bugfix: parse anthropic response content block type logic error.

[FFengIll]

AgentScope Version

1.0.10 via import agentscope; print(agentscope.__version__)

Description

This PR fixes type process error and inconsistent while parsing anthropic response which may return error result.

---

The buggy parser behave different for text block and may cause problem in many edge case,
ref to agentscope.model._anthropic_model.AnthropicChatModel._parse_anthropic_stream_completion_response

Bad case example: anthropic return blocks [text, tool_use], the parser return [text, text(empty)].

Checklist

Please check the following items before code is ready to be reviewed.

• Code has been formatted with pre-commit run --all-files command
• All tests are passing
• Docstrings are in Google style
• Related documentation has been updated (e.g. links, examples, etc.)
• Code is ready for review

[cla-assistant]

All committers have signed the CLA.

[cla-assistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[qbc2016]

Thank you very much for your PR. I have a small question about the implementation. I'm trying to understand how this modification addresses the bad case mentioned in the PR description.

From my understanding, the logic for handling TextBlock and ToolUseBlock appears to remain the same before and after the change:

For TextBlock: both versions extract the text content
For ToolUseBlock: it wouldn't enter this branch in either case
Could you help clarify the connection? I might be missing something in my analysis.

[FFengIll]

Thank you very much for your PR. I have a small question about the implementation. I'm trying to understand how this modification addresses the bad case mentioned in the PR description.非常感谢你的公关。我有一个关于实现的小问题。我正在试图理解这个修改如何解决 PR 描述中提到的糟糕情况。

From my understanding, the logic for handling TextBlock and ToolUseBlock appears to remain the same before and after the change:据我了解，TextBlock 和 ToolUseBlock 的处理逻辑在更改前后似乎保持不变：

For TextBlock: both versions extract the text content对于 TextBlock：两个版本都提取文本内容 For ToolUseBlock: it wouldn't enter this branch in either case对于 ToolUseBlock：无论哪种情况，它都不会进入这个分支 Could you help clarify the connection? I might be missing something in my analysis.你能帮我澄清一下这个联系吗？我可能忽略了什么分析。

@qbc2016 Thanks for your response.

Yes, the logic appears identical when reviewing the old version statically

However, I encountered a runtime bug:

1. Old version: use tool enters as an empty text block.
2. New version: Uses strict determination to correctly render use tool as a tool block.

please check bellow debug snapshot for old version

• look at breakpoint
• current block is tool use
• but it entered into text block process

Furthermore, using new version - type for checking ensures more consistent, reliable, and safe behavior.

[FFengIll]

With my experiment, I believe this bug is trigger by new anthropic module (agentscope use it for anthropic api).
In anthropic, pydantic and pythonic process make each block contains text attr, so we can not use hasattr(x, 'text') simply.

This is a feature of pydantic

[DavdGao]

LGTM