feat(openai-agents): initial instrumentation; collect agent traces

[adharshctr]

• I have added tests that cover my changes.
• If adding a new instrumentation or changing an existing one, I've added screenshots from some observability platform showing the change.
• PR name follows conventional commits format: feat(instrumentation): ... or fix(instrumentation): ....
• (If applicable) I have updated the documentation accordingly.

---

Important

Introduces OpenTelemetry instrumentation for OpenAI Agents, enabling tracing of agent workflows with new instrumentor, tests, and SDK updates.

• Instrumentation:
  • Adds OpenAIAgentsInstrumentor in instrumentation.py to trace OpenAI agent workflows.
  • Implements _wrap_agent_run() to wrap Runner._get_new_response for tracing.
  • Adds functions in patch.py to extract and set span attributes for agent details, model settings, run config, prompts, responses, and token usage.
• SDK Integration:
  • Updates instruments.py to include OPENAI_AGENTS.
  • Modifies tracing.py to initialize OpenAIAgentsInstrumentor in init_instrumentations() and init_openai_agents_instrumentor().
• Configuration:
  • Adds .flake8 and .python-version for code style and Python version management.
  • Adds pyproject.toml and poetry.toml for project configuration and dependencies.
• Documentation:
  • Creates README.md with installation and usage instructions for the new instrumentation.
• Testing:
  • Adds test_openai_agents.py with unit tests for the instrumentor using pytest and unittest.mock.
• Sample Application:
  • Adds openai_agents_using_litellm.py as a sample app demonstrating the use of the new instrumentation.

^{This description was created by for 83787d9. You can customize this summary. It will automatically update as commits are pushed.}

[ellipsis-dev]

Caution

Changes requested ❌

Reviewed everything up to 83787d9 in 2 minutes and 1 seconds. Click for details.

• Reviewed 612 lines of code in 14 files
• Skipped 1 files when reviewing.
• Skipped posting 1 draft comments. View those below.
• Modify your settings and rules to customize what types of comments Ellipsis leaves. And don't forget to react with 👍 or 👎 to teach Ellipsis.

1. packages/opentelemetry-instrumentation-openai_agents/pyproject.toml:14

• Draft comment:
  Typo: The repository URL contains "openllmetry". It appears you meant "opentelemetry". Please correct the typo in the repository URL.
• Reason this comment was not posted:
  Decided after close inspection that this draft comment was likely wrong and/or not actionable: usefulness confidence = 10% vs. threshold = 50% While "openllmetry" might look like a typo of "opentelemetry", it could actually be an intentional name for the repository. The repository URL is part of the project metadata and without being able to verify the actual repository structure or organization naming, we can't be certain this is a mistake. This falls into the category of needing more context to be sure. The repository name could be an intentional branding choice combining "LLM" and "telemetry". Without checking the actual GitHub organization, we can't verify if this is really a mistake. Given that we can't verify the correct repository name and it could be intentional, we should err on the side of not making assumptions about organization/repository naming. Delete the comment as we don't have strong evidence that the repository URL is incorrect, and it could be an intentional naming choice.

Workflow ID: wflow_R5yp44NTeSZlDZ1d

^{You can customize by changing your verbosity settings, reacting with 👍 or 👎, replying to comments, or adding code review rules.}

[nirga]

Thanks @adharshctr! I wonder why you decided to wrap an internal method of the framework and not one of their main APIs?

[nirga]

Thanks @adharshctr - there are 2 major isseues here we need to resolve. I suggest you look at other instrumentations (like the Anthropic one) to get a sense of what is needed here.

1. You're not following semantic conventions - especially around LLM prompts and completions.
2. You should test the spans exist and contain the right set of attributes - again, see other instrumentations for examples.

[nirga]

@adharshctr can you make sure you follow semantic conventions as much as possible?
https://opentelemetry.io/docs/specs/semconv/gen-ai/gen-ai-agent-spans/

[nirga]

@adharshctr can you write proper recorded tests like we have in other instrumentations?

[adharshctr]

Done