_messages = {\"args\": call_args, \"kwargs\": call_kwargs}\n    else:\n        logger.debug(\"No messages provided for LLM call\")\n        input_messages = {\"args\": call_args, \"kwargs\": call_kwargs}\n\n    tool_definitions = None\n    if not call_kwargs.get(\"structured_model\", False):\n        tool_definitions = _get_tool_definitions(\n            call_kwargs.get(\"tools\"), call_kwargs.get(\"tool_choice\")\n        )\n\n    return {\n        \"operation_name\": GenAIAttributes.GenAiOperationNameValues.CHAT.value,\n        \"provider_name\": provider_name,\n        \"request_model\": getattr(call_instance, \"model_name\", \"unknown_model\"),\n        \"request_max_tokens\": call_kwargs.get(\"max_tokens\"),\n        \"request_temperature\": call_kwargs.get(\"temperature\"),\n        \"request_top_p\": call_kwargs.get(\"top_p\"),\n        \"request_top_k\": call_kwargs.get(\"top_k\"),\n        \"request_stop_sequences\": call_kwargs.get(\"stop_sequences\"),\n        \"request_tool_definitions\": tool_definitions,\n        \"input_messages\": _serialize_to_str(input_messages),\n    }\n\n\ndef extract_embedding_attributes(\n    call_instance: EmbeddingModelBase,\n    call_args: Tuple[Any, ...],\n    call_kwargs: Dict[str, Any],\n) -> Dict[str, Any]:\n    \"\"\"Extract Embedding call attributes\"\"\"\n    provider_name = get_embedding_provider_name(call_instance)\n\n    text_for_embedding = call_args[0] if call_args else call_kwargs.get(\"text\")\n\n    if text_for_embedding:\n        input_message = []\n        for text_item in text_for_embedding:\n            input_message.append(\n                {\n                    \"role\": \"user\",\n                    \"content\": [{\"type\": \"text\", \"text\": text_item}],\n                }\n            )\n        input_messages = input_message\n    else:\n        logger.debug(\"No text provided for embedding call\")\n        input_messages = {\"args\": call_args, \"kwargs\": call_kwargs}\n\n    return {\n        \"operation_name\": GenAIAttributes.GenAiOperationNameValues.EMBEDDINGS.value,\n        \"provider_name\": provider_name,\n        \"request_model\": getattr(call_instance, \"model_name\", \"unknown_model\"),\n        \"request_encoding_formats\": call_kwargs.get(\"encoding_formats\"),\n        \"input_messages\": _serialize_to_str(input_messages),\n    }\n\n\ndef extract_agent_attributes(\n    reply_instance: AgentBase,\n    reply_args: Tuple[Any, ...],\n    reply_kwargs: Dict[str, Any],\n) -> Dict[str, Any]:\n    \"\"\"Extract Agent reply call attributes\"\"\"\n    msg = reply_args[0] if reply_args else reply_kwargs.get(\"msg\")\n\n    if msg:\n        try:\n            if isinstance(msg, Msg):\n                input_messages = _format_msg_to_parts(msg)\n            elif isinstance(msg, list):\n                input_messages = [\n                    _format_msg_to_parts(m) for m in msg if isinstance(m, Msg)\n                ]\n            else:\n                input_messages = []\n        except Exception as e:\n            logger.debug(f\"Error formatting agent messages: {e}\")\n            input_messages = []\n    else:\n        logger.debug(\"No msg provided for agent reply\")\n        input_messages = {\"args\": reply_args, \"kwargs\": reply_kwargs}\n\n    request_model = None\n    provider_name = None\n    if hasattr(reply_instance, \"model\") and reply_instance.model:\n        model = reply_instance.model\n        # Only set request_model if model has a model_name attribute\n        request_model = getattr(model, \"model_name\", None)\n        if isinstance(model, ChatModelBase):\n            provider_name = get_provider_name(model)\n\n    return {\n        \"operation_name\": GenAIAttributes.GenAiOperationNameValues.INVOKE_AGENT.value,\n        \"agent_id\": getattr(reply_instance, \"id\", \"unknown\"),\n        \"agent_name\": getattr(reply_instance, \"name\", \"unknown_agent\"),\n        \"agent_description\": inspect.getdoc(reply_instance.__class__)\n        or \"No description available\",\n        \"system_instructions\": reply_instance.sys_prompt\n        if hasattr(reply_instance, \"sys_prompt\")\n        else None,\n        \"request_model\": request_model,\n        \"provider_name\": provider_name,\n        \"conversation_id\": _config.run_id,\n        \"input_messages\": _serialize_to_str(input_messages),\n    }\n\n\ndef get_chatmodel_output_messages(chat_response: Any) -> List[Dict[str, Any]]:\n    \"\"\"Convert ChatResponse to OpenTelemetry standard output message format.