OpenAI Agents root span uses SERVER instead of INTERNAL

[mznet]

Describe your environment

OS: Tahoe 26.3
Python version: 3.13.9
Package version: 1.0b0.dev
GenAI library (e.g. anthropic, openai) and version: openai-agents 0.17.6

What happened?

OpenAI Agents Instrumentation records SpanKind as SERVER in root span. This logic is called in on_trace_start method in GenAISemanticProcessor and its comment says "# Root span is typically server", but I couldn't find any grounds in semantic-conventions-gen-ai and it seems to be incorrect given the definition of SERVER.

Steps to Reproduce

from agents.tracing import trace, set_trace_processors
from opentelemetry.instrumentation.openai_agents import OpenAIAgentsInstrumentor
from opentelemetry.sdk.trace import TracerProvider
from opentelemetry.sdk.trace.export import SimpleSpanProcessor, ConsoleSpanExporter

set_trace_processors([])
provider = TracerProvider()
provider.add_span_processor(SimpleSpanProcessor(ConsoleSpanExporter()))

OpenAIAgentsInstrumentor().instrument(tracer_provider=provider)

with trace("workflow"):
    pass

Console Exporter result is as shown below:

{
    "name": "workflow",
    "context": {
        "trace_id": "0xc2eeeb857a4fd2ef43732413e6b024c9",
        "span_id": "0xccf03ff8895dd4fb",
        "trace_state": "[]"
    },
    "kind": "SpanKind.SERVER",
    "parent_id": null,
    "start_time": "2026-06-19T07:15:07.896064Z",
    "end_time": "2026-06-19T07:15:07.896073Z",
    "status": {
        "status_code": "OK"
    },
    "attributes": {
        "gen_ai.provider.name": "openai",
        "gen_ai.system": "openai",
        "gen_ai.operation.name": "invoke_agent",
        "server.address": "api.openai.com",
        "server.port": 443
    },
    "events": [],
    "links": [],
    "resource": {
        "attributes": {
            "telemetry.sdk.language": "python",
            "telemetry.sdk.name": "opentelemetry",
            "telemetry.sdk.version": "1.42.1",
            "service.name": "unknown_service"
        },
        "schema_url": ""
    }
}

Expected Result

SpanKind should be INTERNAL.

Actual Result

The instrumentation records SpanKind as SERVER.

Additional context

The OpenAI Agents Instrumentation in OpenLLMetry records SpanKind as INTERNAL when it creates a root span.
https://github.com/traceloop/openllmetry/blob/main/packages/opentelemetry-instrumentation-openai-
agents/opentelemetry/instrumentation/openai_agents/_hooks.py#L662

Would you like to implement a fix?

Yes

Tip

_{React with 👍 to help prioritize this issue. Please use comments to provide useful context, avoiding +1 or me too, to help us triage it. Learn more here.}

[bhumikadangayach]

I'd like to take this on. Checked span_processor.py:1315 — the root span is hardcoded to SpanKind. SERVER in on_trace_start, and I don't see anything in the gen-ai-agent-spans semconv that supports SERVER here. Matches what you found with OpenLLmetry using internal.

Went through the existing tests, and nothing asserts on SpanKind. SERVER for this span, so it should be a clean one-line fix to INTERNAL. I'll add a regression test for it.