trace_include_sensitive_data=False does not apply to agents as tools

[panilya]

Question

I have a system with

Runner.run_streamed(
   ....
   run_config=RunConfig(tracing_disabled=False, trace_include_sensitive_data=False)
)

which, according to the documentation shouldn't add input/output of my agents-as-a-tools. Correct?

Whether we include potentially sensitive data (for example: inputs/outputs of tool calls or LLM generations) in traces. If False, we'll still create spans for these events, but the sensitive data will not be included.
Source: https://openai.github.io/openai-agents-python/ref/run#agents.run.RunConfig.trace_include_sensitive_data

I'm asking this, because now, with enabled tracing and disabled logging of sensitive data I can see the system prompt, provided input and generated output of my agent-as-tool.

Please see the screenshots:

With trace_include_sensitive_data=False I would expect to do NOT include inputs/outputs of my agents-as-a-tools.

So, my question is: is it the expected behavior?

Thanks!

[seratch]

Thanks for sharing this. Your observation is correct, and if this PR (or something similar) is merged, this issue can be resolved: #1006

[panilya]

@seratch got it! Would be cool if a fix for this can be included in the next release 😊

[OiPunk]

Likely addressed by PR #2463.\n\nRoot cause in both cases is that nested agent-as-tool runs were not inheriting the parent RunConfig unless run_config was passed explicitly to as_tool(...). That meant nested runs could fall back to default tracing behavior, including trace_include_sensitive_data.\n\nPR #2463 changes this so agent-as-tool invocations inherit parent RunConfig by default, while still honoring explicit overrides.