360 Secure SkillHub
360 安全技能中心

translate-to-zh-hans

仅在用户明确指定具体文章时触发，不会自动运行，翻译前需确认源文件路径、输出位置及文件覆盖规则。遵循双语注释规范：代码、产品名、命令等专有内容原样保留，技术术语首次出现加英文括注，陌生概念补充通俗解释，完整保留原文结构与资源引用。输入为docs目录下指定Markdown文章，输出为同目录带-zh-hans.md后缀的简体中文译文。

j3ffyang
j3ffyang
热度 16.23
最近更新 2026-10-07 21:52:22
分类：效率依赖：读写文件 / 内部调用大模型应用场景：翻译

安装技能

复制下面的话术发给你的 Agent，它会通过 secure-skills 检索并安装该技能。

执行命令安装此 skill： npx secure-skills-cli add translate-to-zh-hans
Metadata
name
translate-to-zh-hans
description
Translate a specific article from ai-thoughts/docs/ into Simplified Chinese, writing the output to an exactly-same-filename "-zh-hans.md" file. Use when the user names a specific article and asks to translate it (e.g. "translate 260803-ollama-to-llamacpp", "翻译 xxx", "make a -zh-hans.md version" — or the legacy "make a -chn.md version"). Never runs automatically; only acts on an explicitly chosen article.

Translate to Chinese (-zh-hans.md)

Translate one article from ai-thoughts/docs/ into Simplified Chinese and write it to a new file with the same filename plus a -zh-hans.md suffix.

Approval gate — read before anything else

  • Never translate, write, or create anything without explicit user approval. This skill converts on demand only; it does not run automatically and it does not pick candidate articles on its own.
  • No action if no article is specifically chosen. The user must point out exactly one article (by filename, slug, or date). If they don't, do nothing except ask which one.
  • Legacy naming: files translated before this rule was standardized are named <YYMMDD-slug>-chn.md and keep that name until the user explicitly migrates them; every new translation uses <YYMMDD-slug>-zh-hans.md.
  • Before writing the output, confirm the plan with the user: the source file, the exact output path, and whether an existing -zh-hans.md should be overwritten.

Inputs

  • source — Path to the source article, e.g. ai-thoughts/docs/260803-ollama-to-llamacpp.md. Required. If the user only names an article by slug or date, locate it in ai-thoughts/docs/.

Outputs

  • outputPath — Path to the translated article: docs/<exact-same-filename>-zh-hans.md (replace the .md suffix with -zh-hans.md, keeping the YYMMDD-slug prefix and all characters exactly). Example: 260803-ollama-to-llamacpp.md → 260803-ollama-to-llamacpp-zh-hans.md.

Procedure

  1. Wait for the user to point out the article to translate. Do not start until they name it explicitly.
  2. Confirm the plan with the user: the source file and the output path. Get their go-ahead before writing anything.
  3. Check for an existing translation. If docs/<same>-zh-hans.md — or the legacy docs/<same>-chn.md — already exists, stop and ask whether to overwrite, diff against it, or skip. Never overwrite silently.
  4. Read the source article from source. Understand the full content before translating.
  5. Translate the body into Simplified Chinese:
    • Keep the title as an H1 with the key term followed by a Chinese translation (match the style of existing Chinese pairs — most are still legacy -chn.md).
    • Right after the H1, add the source link line, using the absolute GitHub URL (the j3ffyang repo), not a relative link: **原文：** [<source-filename>](https://github.com/j3ffyang/ai-thoughts/blob/main/docs/<source-filename>)
    • Translate the running text, headings, and bullets faithfully and naturally. Do not add, remove, or reorder content.
    • Write Simplified Chinese by default; Traditional Chinese only when the source was originally written that way.
    • Apply the bilingual-gloss style below: gloss technical terms and section titles, and add plain-language explanations for unfamiliar concepts.
  6. Do not translate specific terms — leave them verbatim:
    • Product, tool, and agent names: Hermes Agent, OpenCode, OpenClaw, Astro, etc.
    • Technical terms, commands, flags, config and file names (AGENTS.md, SOUL.md, .cursorrules, hermes -c), environment variables, URLs.
    • Code blocks, inline code, CLI keystrokes (Ctrl+C, Alt+Enter).
    • Existing proper nouns already in Latin script. Only translate the surrounding prose; never translate inside code.
  7. Preserve everything else exactly: ../imgs/<file> image references (keep the path verbatim — Chinese files mirror English per AGENTS.md), links, HTML, tables, frontmatter (if any), and Markdown structure.
  8. Write the output to docs/<same-filename>-zh-hans.md with the source link line followed by the translated body.
  9. Arch sign-off: if the source ends with btw, i use arch, append the identical line at the bottom of the -zh-hans.md. Keep it verbatim English — do not translate.
  10. Report outputPath to the user. Do not modify the source file, and do not touch articles.yaml or the READMEs (out of scope for this skill).

