Feature/OpenAI agents content capture

instrumentation-genai/opentelemetry-instrumentation-openai-agents-v2/CHANGELOG.md

@@ -12,3 +12,5 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
   ([#3805](https://github.com/open-telemetry/opentelemetry-python-contrib/pull/3805))
 - Implement OpenAI Agents span processing aligned with GenAI semantic conventions.
   ([#3817](https://github.com/open-telemetry/opentelemetry-python-contrib/pull/3817))
+- Input and output according to GenAI spec.
+  ([#3824](https://github.com/open-telemetry/opentelemetry-python-contrib/pull/3824))

instrumentation-genai/opentelemetry-instrumentation-openai-agents-v2/examples/content-capture/.env.example

@@ -0,0 +1,8 @@
+# Copy to .env and add values before running the sample.
+# Required for OpenAI client (only used if you swap in a real OpenAI call)
+OPENAI_API_KEY=
+
+# Optional overrides for span attributes / exporters
+OTEL_SERVICE_NAME=openai-agents-content-capture-demo
+OTEL_EXPORTER_OTLP_ENDPOINT=http://localhost:4317
+OTEL_EXPORTER_OTLP_PROTOCOL=grpc

instrumentation-genai/opentelemetry-instrumentation-openai-agents-v2/examples/content-capture/README.md

@@ -0,0 +1,42 @@
+# OpenAI Agents Content Capture Demo
+
+This example exercises the `OpenAIAgentsInstrumentor` with message content
+capture enabled, illustrating how prompts, responses, and tool payloads are
+recorded on spans and span events.
+
+> The demo uses the local tracing utilities from the `openai-agents`
+> package—no outbound API calls are made.
+
+## Prerequisites
+
+1. Activate the repository virtual environment:
+
+   ```bash
+   source ../../.venv/bin/activate
+   ```
+
+2. Copy `.env.example` to `.env` and provide any overrides you need (for example,
+   setting `OTEL_EXPORTER_OTLP_ENDPOINT`).
+3. Ensure `openai-agents` is installed in the environment (it is included in
+   the shared development venv for this repository).
+
+## Run the demo
+
+```bash
+python main.py
+```
+
+The script will:
+
+- Configure the OpenTelemetry SDK with an OTLP exporter so spans reach your collector.
+- Instrument the OpenAI Agents tracing hooks with content capture enabled.
+- Simulate an agent invocation that performs a generation and a tool call.
+- Print the resulting spans, attributes, and events (including JSON-encoded
+  prompts and responses) to stdout.
+
+## Customisation tips
+
+- Set `OTEL_SERVICE_NAME` before running to override the default service name.
+- Adjust the OTLP exporter configuration (endpoint, protocol) through `.env`.
+- Modify the prompts, tool payloads, or add additional spans in `run_workflow`
+  to explore different content capture scenarios.

instrumentation-genai/opentelemetry-instrumentation-openai-agents-v2/examples/content-capture/main.py

@@ -0,0 +1,122 @@
+"""
+Content capture demo for the OpenAI Agents instrumentation.
+
+This script spins up the instrumentation with message capture enabled and
+simulates an agent invocation plus a tool call using the tracing helpers from
+the ``openai-agents`` package. Spans are exported to the console so you can
+inspect captured prompts, responses, and tool payloads without making any
+OpenAI API calls.
+"""
+
+from __future__ import annotations
+
+import json
+import os
+from typing import Any
+
+from agents.tracing import agent_span, function_span, generation_span, trace
+from dotenv import load_dotenv
+
+from opentelemetry.exporter.otlp.proto.grpc.trace_exporter import (
+    OTLPSpanExporter,
+)
+from opentelemetry.instrumentation.openai_agents import (
+    OpenAIAgentsInstrumentor,
+)
+from opentelemetry.sdk.resources import Resource
+from opentelemetry.sdk.trace import TracerProvider
+from opentelemetry.sdk.trace.export import BatchSpanProcessor
+
+load_dotenv()  # take environment variables from .env.
+
+
+def configure_tracing() -> None:
+    """Configure a tracer provider that exports spans via OTLP."""
+    resource = Resource.create(
+        {
+            "service.name": os.environ.get(
+                "OTEL_SERVICE_NAME", "openai-agents-content-capture-demo"
+            )
+        }
+    )
+    provider = TracerProvider(resource=resource)
+    provider.add_span_processor(BatchSpanProcessor(OTLPSpanExporter()))
+
+    # Instrument with explicit content capture mode to ensure prompts/responses are recorded.
+    OpenAIAgentsInstrumentor().instrument(
+        tracer_provider=provider,
+        capture_message_content="span_and_event",
+        system="openai",
+        agent_name="Travel Concierge",
+        base_url="https://api.openai.com/v1",
+    )
+
+
+def dump(title: str, payload: Any) -> None:
+    """Pretty-print helper used to show intermediate context."""
+    print(f"\n=== {title} ===")
+    print(json.dumps(payload, indent=2))
+
+
+def run_workflow() -> None:
+    """Simulate an agent workflow with a generation and a tool invocation."""
+    itinerary_prompt = [
+        {"role": "system", "content": "Plan high level travel itineraries."},
+        {
+            "role": "user",
+            "content": "I'm visiting Paris for 3 days in November.",
+        },
+    ]
+
+    tool_args = {"city": "Paris", "date": "2025-11-12"}
+    tool_result = {
+        "forecast": "Mostly sunny, highs 15°C",
+        "packing_tips": ["light jacket", "comfortable shoes"],
+    }
+
+    with trace("travel-booking-workflow"):
+        with agent_span(name="travel_planner") as agent:
+            dump(
+                "Agent span started",
+                {"span_id": agent.span_id, "trace_id": agent.trace_id},
+            )
+
+            with generation_span(
+                input=itinerary_prompt,
+                output=[
+                    {
+                        "role": "assistant",
+                        "content": (
+                            "Day 1 visit the Louvre, Day 2 tour Versailles, "
+                            "Day 3 explore Montmartre."
+                        ),
+                    }
+                ],
+                model="gpt-4o-mini",
+                usage={
+                    "input_tokens": 128,
+                    "output_tokens": 96,
+                    "total_tokens": 224,
+                },
+            ):
+                pass
+
+            with function_span(
+                name="fetch_weather",
+                input=json.dumps(tool_args),
+                output=tool_result,
+            ):
+                pass
+
+    print(
+        "\nWorkflow complete – spans exported to the configured OTLP endpoint."
+    )
+
+
+def main() -> None:
+    configure_tracing()
+    run_workflow()
+
+
+if __name__ == "__main__":
+    main()

