[Feature] 请求增加AI 翻译和AI总结的手动入口 || [Feature] Request to add manual entry for AI translation and AI summary

[qijin-3]

Problem Statement

期望可以在每篇文章中增加AI 翻译和AI总结的按钮
并且允许用户添加自己的api和提示词
如果可以手动修改文章顶部操作区的按钮布局就更好了

Proposed Solution

如上

Use Cases

无

Alternative Solutions

No response

Implementation Ideas

No response

Additional Context

No response

---

Problem Statement

It is hoped that buttons for AI translation and AI summary can be added to each article.
And allows users to add their own api and prompt words
It would be better if you could manually modify the button layout in the operation area at the top of the article

Proposed Solution

As above

Use Cases

None

Alternative Solutions

No response

Implementation Ideas

No response

Additional Context

No response

[github-actions]

This issue is stale because it has been open 30 days with no activity. Remove stale label or comment or this will be closed in 5 days.

[github-actions]

This issue was closed because it has been stalled for 5 days with no activity.

[NekoAria]

感谢你的功能建议。我们实际认真评估并尝试实现了这个功能，但最终决定不合并这部分改动，原因如下。

安全性问题

AI 翻译需要在“富文本 HTML → 纯文本 → 翻译 → 重建 HTML”的链路上确保安全。我们尝试了“文本段提取 + JSON 回填”的方案来隔离 LLM 与 HTML 结构，但这并不能从根本上消除提示注入（prompt injection）风险——恶意文章的正文内容可以通过构造特殊文本诱导 LLM 输出含有 HTML/JS 的字符串，进而在回填时引入 XSS 攻击面。

AI 总结虽然相对简单，但同样存在类似的渲染安全隐患。

要彻底解决这个问题，需要为所有 AI 输出单独设计一套隔离的渲染沙箱，工程量和风险都超出了我们的预期。

维护成本

此次改动涉及 20 余个文件，包括引入 OpenAI 兼容客户端依赖、新增 AI 状态管理、流式渲染逻辑、多语言词条等。此外，在浏览器端直接使用 API Key 本身也是一个需要向用户明确说明的安全 trade-off。综合来看，这些改动的长期维护成本与 ReactFlux 目前的定位不相符。

---

综上，我们不会将此功能纳入后续规划。如果你有 AI 辅助阅读的需求，推荐搭配浏览器扩展（如 https://github.com/josStorer/chatGPTBox ）使用，可以在不改动阅读器本体的前提下实现类似效果。

---

Thanks for your feature suggestions. We actually carefully evaluated and tried to implement this feature, but ultimately decided not to merge this part of the change for the following reasons.

Security issues

AI translation needs to ensure security on the link of "rich text HTML → plain text → translation → reconstruct HTML". We tried the "text segment extraction + JSON backfilling" solution to isolate LLM and HTML structures, but this cannot fundamentally eliminate the risk of prompt injection - the body content of malicious articles can induce LLM to output a string containing HTML/JS by constructing special text, thereby introducing an XSS attack surface during backfilling.

Although AI summary is relatively simple, it also has similar rendering security risks.

To completely solve this problem, we need to design a separate set of isolated rendering sandboxes for all AI outputs. The amount of work and risks exceeded our expectations.

Maintenance cost

This change involves more than 20 files, including the introduction of OpenAI compatible client dependencies, new AI state management, streaming rendering logic, multi-language entries, etc. In addition, using API Key directly on the browser side is itself a security trade-off that needs to be clearly explained to users. Taken together, the long-term maintenance costs of these changes are inconsistent with ReactFlux's current positioning.

---

In summary, we will not include this feature in future plans. If you have AI-assisted reading needs, it is recommended to use it with a browser extension (such as https://github.com/josStorer/chatGPTBox), which can achieve similar effects without changing the reader itself.