Bilingual-gloss style (default)

Every -zh-hans.md uses this style: Simplified Chinese is the primary text, and English is added as glosses so a Chinese-only reader still gets the technical vocabulary. The source article stays as-is; the glosses and explanations live only in the -zh-hans.md.

  1. Gloss technical terms inline. The first meaningful time a technical term appears, follow it with the English in parentheses: 技能 (skill), 无头 (headless), 密钥链 (keychain), 接口额度 (API quota). Don't gloss every repetition or everyday words — only technical or specialized vocabulary.
  2. Gloss section titles. Add the gloss to headings that contain technical terms: ## 技能 (skill) 是什么, ## 无头 (headless) 环境下的认证.
  3. Add plain-language explanations. For concepts a general reader won't know (e.g. what 无头 (headless) 环境 means), add a short, simple explanation in plain Chinese at the first use. Understanding beats literal fidelity here — the explanation may be a small addition to the source, since the goal is a standalone readable article for a Chinese audience.
  4. Gloss inside tables too. The same rules apply to table cells — gloss technical terms on first appearance: | 我——思维流 (thought-flow) | OpenCode + git——工作流 (workflow) |.

The verbatim exceptions from Procedure step 6 still apply: code, commands, product names, and URLs stay untouched, and glosses never go inside code blocks.

Quality rules

  • Avoid hallucination. If a word, reference, or term is unclear, do not guess — ask the user or flag it in the report. Never invent translations, names, dates, or content that is not in the source.
  • Ask if anything is unclear — about the article, an ambiguous phrase, an unknown acronym, or the intended output. Clarify rather than assume.
  • Translate meaning faithfully and naturally in Simplified Chinese; do not perform a word-for-word substitution.

Verification

  • Output is at docs/<exact-source-filename>-zh-hans.md; the prefix and slug match the source filename exactly (only the .md → -zh-hans.md suffix changed).
  • The source link line is present right after the H1 and uses the absolute GitHub URL: **原文：** [<source-filename>](https://github.com/j3ffyang/ai-thoughts/blob/main/docs/<source-filename>).
  • Every ../imgs/<file> reference in the output matches one in the source (verify with a glob/ls against ai-thoughts/imgs/).
  • Code blocks, inline code, commands, and specific terms (e.g. Hermes Agent, OpenCode) are preserved verbatim.
  • The bilingual-gloss style is applied: technical terms and headings carry English glosses, table cells are glossed on first appearance, and unfamiliar concepts get a plain-language explanation at first use.
  • If the source ends with the arch sign-off (btw, i use arch), the -zh-hans.md carries the identical verbatim line.
  • The source file is unmodified (confirm via git status/diff).

Error Handling

  • No article chosen: do nothing; ask the user which article to translate.
  • Source not found: list candidate files in ai-thoughts/docs/ and ask which to use.
  • Output already exists: stop and ask whether to overwrite, diff against it, or skip.
  • Ambiguous or untranslatable content: ask the user instead of guessing.