fix(openai-agents): apply content tracing flag to content

Author: duanyutong

README.md

@@ -180,6 +180,16 @@ Not sure where to get started? You can:
 - [Book a free pairing session with one of our teammates](mailto:[email protected]?subject=Pairing%20session&body=I'd%20like%20to%20do%20a%20pairing%20session!)!
 - Join our <a href="https://traceloop.com/slack">Slack</a>, and ask us any questions there.
 
+Before running tests locally, the following needs to be run.
+
+```shell
+npm ci
+npx nx affected -t install --with dev,test
+# a specific package can be installed and tested individually
+npx nx run opentelemetry-instrumentation-openai:install --with dev,test
+npx nx run opentelemetry-instrumentation-openai:test
+```
+
 ## 💚 Community & Support
 
 - [Slack](https://traceloop.com/slack) (For live discussion with the community and the Traceloop team)

packages/opentelemetry-instrumentation-openai-agents/opentelemetry/instrumentation/openai_agents/_hooks.py

@@ -1,6 +1,6 @@
 """Hook-based instrumentation for OpenAI Agents using the SDK's native callback system."""
 
-from typing import Dict, Any
+from typing import Dict, Any, Final
 import json
 import time
 from collections import OrderedDict
@@ -11,7 +11,7 @@
     gen_ai_attributes as GenAIAttributes
 )
 from agents.tracing.processors import TracingProcessor
-from .utils import dont_throw
+from .utils import dont_throw, should_send_prompts
 
 from traceloop.sdk.tracing import set_agent_name
 
@@ -233,6 +233,7 @@ def on_span_end(self, span):
         if span in self._otel_spans:
             otel_span = self._otel_spans[span]
             span_data = getattr(span, 'span_data', None)
+            should_trace_content: Final[bool] = should_send_prompts()
             if span_data and (
                 type(span_data).__name__ == 'ResponseSpanData' or isinstance(
                     span_data,
@@ -298,7 +299,7 @@ def on_span_end(self, span):
                             otel_span.set_attribute(f"{prefix}.role", role)
 
                         # Set content attribute
-                        if content is not None:
+                        if should_trace_content and content is not None:
                             if not isinstance(content, str):
                                 content = json.dumps(content)
                             otel_span.set_attribute(f"{prefix}.content", content)
@@ -399,7 +400,7 @@ def on_span_end(self, span):
                     if hasattr(response, 'output') and response.output:
                         for i, output in enumerate(response.output):
                             # Handle different output types
-                            if hasattr(output, 'content') and output.content:
+                            if should_trace_content and hasattr(output, 'content') and output.content:
                                 # Text message with content array (ResponseOutputMessage)
                                 content_text = ""
                                 for content_item in output.content:
@@ -430,7 +431,7 @@ def on_span_end(self, span):
                                 otel_span.set_attribute(
                                     f"{GenAIAttributes.GEN_AI_COMPLETION}.{i}.tool_calls.0.id", tool_call_id)
 
-                            elif hasattr(output, 'text'):
+                            elif should_trace_content and hasattr(output, 'text'):
                                 # Direct text content
                                 otel_span.set_attribute(f"{GenAIAttributes.GEN_AI_COMPLETION}.{i}.content", output.text)
                                 otel_span.set_attribute(
@@ -466,7 +467,7 @@ def on_span_end(self, span):
             elif span_data:
                 # Extract prompt data from input and add to response span (legacy support)
                 input_data = getattr(span_data, 'input', [])
-                if input_data:
+                if should_trace_content and input_data:
                     for i, message in enumerate(input_data):
                         if hasattr(message, 'role') and hasattr(message, 'content'):
                             otel_span.set_attribute(f"gen_ai.prompt.{i}.role", message.role)
@@ -508,7 +509,7 @@ def on_span_end(self, span):
                     if hasattr(response, 'output') and response.output:
                         for i, output in enumerate(response.output):
                             # Handle different output types
-                            if hasattr(output, 'content') and output.content:
+                            if should_trace_content and hasattr(output, 'content') and output.content:
                                 # Text message with content array (ResponseOutputMessage)
                                 content_text = ""
                                 for content_item in output.content:
@@ -539,7 +540,7 @@ def on_span_end(self, span):
                                 otel_span.set_attribute(
                                     f"{GenAIAttributes.GEN_AI_COMPLETION}.{i}.tool_calls.0.id", tool_call_id)
 
-                            elif hasattr(output, 'text'):
+                            elif should_trace_content and hasattr(output, 'text'):
                                 # Direct text content
                                 otel_span.set_attribute(f"{GenAIAttributes.GEN_AI_COMPLETION}.{i}.content", output.text)
                                 otel_span.set_attribute(

packages/opentelemetry-instrumentation-openai-agents/opentelemetry/instrumentation/openai_agents/utils.py

@@ -6,6 +6,8 @@
 import traceback
 from opentelemetry import context as context_api
 
+_TRACELOOP_TRACE_CONTENT = "TRACELOOP_TRACE_CONTENT"
+
 
 def set_span_attribute(span, name, value):
     if value is not None:
@@ -18,10 +20,14 @@ def _is_truthy(value):
     return str(value).strip().lower() in ("true", "1", "yes", "on")
 
 
-def should_send_prompts():
-    env_setting = os.getenv("TRACELOOP_TRACE_CONTENT", "true")
+def should_send_prompts() -> bool:
+    """Determine if LLM content tracing should be enabled.
+
+    Content includes not only prompts, but also responses.
+    """
+    env_setting = os.getenv(_TRACELOOP_TRACE_CONTENT, "true")
     override = context_api.get_value("override_enable_content_tracing")
-    return _is_truthy(env_setting) or bool(override)
+    return _is_truthy(env_setting) or _is_truthy(override)
 
 
 class JSONEncoder(json.JSONEncoder):

packages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/utils.py

@@ -174,10 +174,18 @@ def run_async(method):
         asyncio.run(method)
 
 
-def should_send_prompts():
-    return (
-        os.getenv(TRACELOOP_TRACE_CONTENT) or "true"
-    ).lower() == "true" or context_api.get_value("override_enable_content_tracing")
+def _is_truthy(value):
+    return str(value).strip().lower() in ("true", "1", "yes", "on")
+
+
+def should_send_prompts() -> bool:
+    """Determine if LLM content tracing should be enabled.
+
+    Content includes not only prompts, but also responses.
+    """
+    env_setting = os.getenv(TRACELOOP_TRACE_CONTENT, "true")
+    override = context_api.get_value("override_enable_content_tracing")
+    return _is_truthy(env_setting) or _is_truthy(override)
 
 
 def should_emit_events() -> bool: