feat(openai): migrate openai wrapper and add claude agent sdk integration

[Abhijeet Prasad (AbhiPrasad)]

Move OpenAI wrapper tests, cassettes, and tracing code from wrappers/ into integrations/openai/. Split OpenAI Agents SDK tracing processor into its own integrations/openai_agents/ package with a dedicated test_openai_agents nox session.

Add new Claude Agent SDK integration under integrations/ with subprocess transport-level cassette support.

The old wrappers/openai.py now re-exports BraintrustTracingProcessor from the new location for backward compatibility. Test paths in noxfile.py are updated to point at the new integration directories.

[starfolkai]

Drop into this review session: sfk devbox connect pr-206-braintrust-sdk-python-946c --attach

Found 3 issues:

1. setup_openai_agents() uses current_span, NOOP_SPAN, and init_logger without importing them -- calling this function will raise NameError at runtime. The sibling setup_openai() correctly imports these from braintrust.logger on line 3.

braintrust-sdk-python/py/src/braintrust/integrations/openai_agents/__init__.py

Lines 16 to 20 in 50dae47

	project_name: str | None = None,
	) -> bool:
	"""Setup Braintrust tracing for the OpenAI Agents SDK."""
	if current_span() == NOOP_SPAN:
	init_logger(project=project_name or project, api_key=api_key, project_id=project_id)

Compare with the working pattern:

braintrust-sdk-python/py/src/braintrust/integrations/openai/__init__.py

Lines 1 to 6 in 50dae47

	"""Braintrust integration for the OpenAI Python SDK and OpenAI-compatible gateways."""
	from braintrust.logger import NOOP_SPAN, current_span, init_logger
	from .integration import OpenAIIntegration
	from .patchers import wrap_openai

2. auto_instrument() has no openai_agents parameter and never sets "openai_agents" in its return dict, but the auto-test script asserts results.get("openai_agents") == True. This test will always fail. Additionally, OpenAIAgentsIntegration is not exported from integrations/__init__.py, unlike every other integration.

braintrust-sdk-python/py/src/braintrust/integrations/auto_test_scripts/test_auto_openai_agents.py

Lines 24 to 29 in 50dae47

	results = auto_instrument()
	assert results.get("openai_agents") == True
	assert _has_braintrust_processor()
	results2 = auto_instrument()
	assert results2.get("openai_agents") == True

braintrust-sdk-python/py/src/braintrust/integrations/__init__.py

Lines 1 to 26 in 50dae47

	from .adk import ADKIntegration
	from .agentscope import AgentScopeIntegration
	from .agno import AgnoIntegration
	from .anthropic import AnthropicIntegration
	from .claude_agent_sdk import ClaudeAgentSDKIntegration
	from .dspy import DSPyIntegration
	from .google_genai import GoogleGenAIIntegration
	from .langchain import LangChainIntegration
	from .litellm import LiteLLMIntegration
	from .openrouter import OpenRouterIntegration
	from .pydantic_ai import PydanticAIIntegration
	__all__ = [
	"ADKIntegration",
	"AgentScopeIntegration",
	"AgnoIntegration",
	"AnthropicIntegration",
	"ClaudeAgentSDKIntegration",
	"DSPyIntegration",
	"GoogleGenAIIntegration",
	"LiteLLMIntegration",
	"LangChainIntegration",
	"OpenRouterIntegration",
	"PydanticAIIntegration",
	]

3. Four test classes (TestPatchOpenAI, TestPatchOpenAISpans, TestPatchOpenAIAsyncSpans, TestZAICompatibleOpenAI) were dropped during the migration without being moved to the new integration test files. patch_openai() still exists in oai.py but now has no test coverage. TestZAICompatibleOpenAI was a regression guard for a bug fix (None tool call arguments from GLM-4 models) whose fix is still in the code.

braintrust-sdk-python/py/src/braintrust/oai.py

Line 1107 in 50dae47

def patch_openai() -> bool:

Generated with Claude Code

_{- If this code review was useful, please react with 👍. Otherwise, react with 👎.}