fix: redact function tool trace span errors

[Aphroq]

Summary

• redact function tool trace error details when trace_include_sensitive_data=False
• add regression coverage for function tool exception spans under sensitive-data redaction

Test plan

• uv run pytest tests/test_run_step_execution.py::test_function_tool_error_trace_respects_sensitive_data_setting -q
• bash .agents/skills/code-change-verification/scripts/run.sh

Issue number

Closes #3110

Checks

• I've added new tests (if relevant)
• I've added/updated the relevant documentation
• I've run make lint and make format
• I've made sure tests pass

[seratch]

@codex review

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. Breezy!

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[seratch]

Thanks for the fix. The direction looks right, but I think this still needs one more redaction path before we merge.

This PR covers the failure_error_function=None path where the tool exception is re-raised, but the default/handled function-tool error path can still put the raw exception message into the function span error even when RunConfig(trace_include_sensitive_data=False) is set.

I verified this on the PR head with a small probe:

• define a normal @function_tool without failure_error_function=None
• have it raise ValueError("secret-token-123")
• run with RunConfig(trace_include_sensitive_data=False)
• inspect the exported function span

The exported span still includes:

{
  "message": "Error running tool (non-fatal)",
  "data": {
    "tool_name": "error_tool",
    "error": "secret-token-123"
  }
}

The leaking path appears to be the handled-error reporter in src/agents/tool.py, which currently stores str(error) in SpanError.data["error"] without checking the active run config's trace_include_sensitive_data value.

Could you update that path as well and add a regression test for the default @function_tool error handling case, not only the failure_error_function=None case?

[Aphroq]

Thanks for the detailed catch. Agreed — I’ll add regression coverage for this path and update the implementation accordingly.

[Aphroq]

Updated based on the review feedback.

This now covers the default @function_tool handled-error path, so function span errors no longer include the raw exception text when RunConfig(trace_include_sensitive_data=False) is used. I also audited the related tool failure span writers and added redaction for the handled cancellation path; the other tool action paths were already using the shared redaction helper.

Added/updated regression coverage for:

• re-raised tool errors with failure_error_function=None
• default @function_tool handled errors
• handled cancellation errors

Re-ran the full local verification successfully:

bash .agents/skills/code-change-verification/scripts/run.sh

[seratch]

@codex review

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. What shall we delve into next?

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".