Problem

When an AssistantAgent makes multiple tool calls in response to a task, the OpenAI API returns an UnprocessableEntityError(422) with the error message:

'detail': [{'type': 'missing', 'loc': ['body', 'messages', 2, 'content'], 'msg': 'Field required'}]

This occurs because the transformed assistant message contains tool_calls but is missing the required content field.

Root Cause

The issue was in the message transformation pipeline in _message_transform.py. When an AssistantMessage has tool calls but no thought, it uses the tools_assistant_transformer_funcs which only included:

• _set_tool_calls (adds the tool_calls field)
• Base transformers (adds role, name validation)

But it was missing a content field setter, resulting in messages like:

{
  "role": "assistant",
  "tool_calls": [...]
  // Missing required "content" field
}

According to OpenAI's API specification, assistant messages with tool calls must include a content field (which can be null).

Solution

Added a new transformer function _set_null_content_for_tool_calls that sets content: null for tool calls without thought, and updated tools_assistant_transformer_funcs to include it.

Before:

tools_assistant_transformer_funcs = base_assistant_transformer_funcs + [_set_tool_calls]
# Result: {"role": "assistant", "tool_calls": [...]}  # Missing content

After:

tools_assistant_transformer_funcs = base_assistant_transformer_funcs + [
    _set_tool_calls,
    _set_null_content_for_tool_calls,
]
# Result: {"role": "assistant", "tool_calls": [...], "content": null}  # Complete

Testing

Added comprehensive test coverage for both scenarios:

• AssistantMessage with tool calls but no thought → content: null
• AssistantMessage with tool calls and thought → content: thought_text

Impact

• Fixes the exact issue: No more OpenAI UnprocessableEntityError(422) for multiple tool calls
• Minimal and surgical: Only 7 lines added to core code, no existing functionality modified
• Backward compatible: All existing scenarios continue to work unchanged
• API compliant: Follows OpenAI specification requirements exactly

Fixes #6788.

💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.