\"\"\"\n    try:\n        if not isinstance(chat_response, ChatResponse):\n            return []\n\n        parts = []\n        for block in chat_response.content:\n            part = _convert_block_to_part(block)\n            if part:\n                parts.append(part)\n\n        if not parts:\n            parts.append({\"type\": \"text\", \"content\": \"\"})\n\n        return [\n            {\n                \"role\": \"assistant\",\n                \"parts\": parts,\n                \"finish_reason\": \"stop\",\n            }\n        ]\n\n    except Exception as e:\n        logger.warning(\n            f\"Error processing ChatResponse to output messages: {e}\"\n        )\n        return [\n            {\n                \"role\": \"assistant\",\n                \"parts\": [\n                    {\"type\": \"text\", \"content\": \"<error processing response>\"}\n                ],\n                \"finish_reason\": \"error\",\n            }\n        ]\n\n\ndef convert_agentscope_messages_to_genai_format(\n    messages: Any,\n    provider_name: Optional[str] = None,\n) -> List[InputMessage]:\n    \"\"\"Convert AgentScope messages to opentelemetry-util-genai InputMessage format.\n\n    This function is used by ExtendedTelemetryHandler which requires InputMessage objects.\n    \"\"\"\n    if not messages:\n        return []\n\n    if not isinstance(messages, list):\n        messages = [messages]\n\n    input_messages = []\n    for msg in messages:\n        if isinstance(msg, Msg):\n            msg_dict = _format_msg_to_parts(msg)\n        elif isinstance(msg, dict):\n            if provider_name:\n                try:\n                    converted = get_message_converter(provider_name)([msg])\n                    msg_dict = converted[0] if converted else msg\n                except Exception:\n                    msg_dict = msg\n            else:\n                msg_dict = msg\n        else:\n            continue\n\n        role = msg_dict.get(\"role\", \"user\")\n        parts = msg_dict.get(\"parts\", [])\n\n        converted_parts = []\n        for part in parts:\n            part_type = part.get(\"type\")\n            if part_type == \"text\":\n                converted_parts.append(Text(content=part.get(\"content\", \"\")))\n            elif part_type == \"tool_call\":\n                converted_parts.append(\n                    ToolCall(\n                        id=part.get(\"id\"),\n                        name=part.get(\"name\", \"\"),\n                        arguments=part.get(\"arguments\", {}),\n                    )\n                )\n            elif part_type == \"tool_call_response\":\n                converted_parts.append(\n                    ToolCallResponse(\n                        id=part.get(\"id\"),\n                        response=part.get(\"result\", \"\"),\n                    )\n                )\n            else:\n                # Keep other types as-is\n                converted_parts.append(part)\n\n        input_messages.append(InputMessage(role=role, parts=converted_parts))\n\n    return input_messages\n\n\ndef convert_chatresponse_to_output_messages(\n    chat_response: Any,\n) -> List[OutputMessage]:\n    \"\"\"Convert ChatResponse to opentelemetry-util-genai OutputMessage format.\n\n    This function is used by ExtendedTelemetryHandler which requires OutputMessage objects.\n    \"\"\"\n    output_dicts = get_chatmodel_output_messages(chat_response)\n    if not output_dicts:\n        return []\n\n    output_messages = []\n    for msg_dict in output_dicts:\n        role = msg_dict.get(\"role\", \"assistant\")\n        parts_dicts = msg_dict.get(\"parts\", [])\n        finish_reason = msg_dict.get(\"finish_reason\", \"stop\")\n\n        converted_parts = []\n        for part in parts_dicts:\n            part_type = part.get(\"type\")\n            if part_type == \"text\":\n                converted_parts.append(Text(content=part.get(\"content\", \"\")))\n            elif part_type == \"tool_call\":\n                