fix(polish): 过滤模型深度思考输出

[H-Chris233]

摘要

Fixes #25。

本 PR 解决 MiniMax 等带深度思考输出的模型会把 reasoning / thinking 内容一并插入到最终文本里的问题。

部分 OpenAI-compatible 思考模型会在最终回答前返回显式的 <think>...</think> 内容。此前 polish 流程会把这部分内容当作普通文本继续处理，导致用户输入框中被插入很长的思考块，影响实际输入体验。

本次改动在 response parsing 之后、最终插入文本之前，增加保守的清洗逻辑：只移除明确的 <think>...</think> 块，保留最终润色文本。

修复 / 新增 / 改进

• 在 clean_polish_output 中增加 thinking block 清洗步骤。

• 新增 strip_thinking_blocks：

  • 只处理显式 <think>...</think> 块
  • 支持移除模型返回的 reasoning 内容
  • 保留 thinking block 后面的最终文本

• 保持现有 provider 请求逻辑不变。

• 保持现有 UI 行为不变。

• 保持现有 fallback 行为不变。

• 新增测试覆盖：

  • 验证 <think>...</think> 内容会被移除
  • 验证最终润色文本会被保留

兼容

• 不包含：

  • 不关闭模型的 thinking 能力。
  • 不新增 provider-specific 开关。
  • 不修改模型配置 UI。
  • 不修改请求参数。
  • 不按中文标题、自然语言段落或非稳定文本模式做激进清洗。

• 对现有用户 / 本地环境 / 构建流程的影响：

  • 对使用 MiniMax 等思考模型的用户，插入文本中不再包含显式 <think>...</think> 思考块。
  • 对不返回 thinking block 的模型无影响。
  • 对现有 provider 配置、fallback 逻辑和 UI 使用方式无影响。
  • 构建流程无变化。

测试计划

• 命令：cargo test polish::tests -- --nocapture

• 结果：通过

• 证据路径：本地测试输出

• 命令：cargo check

• 结果：通过

• 证据路径：本地检查输出

主要改动文件

• openless-all/app/src-tauri/src/polish.rs

备注

本 PR 采用保守清洗策略，只移除有稳定标签边界的 <think>...</think> 内容。没有选择按“深度思考”“思考过程”等本地化标题文本清洗，避免误删正常用户可见内容。

Summary by Sourcery

Filter out model reasoning blocks from polish output so only the final user-facing text is inserted.

Bug Fixes:

• Remove explicit <think>...</think> reasoning sections from polish results to prevent deep-thinking model internals from being inserted into the user input field.

Tests:

• Add unit tests covering stripping of simple, attributed, mixed-case, multiple, and malformed <think> blocks while preserving the final polished text.

[sourcery-ai]

Reviewer's Guide

Adds conservative stripping of <think>...</think> reasoning blocks from polish outputs so only the final user-facing text is kept, plus tests covering these cases.

Flow diagram for clean_polish_output with strip_thinking_blocks

flowchart TD
  A[Start_clean_polish_output_with_model_content] --> B[Call_strip_thinking_blocks]
  B --> C{Any_think_blocks_found?}
  C -- Yes --> D[Build_output_without_think_blocks]
  D --> E[Return_cleaned_text_from_strip_thinking_blocks]
  C -- No --> F[Return_original_text_as_borrowed_Cow]
  E --> G[Trim_whitespace]
  F --> G[Trim_whitespace]
  G --> H[Strip_markdown_fence_if_present]
  H --> I[Remove_leading_trailing_boilerplate_sentences]
  I --> J[Return_final_polish_output_without_reasoning_blocks]
  J --> K[End]

Loading

File-Level Changes

Change	Details	Files
Strip <think>...</think> reasoning blocks from polish output before further cleanup so only final user text is inserted.	Call a new strip_thinking_blocks helper at the start of clean_polish_output before trimming and markdown fence stripping. Return a borrowed Cow when no <think> tags are found to avoid unnecessary allocation, and an owned string when cleaning is applied. Ensure existing polish post-processing (trimming, markdown fence stripping, boilerplate removal) runs on the cleaned text.	openless-all/app/src-tauri/src/polish.rs
Implement robust parsing for <think> open/close tags, supporting attributes, ASCII case-insensitivity, and multiple blocks.	Scan the text for <think> open tags with find_think_open, then pair them with closing tags via find_think_close, copying only non-thinking segments into the output buffer. Use parse_think_open_at / parse_think_close_at to validate positions as actual <think> tags and skip non-matching < occurrences. Implement parse_think_tag_end to match think in a case-insensitive way, allow optional attributes on the opening tag, and require proper > termination while ignoring malformed or unclosed tags.	openless-all/app/src-tauri/src/polish.rs
Add unit tests verifying stripping behavior and safety around <think> blocks.	Test that a simple <think>...</think> block before the final sentence is removed while the final Chinese text is preserved. Test that <THINK> tags with attributes and mixed case are removed correctly, leaving only the final text. Test handling of multiple <think> blocks interleaved with visible text. Test that non-<think> tags and unclosed <think> tags are left untouched, and that strip_thinking_blocks can return a borrowed Cow for unchanged input.	openless-all/app/src-tauri/src/polish.rs

Assessment against linked issues

Issue	Objective	Addressed	Explanation
#25	Filter out MiniMax and other models' deep-thinking / reasoning content (e.g., ... blocks) so that only the final polished answer is inserted into the input field.	✅

Possibly linked issues

• [macos] MiniMax 模型会把深度思考也一并输入进去 #25: PR 在 clean_polish_output 中移除 <think>...</think> 块，避免 MiniMax 深度思考内容插入输入框

---

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 - I've left some high level feedback:

• In strip_thinking_blocks, consider operating on a mutable String passed in (or using a single pass over &str with indices) to avoid repeated to_string allocations and potential O(n²) behavior when many <think> blocks are present.
• If you expect models to sometimes include attributes or casing variants on the thinking tag (e.g. <think reason="true"> or <THINK>), you may want to extend strip_thinking_blocks to handle those forms as well while still keeping the conservative behavior.

Prompt for AI Agents

Please address the comments from this code review:

## Overall Comments
- In `strip_thinking_blocks`, consider operating on a mutable `String` passed in (or using a single pass over `&str` with indices) to avoid repeated `to_string` allocations and potential O(n²) behavior when many `<think>` blocks are present.
- If you expect models to sometimes include attributes or casing variants on the thinking tag (e.g. `<think reason="true">` or `<THINK>`), you may want to extend `strip_thinking_blocks` to handle those forms as well while still keeping the conservative behavior.

---

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.}

[H-Chris233]

@sourcery-ai review

[sourcery-ai]

Hey - I've reviewed your changes and they look great!

---

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.}