fix: attach thinking content to agent execution message

[xinquiry]

Summary

• Fix bug where thinking content appears as a separate message bubble below the agent response
• When agent_start consumes the loading message first, thinking_start now attaches to the agent execution message instead of creating a separate message
• Updated thinking_chunk and thinking_end handlers to also find agent messages with isThinking flag

Test plan

• Send message to an agent using a model with extended thinking (Claude with thinking enabled)
• Verify thinking content appears in "Show thinking" collapsible ON the same message bubble as the agent response
• Verify no separate empty message bubble is created
• Refresh page and verify message still displays correctly

🤖 Generated with Claude Code

Summary by Sourcery

确保「思考内容」附加到对应的代理消息上，而不是作为单独的气泡渲染。

Bug 修复：

• 当代理执行正在运行时，防止思考内容出现在单独的空助手消息中。

增强：

• 在可用的情况下，将思考状态附加到当前正在运行的代理执行消息上，仅在必要时才退回到创建专门的思考消息。
• 更新思考流式处理程序，以便在缺少原始事件 ID 的情况下，也能定位被标记为「思考中」的代理消息。

Original summary in English

Summary by Sourcery

Ensure thinking content is attached to the appropriate agent message instead of rendering as a separate bubble.

Bug Fixes:

• Prevent thinking content from appearing in a separate empty assistant message when an agent execution is running.

Enhancements:

• Attach thinking state to the current running agent execution message when available, falling back to creating a dedicated thinking message only when necessary.
• Update thinking streaming handlers to locate agent messages marked as thinking, even when they lack the original event ID.

[sourcery-ai]

Sorry @xinquiry, your pull request is larger than the review limit of 150000 diff characters

[xinquiry]

@sourcery-ai Review again

[sourcery-ai]

审阅者指南

确保对于需要扩展推理的代理，其思考内容附加在主代理执行消息（或加载消息）上，而不是创建一个单独的空助手消息气泡，并更新流式处理程序以能够正确定位并更新这些附加了思考内容的消息。

将 thinking_start 附加到代理执行或加载消息的时序图

sequenceDiagram
  participant StreamHandler
  participant Channel
  participant Messages

  StreamHandler->>Channel: receive event type thinking_start(id)
  Channel->>Messages: findIndex m.isLoading
  alt loading message found
    Channel->>Messages: attach thinking to loading message
    Messages-->>Channel: message.isThinking = true
  else no loading message
    Channel->>Messages: findLastIndex m.agentExecution.status == running
    alt running agent message found
      Channel->>Messages: attach thinking to running agent message
      Messages-->>Channel: message.isThinking = true
    else no agent execution message
      Channel->>Messages: push new assistant message
      Messages-->>Channel: new message with isThinking = true
    end
  end
  Channel-->>StreamHandler: channel.responding = true

Loading

为 thinking_chunk 和 thinking_end 定位思考消息的时序图

sequenceDiagram
  participant StreamHandler
  participant Channel
  participant Messages

  rect rgb(230,230,250)
    StreamHandler->>Channel: receive event type thinking_chunk(id, content)
    Channel->>Messages: findIndex m.id == id
    alt message found by id
      Channel->>Messages: append content to message.thinkingContent
    else not found by id
      Channel->>Messages: findLastIndex m.isThinking && m.agentExecution.status == running
      alt agent thinking message found
        Channel->>Messages: append content to message.thinkingContent
      else no thinking message
        Channel-->>StreamHandler: ignore chunk
      end
    end
  end

  rect rgb(230,255,230)
    StreamHandler->>Channel: receive event type thinking_end(id)
    Channel->>Messages: findIndex m.id == id
    alt message found by id
      Channel->>Messages: set message.isThinking = false
    else not found by id
      Channel->>Messages: findLastIndex m.isThinking && m.agentExecution
      alt agent thinking message found
        Channel->>Messages: set message.isThinking = false
      else no thinking message
        Channel-->>StreamHandler: ignore end
      end
    end
  end

Loading

文件级变更

Change	Details	Files
将 thinking_start 的处理逻辑修改为优先复用已有的加载消息或正在运行的代理执行消息，仅在找不到时才创建独立的思考消息。	在接收到 thinking_start 事件后，首先在 channel.messages 中查找 isLoading 消息，如果找到，则将其转换为支持思考的助手消息，并初始化 isThinking 和 thinkingContent。 如果不存在加载消息，则从尾部开始在 channel.messages 中查找 status === "running" 的 agentExecution 消息，并将思考状态（isThinking 和 thinkingContent）附加到该消息上。 如果既没有加载消息，也没有正在运行的代理执行消息，则作为回退创建一个新的助手消息，其中 isThinking: true 且 thinkingContent 为空字符串。	web/src/store/slices/chatSlice.ts
更新 thinking_chunk 的处理，以便在思考内容附加到代理执行消息时，能够定位到正确的消息。	在 thinking_chunk 事件中，首先尝试在 channel.messages 中通过 id 查找目标消息。 如果通过 id 未找到消息，则回退为从尾部开始查找满足 isThinking && agentExecution?.status === "running" 的消息，以支持思考内容附加到进行中的代理执行的场景。 将接收到的思考内容追加到 message.thinkingContent 字符串中，同时保留已有内容。	web/src/store/slices/chatSlice.ts
更新 thinking_end 的处理，以清除可能附加了思考内容的消息（包括代理执行消息）上的思考状态。	在 thinking_end 事件中，首先尝试在 channel.messages 中通过 id 查找目标消息。 如果通过 id 未找到消息，则回退为从尾部开始查找满足 isThinking && m.agentExecution 的消息，以处理附加到代理执行消息上的思考内容。 找到消息后，将 isThinking 设为 false，以停止为该消息渲染思考 UI。	web/src/store/slices/chatSlice.ts

---

技巧与命令

与 Sourcery 交互

• 触发新的代码审查： 在拉取请求上评论 @sourcery-ai review。
• 继续讨论： 直接回复 Sourcery 的审查评论。
• 从审查评论生成 GitHub issue： 在某条审查评论下回复，要求 Sourcery 基于该评论创建一个 issue。你也可以回复 @sourcery-ai issue 来从该评论创建一个 issue。
• 生成拉取请求标题： 在拉取请求标题的任何位置写上 @sourcery-ai，即可随时生成标题。你也可以在拉取请求上评论 @sourcery-ai title 来（重新）生成标题。
• 生成拉取请求总结： 在拉取请求正文的任意位置写上 @sourcery-ai summary，即可在该位置生成 PR 总结。你也可以在拉取请求上评论 @sourcery-ai summary 来在任意时间（重新）生成总结。
• 生成审阅者指南： 在拉取请求上评论 @sourcery-ai guide，即可随时（重新）生成审阅者指南。
• 一键解决所有 Sourcery 评论： 在拉取请求上评论 @sourcery-ai resolve，即可批量标记所有 Sourcery 评论为已解决。当你已经处理完所有评论且不再希望看到它们时非常有用。
• 清除所有 Sourcery 审查： 在拉取请求上评论 @sourcery-ai dismiss，以清除所有现有的 Sourcery 审查记录。特别适用于你想要从头开始进行新的审查时——别忘了再评论 @sourcery-ai review 来触发新一轮审查！

自定义你的使用体验

访问你的 控制面板 以：

• 启用或禁用审查功能，例如 Sourcery 自动生成的拉取请求总结、审阅者指南等。
• 更改审查语言。
• 添加、移除或编辑自定义审查指令。
• 调整其他审查设置。

获取帮助

• 通过 联系我们的支持团队 提出问题或反馈意见。
• 访问我们的 文档，获取详细指南和信息。
• 通过关注我们的 X/Twitter、LinkedIn 或 GitHub 与 Sourcery 团队保持联系。

Original review guide in English

Reviewer's Guide

Ensure thinking content for extended-thinking agents is attached to the main agent execution message (or loading message) instead of creating a separate empty assistant bubble, and update streaming handlers to correctly locate and update these thinking-attached messages.

Sequence diagram for thinking_start attaching to agent execution or loading message

sequenceDiagram
  participant StreamHandler
  participant Channel
  participant Messages

  StreamHandler->>Channel: receive event type thinking_start(id)
  Channel->>Messages: findIndex m.isLoading
  alt loading message found
    Channel->>Messages: attach thinking to loading message
    Messages-->>Channel: message.isThinking = true
  else no loading message
    Channel->>Messages: findLastIndex m.agentExecution.status == running
    alt running agent message found
      Channel->>Messages: attach thinking to running agent message
      Messages-->>Channel: message.isThinking = true
    else no agent execution message
      Channel->>Messages: push new assistant message
      Messages-->>Channel: new message with isThinking = true
    end
  end
  Channel-->>StreamHandler: channel.responding = true