instrumentation-genai/opentelemetry-instrumentation-openai-agents-v2/examples/manual/.env.example

@@ -1,11 +1,5 @@
-# Update this with your real OpenAI API key
-OPENAI_API_KEY=sk-YOUR_API_KEY
-
-# Uncomment and adjust if you use a non-default OTLP collector endpoint
-# OTEL_EXPORTER_OTLP_ENDPOINT=http://localhost:4317
-# OTEL_EXPORTER_OTLP_PROTOCOL=grpc
-
-OTEL_SERVICE_NAME=opentelemetry-python-openai-agents-manual
-
-# Optionally override the agent name reported on spans
-# OTEL_GENAI_AGENT_NAME=Travel Concierge
+# Copy to .env and add real values before running main.py
+OPENAI_API_KEY=
+OTEL_EXPORTER_OTLP_ENDPOINT=http://localhost:4317
+OTEL_EXPORTER_OTLP_PROTOCOL=grpc
+OTEL_SERVICE_NAME=openai-agents-manual-demo

instrumentation-genai/opentelemetry-instrumentation-openai-agents-v2/examples/manual/README.rst

@@ -22,7 +22,7 @@ Setup
        python3 -m venv .venv
        source .venv/bin/activate
        pip install "python-dotenv[cli]"
-       pip install -r requirements.txt
+       uv pip install -r requirements.txt --prerelease=allow
 
 Run
 ---
@@ -34,6 +34,8 @@ are applied:
 
     dotenv run -- python main.py
 
+Ensure ``OPENAI_API_KEY`` is present in your environment (or ``.env`` file); the OpenAI client raises ``OpenAIError`` if the key is missing.
+
 The script automatically loads environment variables from ``.env`` so running
 ``python main.py`` directly also works if the shell already has the required
 values exported.

instrumentation-genai/opentelemetry-instrumentation-openai-agents-v2/examples/zero-code/README.rst

@@ -22,7 +22,7 @@ Setup
        python3 -m venv .venv
        source .venv/bin/activate
        pip install "python-dotenv[cli]"
-       pip install -r requirements.txt
+       uv pip install -r requirements.txt --prerelease=allow
 
 Run
 ---
@@ -34,6 +34,8 @@ instrumentation is activated automatically:
 
     dotenv run -- opentelemetry-instrument python main.py
 
+Ensure ``OPENAI_API_KEY`` is set in your shell or `.env`; the OpenAI client raises ``OpenAIError`` if the key is missing.
+
 Because ``main.py`` invokes ``load_dotenv``, running ``python main.py`` directly
 also works when the required environment variables are already exported.
 

instrumentation-genai/opentelemetry-instrumentation-openai-agents-v2/pyproject.toml

@@ -26,7 +26,8 @@ classifiers = [
 dependencies = [
   "opentelemetry-api >= 1.37",
   "opentelemetry-instrumentation >= 0.58b0",
-  "opentelemetry-semantic-conventions >= 0.58b0"
+  "opentelemetry-semantic-conventions >= 0.58b0",
+  "opentelemetry-util-genai"
 ]
 
 [project.optional-dependencies]