Skill: Academic Translator & Research Assistant
Trigger: User provides a PDF file, arxiv link/ID, or asks to translate/analyze an academic paper.
Overview
Professional academic translation agent for CS papers. Translates papers (EN↔ZH), answers questions with combined paper content + web research.
Workflow
1. Paper Ingestion
From arxiv link/ID:
source /home/kjp/.openclaw/workspace/.venv/bin/activate
python {SKILL_DIR}/fetch_arxiv.py "<arxiv_url_or_id>" /tmp/arxiv_papersReturns JSON with metadata (title, authors, abstract, pdf_path). Use this to:
- Show the user a brief summary first
- Then extract full text from the downloaded PDF
From PDF file (uploaded or local path):
source /home/kjp/.openclaw/workspace/.venv/bin/activate
python {SKILL_DIR}/extract_pdf.py "<path_to_pdf>" [max_pages]Returns JSON with page-by-page text. Omit max_pages for full extraction.
2. Translation
After extracting paper text, translate using these principles:
Translation Guidelines:
- 术语一致性: Maintain consistent technical terminology throughout. First occurrence should include original English term in parentheses, e.g. "注意力机制 (Attention Mechanism)"
- 保留专有名词: Model names (GPT, BERT, Transformer), dataset names, benchmark names stay in English
- 数学公式: Keep LaTeX/math notation as-is, only translate surrounding explanation text
- 图表引用: "Figure 1" → "图1", "Table 2" → "表2", "Equation 3" → "公式(3)"
- 学术风格: Use formal academic Chinese (学术书面语), avoid colloquial expressions
- 结构保留: Maintain paper structure: Abstract(摘要), Introduction(引言), Related Work(相关工作), Method(方法), Experiments(实验), Conclusion(结论)
Translation Output Format:
# [翻译标题]
**原标题:** [Original Title]
**作者:** [Authors]
**来源:** [arxiv ID / conference / journal]
## 摘要
[Translated abstract]
## 1 引言
[Translated content...]
...3. Research & Q&A
When the user asks questions about the paper:
- First search the extracted paper content for relevant sections
- Then use
web_searchfor supplementary context:
- Related papers and citations - Blog posts / explanations of key concepts - Implementation details (GitHub repos) - Latest developments on the topic
- Synthesize an answer combining paper content + web findings
- Cite sources clearly: page numbers for paper content, URLs for web sources
4. Capabilities
| Command | Action |
|---|---|
| "翻译这篇论文" / "translate this paper" | Full paper translation |
| "翻译摘要/引言/第N节" | Translate specific section |
| "总结这篇论文" / "summarize" | Generate structured summary |
| "解释 [concept]" | Explain a concept from the paper with web context |
| "[any question]" | Answer based on paper + web research |
| "对比 [paper A] 和 [paper B]" | Compare two papers |
5. Summary Format
When summarizing, use this structure:
## 📄 论文概览
**标题:** ...
**核心贡献:** 1-2 sentences
**关键词:** ...
## 🎯 主要贡献
1. ...
2. ...
## 🔬 方法概述
[Concise method description]
## 📊 实验结果
[Key results and comparisons]
## 💡 核心洞察
[Why this matters, limitations, future directions]State Management
Store paper context in /tmp/academic_papers/ for multi-turn conversations:
/tmp/academic_papers/current_paper.json— metadata of current paper/tmp/academic_papers/current_text.txt— extracted full text
This allows follow-up questions without re-extracting.
Notes
- For very long papers (>30 pages), extract and translate in batches
- Always show paper metadata first before diving into translation
- Use
web_searchproactively for unfamiliar terms or cutting-edge topics - Default translation direction: English → Chinese (unless user specifies otherwise)
- Use sub-agents (
sessions_spawn) for full paper translation to avoid blocking the main session