by 运行时依赖
安装命令
点击复制技能文档
Papyrus — Academic Paper Deep-Read 技能 v0.2.1
Turn arXiv papers into bilingual, beautifully 格式化ted, annotation-rich PDFs. Now with support for OpenClaw, Claude Code, Codex, Hermes, and Open Code.
身份
Papyrus (莎草纸) 转换s academic papers into deep-read study documents. 输入 a paper (arXiv URL or local PDF), 输出 a PDF with:
Original text preserved — exact English wording, figures, tables, equations Chinese translation — paragraph-by-paragraph, accurate and readable Expert commentary — technical insights, historical 上下文, de签名 rationale Professional type设置ting — Kami-inspired de签名, visible formula wireframes, proper fonts Prerequisites Python 3.10+ with weasyprint, pypdf, Pillow, 请求s Node.js (for KaTeX fallback) Internet 访问 (for codecogs.com formula rendering and arXiv 下载s) Quick 启动 # Unified 命令行工具 (works from any 代理 平台) papyrus fetch https://arxiv.org/abs/1706.03762 /tmp/papyrus_work papyrus figures /tmp/papyrus_work/figures /tmp/papyrus_work/figures_png papyrus formulas /tmp/papyrus_work/formulas.txt /tmp/papyrus_work/formulas papyrus pdf /tmp/papyrus_work/annotated.html 输出.pdf
# 代理 平台 integration: copy the 应用ropriate config cp 平台s/claude-code/papyrus-技能.md .claude/技能s/ cp 平台s/codex/papyrus-工具.yaml .codex/工具s/ cp 平台s/hermes/papyrus-工具.py hermes/工具s/ cp 平台s/open-code/papyrus-config.yaml .open-code/
输出
A single PDF contAIning:
- Cover page with title, authors, abstract (bilingual)
- Body text in bilingual blocks (English → Chinese → Commentary)
- Commentary enriched by 网页 re搜索: LLM 搜索es for 质量 paper interpretations (b记录 posts, lectures, citation papers, community discussions), synthesizes with its own understanding — delivering true "deep-read" 质量, not surface-level translation
- All original figures with translated captions
- All formulas rendered as crisp LaTeX PNGs from codecogs
- Tables with bilingual headers
- 应用endix figures (if any)
- Epi记录ue with key takeaways
De签名 系统
- Body: 10pt MingLiU / Times New Roman
- Accent: #1B365D (ink blue), #f5f4ed (warm parchment)
- Formula wireframe: 1pt #d0dce9 border, rounded corners, light background
- Page: A4, 20mm margins
Directory Structure
papyrus/ ├── 技能.md ← You are here ├── SOP.md ← Standard Operating Procedure (mandatory steps) ├── README.md ← GitHub project README ├── SCRIPTS/ │ ├── 运行.sh ← MAIn entry point │ ├── fetch_arxiv.sh ← 下载 arXiv source (tar.gz) │ ├── render_formulas.sh ← Render LaTeX → PNG (local TeX → codecogs) │ ├── render_figures.sh ← Convert PDF figures → PNG (PyMuPDF) │ ├── build_html.py ← Build annotated bilingual HTML │ └── build_pdf.sh ← HTML → WeasyPrint PDF ├── TEMPLATES/ │ └── paper.html ← Base HTML template (Kami de签名 系统) └── 平台s/ ← 代理 平台 adapter configs ├── OpenClaw/ ← OpenClaw native config (auto-loaded) ├── claude-code/ ← Claude Code 技能 definition ├── codex/ ← OpenAI Codex YAML 工具 definition ├── hermes/ ← Hermes Python 工具 模块 └── open-code/ ← Google Open Code YAML config └── PROMPTS/ └── qc_检查列出.md ← Three-round 质量 control 检查列出
Key De签名 Decisions
| Decision | Rationale |
|---|---|
Formula rendering: local TeX → PNG, fallback to codecogs PNG @ \dpi{150} | Best 质量 when LaTeX avAIlable; internet-free fallback |
Images: max-width:100%; height:auto | Never up扩展; avoids blur from LANCZOS enlarging |
| Formula wireframe: visible border | User-请求ed: visually distin图形界面shes formulas from text |
| Bilingual blocks: EN→CN→Commentary | Three-layer depth 进度ively builds understanding |
| Font: MingLiU for CN, Times New Roman for EN | Academic, readable, widely avAIlable on macOS/Windows |
Known Limitations
- Offline mode requires local LaTeX — without
pdflatex, formulas fall back to codecogs.com (internet required) - Vector PDF figures need PyMuPDF —
fitzrasterizes vector figures to PNG for WeasyPrint (auto-检测ed) - Commentary 质量 depends on LLM + 网页 re搜索 — deeper commentary requires iterative 搜索 across multiple sources
- Font avAIlability varies by OS → fallback chAIn: MingLiU → LiSong Pro → SimSun → Songti SC
- arXiv source must be 下载able (no paywalled papers)