[OpenAI Agents Auto-Instrument] Auto-instrumented spans have hardcoded agent identity instead of actual agent details from baggage

[fpfp100]

Summary

Spans created by GenAISemanticProcessor use hardcoded default values for gen_ai.agent.name, gen_ai.agent.id, and gen_ai.agent.description instead of the actual agent identity propagated via baggage.

Details

Observed on auto-instrumented spans (chat, invoke_agent Assistant, unknown, Agent workflow):

Attribute	Expected (from baggage)	Actual (distro)
gen_ai.agent.name	nikhilcagent0416 Agent User	OpenAI Agent
gen_ai.agent.id	d6d31512-379f-4994-b901-c098dfd9293f	agent
gen_ai.agent.description	agenticUser	OpenAI Agents instrumentation

The base A365 SDK correctly propagates the real agent identity from baggage to all spans. The distro's GenAISemanticProcessor uses constructor defaults (agent_name, agent_id, agent_description) which are set at instrumentor init time, not from the per-request baggage context.

The invoke_agent Assistant span does correctly pick up the agent name "Assistant" from the AgentSpanData.name field, but gen_ai.agent.id and gen_ai.agent.description still show hardcoded defaults.

Expected behavior

Auto-instrumented spans should reflect the actual agent identity from baggage/context, matching the base A365 SDK behavior.

[github-actions]

cc Hector Hernandez (@hectorhdzg) Jackson Weber (@JacksonWeber) Leighton Chen (@lzchen) Nev (@MSNev) Radhika Gupta (@rads-1996) Jeremy Voss (@jeremydvoss)

[fpfp100]

Distro version: microsoft-opentelemetry==0.1.0b3, opentelemetry-instrumentation-openai-agents (bundled)

[alex-pathcourse]

Your auto-instrumented spans are showing default agent values instead of the actual agent identity from baggage. PathCourse Health captures the real agent context automatically, so you don’t have to maintain custom instrumentation or propagate identity manually.

Working example: https://github.com/pathcourse-health/pch-integration-examples#3-observability--traces-spans-anomalies