fix(formatter): ensure assistant messages with tool_calls always include content field

[uuuyuqi]

Description

When Qwen3 and similar models operate in thinking mode, assistant messages may contain reasoning_content + tool_calls but null content. Due to @JsonInclude(NON_NULL) on OpenAIMessage and DashScopeMessage, the null content field was entirely omitted from the JSON request body, causing strict OpenAI-compatible APIs (e.g. vLLM) to reject the request with:

Input error. Field required: input.messages.N.content

Root Cause

In both OpenAIMessageConverter and DashScopeMessageConverter, when an assistant message has tool_calls but no TextBlock, the content field was left as null on the DTO. Jackson's @JsonInclude(NON_NULL) then omitted the field entirely during serialization.

Fix

Explicitly set content to an empty string "" (instead of leaving it null) when an assistant message has tool_calls but no text content. This ensures the content field is always present in the serialized JSON.

Changed files:

• OpenAIMessageConverter.convertAssistantMessage() — set content="" when hasTextContent is false and tool calls exist
• DashScopeMessageConverter.convertToSimpleContent() — replace null content with "" for assistant messages with tool calls

Close #1268

Checklist

• Code has been formatted with mvn spotless:apply
• All tests are passing (mvn test)
• Javadoc comments are complete and follow project conventions
• Related documentation has been updated (e.g. links, examples, etc.)
• Code is ready for review

Made with Cursor

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!