Opentelemetry traces should include agent names and the executed Python code

[jploski]

The current OTEL instrumentation does not output the agent names, which makes understanding multi-agent traces difficult (although confusingly enough it does output managed_agents). Instead of outputting "Step 1", it would be better to have "{agent.name} Step 1", same for "CodeAgent.run" - better "{agent.name} CodeAgent.run". Based on a quick hack, it seems that this can be improved simply by replacing span_name assignments in _wrappers.py (and maybe also by adding it to _smolagent_run_attributes).

Furthermore, given that Python execution errors are the most common type of errors occurring for CodeAgents during the runs, it is important that the actual Python code (not just error messages) is captured via OTEL. For that it seems like adding a _PythonExecutorWrapper similar to _ToolCallWrapper in _wrappers.py should be sufficient. (I got it to work with my LocalPythonExecutor, but I can't test it with other types, which is why I'm not providing a direct patch/PR.)

[albertvillanova]

Thanks, @jploski.

Feel free to open an issue in the GitHub repository of OpenTelemetry:

• OpenInference smolagents Instrumentation: https://github.com/Arize-ai/openinference/tree/main/python/instrumentation/openinference-instrumentation-smolagents