fix: prevent empty text content blocks in image agentFix/empty text content blocks

[Yi-Lyu]

🐛 Problem

Fixed a critical bug in the image agent that could cause messages: text content blocks must be non-empty API errors.

Root Cause

In src/agents/image.agent.ts:225, when processing text messages containing [Image #N] markers, the code would completely remove these markers using replace(/\[Image #\d+\]/g, ''). If the original text contained only these markers, the result would be an empty string, violating Anthropic API validation requirements.

Solution

• Added safety check to ensure cleaned text is not empty
• Provide fallback text [Image reference removed] when cleaned text becomes empty
• Added .trim() to handle whitespace-only cases
• Added explanatory comment for future maintainers

Changes

• File: src/agents/image.agent.ts
• Lines: 224-227
• Type: Bug fix (non-breaking)

Testing

• ✅ Code builds successfully
• ✅ No linting errors
• ✅ Maintains existing functionality while preventing empty text blocks

Impact

• Prevents API errors when processing image-related messages
• Improves reliability of the image agent
• Maintains backward compatibility

Fixes the issue reported in logs where claude-jp,claude-sonnet-4-5-20250929 provider returned 400: messages: text content blocks must be non-empty error.

[Yi-Lyu]

主要解决这个报错问题

定位了一下，主要是用户用 cc 时候直接贴了张图让 cc 分析，当前请求没问题，但是后续再碰到 [Image #数字] 的时候就会被清空，但是有的 claude 模型渠道对于 user 和 ai 的消息内容要求严格必须有内容，这就出问题了。ai 分析如下图：

请伟大的作者审核代码并修复，谢谢。 祝 ccr 越来越好。