Loading

Sequence diagram for thinking_chunk and thinking_end locating thinking message

sequenceDiagram
  participant StreamHandler
  participant Channel
  participant Messages

  rect rgb(230,230,250)
    StreamHandler->>Channel: receive event type thinking_chunk(id, content)
    Channel->>Messages: findIndex m.id == id
    alt message found by id
      Channel->>Messages: append content to message.thinkingContent
    else not found by id
      Channel->>Messages: findLastIndex m.isThinking && m.agentExecution.status == running
      alt agent thinking message found
        Channel->>Messages: append content to message.thinkingContent
      else no thinking message
        Channel-->>StreamHandler: ignore chunk
      end
    end
  end

  rect rgb(230,255,230)
    StreamHandler->>Channel: receive event type thinking_end(id)
    Channel->>Messages: findIndex m.id == id
    alt message found by id
      Channel->>Messages: set message.isThinking = false
    else not found by id
      Channel->>Messages: findLastIndex m.isThinking && m.agentExecution
      alt agent thinking message found
        Channel->>Messages: set message.isThinking = false
      else no thinking message
        Channel-->>StreamHandler: ignore end
      end
    end
  end

Loading

File-Level Changes

Change	Details	Files
Change thinking_start handling to prefer existing loading or running agent execution messages and only create a standalone thinking message as a fallback.	After receiving a thinking_start event, first search channel.messages for an isLoading message and, if found, convert it into a thinking-enabled assistant message with isThinking and thinkingContent initialized. If no loading message is present, search channel.messages from the end for an agentExecution message with status === "running" and attach thinking state (isThinking and thinkingContent) to that message. If neither a loading nor a running agent execution message is found, create a new assistant message with isThinking: true and empty thinkingContent as a fallback.	web/src/store/slices/chatSlice.ts
Update thinking_chunk handling to locate the correct message when thinking is attached to an agent execution message.	On thinking_chunk events, first attempt to find the target message by id in channel.messages. If no message is found by id, fall back to searching from the end for a message with isThinking && agentExecution?.status === "running" to support cases where thinking is attached to an in-progress agent execution. Append incoming thinking content to the message.thinkingContent string while preserving existing content.	web/src/store/slices/chatSlice.ts
Update thinking_end handling to clear thinking state on messages that may have thinking attached, including agent execution messages.	On thinking_end events, first try to find the target message by id in channel.messages. If no message is found by id, fall back to searching from the end for a message with isThinking && m.agentExecution to handle thinking attached to agent execution messages. When the message is found, set isThinking to false to stop rendering the thinking UI for that message.	web/src/store/slices/chatSlice.ts

---

Tips and commands

Interacting with Sourcery

• Trigger a new review: Comment @sourcery-ai review on the pull request.
• Continue discussions: Reply directly to Sourcery's review comments.
• Generate a GitHub issue from a review comment: Ask Sourcery to create an
  issue from a review comment by replying to it. You can also reply to a
  review comment with @sourcery-ai issue to create an issue from it.
• Generate a pull request title: Write @sourcery-ai anywhere in the pull
  request title to generate a title at any time. You can also comment
  @sourcery-ai title on the pull request to (re-)generate the title at any time.
• Generate a pull request summary: Write @sourcery-ai summary anywhere in
  the pull request body to generate a PR summary at any time exactly where you
  want it. You can also comment @sourcery-ai summary on the pull request to
  (re-)generate the summary at any time.
• Generate reviewer's guide: Comment @sourcery-ai guide on the pull
  request to (re-)generate the reviewer's guide at any time.
• Resolve all Sourcery comments: Comment @sourcery-ai resolve on the
  pull request to resolve all Sourcery comments. Useful if you've already
  addressed all the comments and don't want to see them anymore.
• Dismiss all Sourcery reviews: Comment @sourcery-ai dismiss on the pull
  request to dismiss all existing Sourcery reviews. Especially useful if you
  want to start fresh with a new review - don't forget to comment
  @sourcery-ai review to trigger a new review!

Customizing Your Experience

Access your dashboard to:

• Enable or disable review features such as the Sourcery-generated pull request
  summary, the reviewer's guide, and others.
• Change the review language.
• Add, remove or edit custom review instructions.
• Adjust other review settings.

Getting Help

• Contact our support team for questions or feedback.
• Visit our documentation for detailed guides and information.
• Keep in touch with the Sourcery team by following us on X/Twitter, LinkedIn or GitHub.

[sourcery-ai]

Hey - 我发现了 1 个问题，并留下了一些高层反馈：

• 在 thinking_end 中，用于回退查找 thinking 消息的条件是 m.isThinking && m.agentExecution，而 thinking_chunk 使用的是 status === 'running' 作为约束；建议对齐这两个条件（或说明它们为何不同），以避免对代理消息的定位不一致。
• thinking_chunk 和 thinking_end 现在都依赖 findLastIndex，这是一个相对较新的方法；如果这段代码会在不支持该方法的运行环境中执行，你可能需要增加保护或 polyfill 来避免运行时错误。

给 AI Agent 的提示

Please address the comments from this code review:

## Overall Comments
- 在 `thinking_end` 中，用于回退查找 thinking 消息的条件是 `m.isThinking && m.agentExecution`，而 `thinking_chunk` 使用的是 `status === 'running'` 作为约束；建议对齐这两个条件（或说明它们为何不同），以避免对代理消息的定位不一致。
- `thinking_chunk` 和 `thinking_end` 现在都依赖 `findLastIndex`，这是一个相对较新的方法；如果这段代码会在不支持该方法的运行环境中执行，你可能需要增加保护或 polyfill 来避免运行时错误。

## Individual Comments

### Comment 1
<location> `web/src/store/slices/chatSlice.ts:1343` </location>
<code_context>
+                  channel.messages[loadingIndex] = {
</code_context>

<issue_to_address>
**issue (bug_risk):** 在将 loading 消息转换为 thinking 消息时，保留现有的消息元数据。

像这样直接覆盖 `channel.messages[loadingIndex]` 会丢失该 loading 消息上已有的属性（例如 `agentExecution`、附件、元数据）。在 agent 的场景中你有对已有消息做展开，但这里没有。可以使用类似下面的写法：

```ts
channel.messages[loadingIndex] = {
  ...channel.messages[loadingIndex],
  isThinking: true,
  thinkingContent: '',
  content: '',
};
```

从而保留之前关联的所有数据。
</issue_to_address>

---

Sourcery 对开源项目是免费的——如果你觉得我们的评审有帮助，欢迎帮忙分享 ✨

• X
• Mastodon
• LinkedIn
• Facebook

_{帮我变得更有用！请在每条评论上点 👍 或 👎，我会根据这些反馈改进后续的评审。}

Original comment in English

Hey - I've found 1 issue, and left some high level feedback:

• In thinking_end, the fallback lookup for a thinking message uses m.isThinking && m.agentExecution, whereas thinking_chunk constrains to status === 'running'; consider aligning these conditions (or documenting why they differ) to avoid inconsistent targeting of agent messages.
• Both thinking_chunk and thinking_end now rely on findLastIndex, which is relatively new; if this code runs in environments without native support, you may want to guard or polyfill this to prevent runtime errors.

Prompt for AI Agents

Please address the comments from this code review:

## Overall Comments
- In `thinking_end`, the fallback lookup for a thinking message uses `m.isThinking && m.agentExecution`, whereas `thinking_chunk` constrains to `status === 'running'`; consider aligning these conditions (or documenting why they differ) to avoid inconsistent targeting of agent messages.
- Both `thinking_chunk` and `thinking_end` now rely on `findLastIndex`, which is relatively new; if this code runs in environments without native support, you may want to guard or polyfill this to prevent runtime errors.

## Individual Comments

### Comment 1
<location> `web/src/store/slices/chatSlice.ts:1343` </location>
<code_context>
+                  channel.messages[loadingIndex] = {
</code_context>

<issue_to_address>
**issue (bug_risk):** Preserve existing message metadata when converting a loading message into a thinking message.

Overwriting `channel.messages[loadingIndex]` like this drops any existing properties on the loading message (e.g., `agentExecution`, attachments, metadata). In the agent case you spread the existing message, but not here. Use something like:

```ts
channel.messages[loadingIndex] = {
  ...channel.messages[loadingIndex],
  isThinking: true,
  thinkingContent: '',
  content: '',
};
```

to preserve all previously associated data.
</issue_to_address>

---

Sourcery is free for open source - if you like our reviews please consider sharing them ✨

• X
• Mastodon
• LinkedIn
• Facebook

_{Help me be more useful! Please click 👍 or 👎 on each comment and I'll use the feedback to improve your reviews.}

[sourcery-ai]

issue (bug_risk): 在将 loading 消息转换为 thinking 消息时，保留现有的消息元数据。

像这样直接覆盖 channel.messages[loadingIndex] 会丢失该 loading 消息上已有的属性（例如 agentExecution、附件、元数据）。在 agent 的场景中你有对已有消息做展开，但这里没有。可以使用类似下面的写法：

channel.messages[loadingIndex] = {
  ...channel.messages[loadingIndex],
  isThinking: true,
  thinkingContent: '',
  content: '',
};

从而保留之前关联的所有数据。

Original comment in English

issue (bug_risk): Preserve existing message metadata when converting a loading message into a thinking message.

Overwriting channel.messages[loadingIndex] like this drops any existing properties on the loading message (e.g., agentExecution, attachments, metadata). In the agent case you spread the existing message, but not here. Use something like:

channel.messages[loadingIndex] = {
  ...channel.messages[loadingIndex],
  isThinking: true,
  thinkingContent: '',
  content: '',
};

to preserve all previously associated data.

[Mile-Away]

🎉 This PR is included in version 1.1.1 🎉

The release is available on GitHub release

Your semantic-release bot 📦🚀