OpenAI agents: address conformance test issues

[lmolkova]

opentelemetry.test.weaver_live_check.LiveCheckError: Semconv violations found:

• [missing_attribute] Attribute 'gen_ai.system' does not exist in the registry. (4 occurrence(s) on span 'unknown')
• [genai_operation_name_unknown] Span 'unknown' has gen_ai.operation.name='unknown' which is not a documented enum value (4 occurrence(s) on span 'unknown')
• [genai_span_status_ok_set_by_instrumentation] Span 'unknown' has status.code='ok'; instrumentations must leave status UNSET on success (OK is reserved for application code). (4 occurrence(s) on span 'unknown')
• [genai_content_schema] Attribute 'gen_ai.output.messages' value does not conform to the GenAI schema: "finish_reason" is a required property (3 occurrence(s) on span 'chat gpt-4o-mini-2024-07-18')
• [missing_attribute] Attribute 'gen_ai.system' does not exist in the registry. (3 occurrence(s) on span 'chat gpt-4o-mini-2024-07-18')
• [genai_expected_attribute_missing] Span 'chat gpt-4o-mini-2024-07-18' (operation 'chat') is missing expected attribute 'gen_ai.response.finish_reasons' (3 occurrence(s) on span 'chat gpt-4o-mini-2024-07-18')
• [genai_span_status_ok_set_by_instrumentation] Span 'chat gpt-4o-mini-2024-07-18' has status.code='ok'; instrumentations must leave status UNSET on success (OK is reserved for application code). (3 occurrence(s) on span 'chat gpt-4o-mini-2024-07-18')
• [missing_attribute] Attribute 'gen_ai.handoff.from_agent' does not exist in the registry. (1 occurrence(s) on span 'agent_handoff OpenAI Agent')
• [missing_attribute] Attribute 'gen_ai.handoff.to_agent' does not exist in the registry. (1 occurrence(s) on span 'agent_handoff OpenAI Agent')
• [genai_content_schema] Attribute 'gen_ai.output.messages' value does not conform to the GenAI schema: "finish_reason" is a required property (1 occurrence(s) on span 'invoke_agent triage')
• [genai_content_schema] Attribute 'gen_ai.output.messages' value does not conform to the GenAI schema: "finish_reason" is a required property (1 occurrence(s) on span 'invoke_agent weather_specialist')
• [missing_attribute] Attribute 'gen_ai.system' does not exist in the registry. (1 occurrence(s) on span 'agent_handoff OpenAI Agent')
• [missing_attribute] Attribute 'gen_ai.system' does not exist in the registry. (1 occurrence(s) on span 'invoke_agent triage')
• [missing_attribute] Attribute 'gen_ai.system' does not exist in the registry. (1 occurrence(s) on span 'execute_tool get_weather')
• [missing_attribute] Attribute 'gen_ai.system' does not exist in the registry. (1 occurrence(s) on span 'invoke_agent weather_specialist')
• [missing_attribute] Attribute 'gen_ai.system' does not exist in the registry. (1 occurrence(s) on span 'Agent workflow')
• [genai_span_status_ok_set_by_instrumentation] Span 'Agent workflow' has status.code='ok'; instrumentations must leave status UNSET on success (OK is reserved for application code). (1 occurrence(s) on span 'Agent workflow')
• [genai_operation_name_unknown] Span 'agent_handoff OpenAI Agent' has gen_ai.operation.name='agent_handoff' which is not a documented enum value (1 occurrence(s) on span 'agent_handoff OpenAI Agent')
• [genai_span_status_ok_set_by_instrumentation] Span 'agent_handoff OpenAI Agent' has status.code='ok'; instrumentations must leave status UNSET on success (OK is reserved for application code). (1 occurrence(s) on span 'agent_handoff OpenAI Agent')
• [genai_expected_attribute_missing] Span 'execute_tool get_weather' (operation 'execute_tool') is missing expected attribute 'gen_ai.tool.call.id' (1 occurrence(s) on span 'execute_tool get_weather')
• [genai_span_status_ok_set_by_instrumentation] Span 'execute_tool get_weather' has status.code='ok'; instrumentations must leave status UNSET on success (OK is reserved for application code). (1 occurrence(s) on span 'execute_tool get_weather')
• [genai_span_status_ok_set_by_instrumentation] Span 'invoke_agent triage' has status.code='ok'; instrumentations must leave status UNSET on success (OK is reserved for application code). (1 occurrence(s) on span 'invoke_agent triage')
• [genai_span_status_ok_set_by_instrumentation] Span 'invoke_agent weather_specialist' has status.code='ok'; instrumentations must leave status UNSET on success (OK is reserved for application code). (1 occurrence(s) on span 'invoke_agent weather_specialist')
• [genai_span_name_format] invoke_agent span name should be 'invoke_agent' or 'invoke_agent ', got 'Agent workflow' (1 occurrence(s) on span 'Agent workflow')

.venv/lib/python3.13/site-packages/opentelemetry/test/weaver_live_check.py:441: LiveCheckError

---

Also no metrics are emitted.
The lib needs to switch to genai utils.

[lmolkova]

a trace from the instrumentation

[alfozan]

what agents-sdk version was used?

[alfozan]

I updated PR #49 to cover the concrete conformance failures listed here:

• removes invalid gen_ai.system
• avoids gen_ai.operation.name=unknown
• leaves successful spans with unset status instead of setting OK
• adds required output-message finish_reason and gen_ai.response.finish_reasons fallbacks
• removes undocumented handoff operation/attributes
• emits gen_ai.tool.call.id for tool spans, with a span-id fallback when the SDK does not provide a call ID
• stops labeling the root workflow span as invoke_agent
• enables metrics using the shared GenAI duration/token histogram helpers and configured meter_provider

I am not marking this as closing #86 yet because the PR does not add the live weaver scenario for OpenAI Agents, and it is not a full migration to the OpenAI utils path. I think those can be follow-up unless maintainers want them included in this PR.

[dhruv-ahuja]

Sharing here in case it is not covered as part of the semconv violation list, but the conversation retrieval function is also marked unknown.

using:

• opentelemetry-instrumentation-openai-agents-v2==0.1.0
• openai-agents==0.17.3