From paper-skills
把一篇论文拆解成结构化研究卡片。 当用户说"做个卡片"、"写阅读笔记"、"总结这篇论文"、 "帮我读一下"、"结构化分析"、"拆解论文"时触发。 输出 card.md(快速概览)和 card-deep.md(深度分析)。
How this skill is triggered — by the user, by Claude, or both
Slash command
/paper-skills:paper-card <论文引用 | folder_slug | PDF路径> [--lang zh|en]<论文引用 | folder_slug | PDF路径> [--lang zh|en]The summary Claude sees in its skill listing — used to decide when to auto-load this skill
Generate structured research cards from an acquired paper.
Generate structured research cards from an acquired paper.
This skill owns:
$PAPERS_DIR/{folder_slug}/card.md — quick card$PAPERS_DIR/{folder_slug}/card-deep.md — deep cardThis skill does not:
paper-acquire)paper-search)paper-repo)metadata.yaml按以下优先级检测输入类型:
$PAPERS_DIR/{input}/metadata.yaml 是否存在。
存在 → 检查 paper/paper.md → 生成卡片。
输入看起来像 folder_slug(无 .pdf 后缀、无 URL scheme、无 DOI/arXiv 前缀)但目录不存在时,告知用户该论文未导入,询问是否需要导入。.pdf 文件路径(Path(input).suffix == '.pdf' and Path(input).is_file())。
→ 调 pdf-to-md 转 markdown → 调 paper-search 解析身份 → 复制 PDF 和 markdown 到 $PAPERS_DIR/{folder_slug}/paper/ → 生成卡片。paper-import 完成 search+acquire → 用生成的 folder_slug → 生成卡片。用户给本地 PDF(如 ~/Downloads/paper.pdf)时:
pdf-to-md 把 PDF 转成 markdown,输出到 PDF 同目录2301.13688.pdf → arXiv ID,attention_is_all_you_need.pdf → 标题关键词)paper-search 生成 metadata.yaml$PAPERS_DIR/{folder_slug}/paper/ 下创建标准目录结构,复制(不移动)PDF 和 markdown非 folder_slug、非本地 PDF 的输入,agent 应:
paper-import skill,传入用户的论文引用在 Windows 上执行 paper-acquire / pdf-to-md 的 bash 命令时:
PYTHONIOENCODING=utf-8 环境变量避免 GBK 编码错误MINERU_API_TOKEN 在当前 shell session 中可用(检查 $env:MINERU_API_TOKEN 或 os.environ)python3 可能不存在,使用 python 替代检测输入类型并确保前置条件满足(详见 Input Handling)。
完成后,以下文件必须存在:
$PAPERS_DIR/{folder_slug}/metadata.yaml$PAPERS_DIR/{folder_slug}/paper/paper.mdRead both files:
$PAPERS_DIR/{folder_slug}/metadata.yaml — for frontmatter (title, authors, year, venue, url, code, tags)$PAPERS_DIR/{folder_slug}/paper/paper.md — full paper textRead both templates (templates/paper-card-deep.md and templates/paper-card-quick.md). The deep card template contains inline extraction rules (in HTML comments and annotation blocks) that guide how to fill each section. Follow them.
Read the full paper.md and fill both templates section by section:
metadata.yaml + your assessment (tags, builds_on, contrasts_with, ratings, verdict)Optional parallel execution: If the caller supports subagent dispatch, run two independent extractions in parallel for cross-validation. Both read paper.md independently and follow the same templates. When two results are available, use cross-validation rules:
If subagents are not available, the main agent performs the extraction directly — same templates, single pass.
Write to:
$PAPERS_DIR/{folder_slug}/card.md$PAPERS_DIR/{folder_slug}/card-deep.mdBoth files should be self-contained Markdown with YAML frontmatter.
卡片输出语言由 --lang 参数决定。未指定时跟随用户对话语言(用户用中文提问则输出中文,用英文则输出英文)。
语言规则:
模板文件保持英文不变。agent 在生成卡片时根据语言指令翻译 section headers 和正文内容。
Both templates contain inline extraction rules (HTML comments). Read them for guidance, but do NOT include them in the output. Preserve all section headers, numbering, annotations, and callout syntax exactly as they appear.
This skill runs after paper-acquire (which runs after paper-search).
Use paper-import for end-to-end workflow including card generation.
npx claudepluginhub huangyrcn/toolshed --plugin paper-skillsGuides creation, editing, and verification of skills for AI coding agents using test-driven development with subagent scenarios. Use when authoring or debugging skills.