| name | article-translate |
| description | Translate web articles, blog posts, essays, and technical posts from a URL into Chinese Markdown files. Use this skill only when the user explicitly asks to translate, with requests like '翻译这篇文章', '翻译一下这个链接', '把这个网页翻成中文', 'translate this post', or asks to continue translating an existing article directory. Do not trigger this skill for a bare URL alone without translation intent. |
Article Translate
Translate a high-quality article from a URL into polished Chinese Markdown files.
When To Use This Skill
Use this skill when the user wants to:
- translate an article from a URL with explicit translation intent
- translate a blog post, essay, or technical post into Chinese
- fetch an article and translate it
- continue translating an existing article directory
Do not use this skill for:
- a bare URL without words that mean translate, such as
翻译, 翻成中文, or translate
- requests to open, read, summarize, explain, or analyze a URL without asking for translation
- translating a short paragraph directly in chat
- explaining a single sentence or phrase
- non-article writing tasks
Workflow
The skill directory is at SKILL_DIR (the directory containing SKILL.md).
1. Prepare
Check the local environment first:
node SKILL_DIR/scripts/prepare-workspace.mjs
If articrab is missing, guide the user to run:
npm install -g articrab
articrab set-jina-token jina_xxx
If the Jina token is missing, STOP and call AskUserQuestion to ask the user to either:
- provide an existing Jina token
- or go to the official Jina site to apply for one
After the user provides the token, run:
articrab set-jina-token <token>
Help the user complete the configuration before continuing.
Do not assume token setup is correct until fetching succeeds.
2. Fetch
If the user provided a URL, fetch the article:
node SKILL_DIR/scripts/fetch-article.mjs "<url>" --out "<workspace-root>"
This creates an article directory and writes the source files needed for translation.
3. Analyze And Build Prompt
Build the analysis and translation prompt:
node SKILL_DIR/scripts/build-translation-prompt.mjs --article-dir "<article-dir>" --lang zh
Do not translate directly from source.md.
If the user already has an article directory with source.clean.md, resume here instead of fetching again.
4. Translate
Read 02-translate.<lang>.prompt.txt and write a first complete draft of the translation.
The draft must:
- start with a translated H1 title using
#
- leave one blank line after the title
- preserve Markdown structure
- keep code blocks, commands, URLs, and file paths unchanged
- preserve generated placeholders unchanged when they appear
Write this draft to 03-draft.md.
5. Critique Draft
Generate critique notes for 03-draft.md:
node SKILL_DIR/scripts/stage-draft.mjs --article-dir "<article-dir>" --lang zh
This writes:
The script writes the deterministic checks. Then the Agent must read the source and the draft, and complete the ## Agent Critique section inside 04-critique.md.
Replace every pending agent review line with a concrete review result, then set review-decision to revise or pass.
The critique should cover:
- terminology consistency
- title quality
- fluency in Chinese
- missing or distorted meaning
- Markdown integrity
6. Revise And Repeat
Read 04-critique.md and decide based on its contents:
- If
review-decision: pass, continue to Apply Final.
- If there are problems, revise
03-draft.md and run Critique Draft again.
Do not regenerate the draft from scratch, and do not publish until the critique issues are resolved.
Do not run Apply Final while 04-critique.md still contains any pending agent review placeholder.
7. Apply Final
After critique issues are resolved, apply the revised 03-draft.md as the final translation:
node SKILL_DIR/scripts/apply-final.mjs --article-dir "<article-dir>" --lang zh --path-only
This writes:
05-revision.md06-quality-report.json- final
<lang>.md
After Apply Final succeeds, respond with the absolute path of the translated file only.
Do not add status text, bullets, summaries, explanations, or any other surrounding text.
Do not describe the generated files.
Output
Each article directory should contain these source files:
source.mdmeta.jsonsource.clean.md
And these workflow files:
01-analysis.md02-translate.<lang>.prompt.txt03-draft.md04-critique.md05-revision.md06-quality-report.json<lang>.md
Failure Handling
If environment setup fails:
- install
articrab
- configure the Jina token
- rerun the prepare command above
If fetch fails:
- confirm the URL is reachable
- confirm
articrab works in the shell
- confirm the Jina token is valid
If draft staging or final apply fails:
- check that the draft still contains every required placeholder
- check that the draft still starts with a Markdown H1
- check that
04-critique.md no longer contains pending agent review
- check that
04-critique.md ends with review-decision: pass
- fix the draft and rerun the appropriate stage
Do not skip failed stages. Resume from the last valid file.
Translation Rules
Always:
- translate the title
- preserve Markdown headings, lists, quotes, tables, and links
- keep code blocks, commands, URLs, file paths, and API names unchanged
- keep generated placeholders unchanged when they appear
- translate the title instead of copying the English title unchanged
- write critique and revision notes before applying the final file
Examples
Example 1
Input: 翻译 https://openai.com/index/harness-engineering/
Behavior: run the full workflow from prepare to final zh.md.
Final response: output only the absolute path to zh.md.
Example 2
Input: 把这篇文章翻成中文：https://martinfowler.com/articles/agentic-ai.html
Behavior: fetch the article, build the translation prompt file, write a draft, critique it, revise it as needed, and then apply the final files.
Final response: output only the absolute path to the translated file.
Example 3
Input: https://openai.com/index/harness-engineering/
Behavior: do not use this skill, because a bare URL alone does not express translation intent.
