Fix: Add missing support for 'assistant' Role in Converter.items_to_messages used by Runner.run_sync

[mjunaidca]

Why are these changes needed?

The Agents SDK currently rejects messages with the 'assistant' role when processing conversation histories. Specifically, the _Converter.items_to_messages method raises a UserError with the message "Unexpected role in easy_input_message: assistant". This affects both direct usage of the converter and higher-level APIs like Runner.run_sync.

This is a critical issue because the 'assistant' role is a standard part of the OpenAI Chat Completions API and must be supported for building conversational agents that maintain context across multiple turns. Currently, any attempt to process a conversation history containing assistant responses fails, breaking functionality for many common use cases.

[Reproducible Example Code - Failing Case]

Minimal Reproducible Example 1

from agents.models.openai_chatcompletions import _Converter

messages = _Converter.items_to_messages(
    [
        {"role": "user", "content": "Hello"},
        {"role": "assistant", "content": "Hello?"},
        {"role": "user", "content": "What was my Name?"},
    ]
)

# Results in: agents.exceptions.UserError: Unexpected role in easy_input_message: assistant

Minimal Reproducible Example 2

from agents import Agent, Runner, AsyncOpenAI, OpenAIChatCompletionsModel


external_client = AsyncOpenAI()

model = OpenAIChatCompletionsModel(
    model="gpt-4o",
    openai_client=external_client
)

agent: Agent = Agent(name="Assistant", instructions="You are a helpful assistant", model=model)

history = [{"role": "user", "content": "I am Junaid"}, {"role": "assistant", "content": "Hello?"}, {"role": "user", "content": "What was my Name?"}]

result = Runner.run_sync(agent, history)

print("\nCALLING AGENT\n")
print(result.final_output)

Changes Made:

• Updated the role validation logic in items_to_messages to accept 'assistant' as a valid role
• Ensured proper conversion of assistant messages to maintain format consistency
• Added test case for the failing case and implemented above fix.

There are no breaking changes @rm-openai . Now the method successfully processes message lists containing the 'assistant' role, enabling proper handling of multi-turn conversations with context preservation.

• Testing:
  Unit tests have been added to verify correct handling of various message formats:
  • Messages with only user roles (existing functionality)
  • Messages with mixed user and assistant roles (new functionality)
  • Messages with invalid roles (error case)

Related issue number

#92 #64

Checks

• No breaking changes
• I've added tests corresponding to the changes introduced in this PR.
• I've made sure all auto checks have passed.

[benjaminbascary97]

Waiting on this to be merged! Hopefully

[rm-openai]

Looks like lint is failing

[mjunaidca]

Looks like lint is failing

@rm-openai Fixed the lint errors