Skip to content

docs: fix mistranslations and awkward wording in README-zh - #612

Open
xqs-xqs wants to merge 1 commit into
shareAI-lab:mainfrom
xqs-xqs:docs/fix-readme-zh-translation
Open

xqs-xqs wants to merge 1 commit into
shareAI-lab:mainfrom
xqs-xqs:docs/fix-readme-zh-translation

Conversation

@xqs-xqs

@xqs-xqs xqs-xqs commented Nov 3, 2026

Copy link
Copy Markdown

Fixes: #611

Summary

While learning from the Chinese README, I found several places that are hard to follow for Chinese readers. This PR fixes them in README-zh.md (docs only, one file):

  • Mistranslations: prompt-plumbing, curate, focused work, trajectory data
  • Word-for-word sentences rewritten in natural Chinese (5 places)
  • Out of sync with README.md: "the model will do the rest", and two sentences the English README has already removed (「那个模型就是 agent」 and an SDK sentence)
  • Typo in the learning-path diagram

Changes

Original Reason Revised
「没有游戏专属规则。没有决策树。一个模型,从经验中学习。那个模型就是 agent。」 Word-for-word fragments merged into one sentence, with 「游戏专属规则」 reworded as 「针对具体游戏的规则」; the closing 「那个模型就是 agent」 is removed because the English README no longer has it 「没有针对具体游戏的规则,没有决策树,只有一个从经验中学习的模型。」
「"Agent" 这个词已经被一整个提示词水管工产业劫持了。」 Mistranslation: prompt-plumbing means piping prompts between LLM calls; 「水管工」 (plumber) turns the activity into a job title, and the metaphor does not carry over into Chinese. 「劫持」 for hijacked is also literal; 「挪用」 (appropriated) is the natural verb for a word 「"Agent" 这个词,已经被一整个靠拼接提示词吃饭的行业挪用了。」
「提示词水管工式 "Agent"」 Mistranslation: same term as above, reworded with the same 「拼接提示词」 「靠拼接提示词堆出来的 "Agent"」
「收集任务过程数据」 (2 places) Mistranslation: trajectory data is usually rendered as 「轨迹数据」, and the same paragraph already uses 「轨迹」 「收集轨迹数据」
「Agent -- 那个智能、那个决策者 -- 永远是模型。」 Word-for-word appositive (「那个……、那个……」), restructured into a natural sentence; 「永远」 ("forever") replaced with 「始终」 ("always") 「真正有智能、做决策的,始终是模型。」
「这个仓库教你造载具。编程用的载具。但设计模式可以泛化到任何领域:」 Word-for-word fragments ("…the vehicle. A vehicle for coding.") merged into one sentence; 「泛化」 is an ML term, while the meaning here is simply "apply to", so 「推广」 is used 「这个仓库教你造的,就是一辆编程用的载具。但其中的设计模式可以推广到任何领域:」
「策划知识」 Mistranslation: curate means to select and organize; 「策划」 means to plan, as in an event 「整理知识」
「子 Agent 把明确的工作留在另一份消息列表中」 Mistranslation: focused work means a focused subtask, not 「明确的工作」 ("clear-cut work"); separate is rendered as 「独立的」 instead of 「另一份」 「子 Agent 在独立的消息列表里专心处理子任务」
「造好 Harness。Agent 会完成剩下的。」 Out of sync: the English now says "The model will do the rest", as does the closing line of this file 「造好 Harness。模型会完成剩下的。」
「然后让开了。」 Word-for-word: "gets out of the way" 「然后就不再插手。」
「由 Anthropic 在人类推理和代码的全部广度上训练而成」 Word-for-word: "trained on the full breadth of human reasoning and code"; reworded to say what the training data covers 「由 Anthropic 用覆盖人类推理和代码方方面面的数据训练而成」
「长下文腾空间」 Typo in the learning-path diagram 「长上下文腾空间」

Removed

Section Removed text Reason
Kode Agent SDK 「官方 Claude Code Agent SDK 底层与完整 CLI 进程通信 -- 每个并发用户 = 一个终端进程。」 Out of sync: already removed from the English README; the official SDK is now named Claude Agent SDK

AI assistance

I used AI to compare README-zh.md with README.md, to check each discrepancy I had found against the English source, and to refine the revised Chinese wording.

I reviewed every change in this PR myself and decided which ones to include.

- Fix mistranslations: prompt-plumbing, curate, focused work,
  trajectory data
- Rewrite word-for-word sentences in natural Chinese (5 places)
- Sync with README.md: "the model will do the rest", and drop two
  sentences the English README has already removed ("that model
  is the agent" and an SDK sentence)
- Fix a typo in the learning-path diagram

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

README-zh.md: mistranslations, word-for-word phrasing, and lines out of sync with README.md

1 participant