converted_parts.append(\n                    ToolCall(\n                        id=part.get(\"id\"),\n                        name=part.get(\"name\", \"\"),\n                        arguments=part.get(\"arguments\", {}),\n                    )\n                )\n            else:\n                converted_parts.append(part)\n\n        output_messages.append(\n            OutputMessage(\n                role=role,\n                parts=converted_parts,\n                finish_reason=finish_reason,\n            )\n        )\n\n    return output_messages\n\n\ndef convert_agent_response_to_output_messages(\n    agent_response: Any,\n) -> List[OutputMessage]:\n    \"\"\"Convert agent Msg response to opentelemetry-util-genai OutputMessage format.\n\n    This function is used by ExtendedTelemetryHandler which requires OutputMessage objects.\n    \"\"\"\n    if not hasattr(agent_response, \"content\"):\n        return []\n\n    try:\n        # Use _format_msg_to_parts to convert Msg to dict, then convert to OutputMessage\n        msg_dict = _format_msg_to_parts(agent_response)\n        parts_dicts = msg_dict.get(\"parts\", [])\n\n        converted_parts = []\n        for part in parts_dicts:\n            part_type = part.get(\"type\")\n            if part_type == \"text\":\n                converted_parts.append(Text(content=part.get(\"content\", \"\")))\n            elif part_type == \"tool_call\":\n                converted_parts.append(\n                    ToolCall(\n                        id=part.get(\"id\"),\n                        name=part.get(\"name\", \"\"),\n                        arguments=part.get(\"arguments\", {}),\n                    )\n                )\n            else:\n                converted_parts.append(Text(content=str(part)))\n\n        if not converted_parts:\n            converted_parts.append(Text(content=\"\"))\n\n        return [\n            OutputMessage(\n                role=\"assistant\",\n                parts=converted_parts,\n                finish_reason=\"stop\",\n            )\n        ]\n    except Exception as e:\n        logger.debug(f\"Failed to convert agent response: {e}\")\n        return []\n"}, "45004380c53c019e8e8ee07a73d66997b00a8ba2": {"t": 1785758593.6540606, "n": 777, "h": "6380652f904c3509", "c": "import os\nfrom importlib.metadata import version\n\nfrom packaging import version as pkg_version\n\ntry:\n    from pydantic import BaseModel\nexcept ImportError:\n    BaseModel = None\n\n# 获取 AgentScope 版本\ntry:\n    _AGENTSCOPE_VERSION = version(\"agentscope\")\nexcept Exception:\n    _AGENTSCOPE_VERSION = \"0.0.0\"\n\n# 环境变量配置\nOTEL_INSTRUMENTATION_GENAI_CAPTURE_MESSAGE_CONTENT = (\n    \"OTEL_INSTRUMENTATION_GENAI_CAPTURE_MESSAGE_CONTENT\"\n)\n\n\ndef is_agentscope_v1():\n    \"\"\"检查是否为 AgentScope v1.0.0 及以上版本\"\"\"\n    return pkg_version.parse(_AGENTSCOPE_VERSION) >= pkg_version.parse(\"1.0.0\")\n\n\ndef is_content_enabled() -> bool:\n    \"\"\"检查是否应该捕获消息内容\"\"\"\n    return (\n        os.getenv(\n            OTEL_INSTRUMENTATION_GENAI_CAPTURE_MESSAGE_CONTENT, \"false\"\n        ).lower()\n        == \"true\"\n    )\n"}, "fe887cf3445eb8fe548b7e3a1c5a20995d0e1dc8": {"t": 1785758595.067652, "n": 981, "h": "ac4d4a8925345687", "c": "import os\nimport json\nimport datetime\nimport enum\nimport inspect\nfrom importlib.metadata import version\nfrom packaging import version as pkg_version\nfrom dataclasses import is_dataclass\nfrom typing import Any, Dict, List, Optional, Union, Tuple, Iterable\n\nfrom opentelemetry.util.types import AttributeValue\n\ntry:\n    from pydantic import BaseModel\nexcept ImportError:\n    BaseModel = None\n\n# 获取 AgentScope 版本\ntry:\n    import agentscope\n    _AGENTSCOPE_VERSION = version(\"agentscope\")\nexcept Exception:\n    _AGENTSCOPE_VERSION = \"0.0.0\"\n\n# 环境变量配置\nOTEL_INSTRUMENTATION_GENAI_CAPTURE_MESSAGE_CONTENT = (\n    \"OTEL_INSTRUMENTATION_GENAI_CAPTURE_MESSAGE_CONTENT\"\n)\n\n\ndef is_agentscope_v1():\n    \"\"\"检查是否为 AgentScope v1.0.0 及以上版本\"\"\"\n    return pkg_version.parse(_AGENTSCOPE_VERSION) >= pkg_version.parse(\"1.0.0\")\n\ndef is_content_enabled() -> bool:\n    \"\"\"检查是否应该捕获消息内容\"\"\"\n    return (\n        os.getenv(OTEL_INSTRUMENTATION_GENAI_CAPTURE_MESSAGE_CONTENT, \"false\").lower() == \"true\"\n    )\n"}}}}