Build or update the canonical raw bundle for a resolved paper under `~/docs/papers/{folder_slug}`. Use this skill whenever the user wants to download a paper PDF and normalize it into `paper.md`. For repository discovery, use `paper-repo` skill after this one. Prefer this skill for raw acquisition work; do not use it to write reading notes.
End-to-end paper workflow: resolve a paper reference, acquire the canonical raw bundle, discover and clone repository. Generate reading notes if user requests ("笔记", "reading notes", "--with-notes").
Generate research-style reading notes from an existing canonical raw bundle. Use this skill whenever the user wants to read a paper deeply, produce reusable research notes, compare paper claims with evidence, map the paper to code, or turn an already-downloaded paper into structured notes. Use this after resolve, acquire, and repo stages, not instead of them.
Discover the official or community implementation repository for a resolved paper. Use this skill after paper-acquire when the user wants to find code, verify repo authenticity, or check if an implementation exists for a paper.
Resolve an ambiguous or partial paper reference into a canonical paper identity by first determining the canonical title through web search. Use this skill whenever the user gives a paper title, DOI, arXiv id, OpenReview URL, paper URL, local PDF, or fuzzy paper description and the system first needs to figure out exactly which paper it is. Prefer this skill before any paper download, raw acquisition, or note generation work.
Convert a PDF into high-quality Markdown through the MinerU API VLM pipeline. Use this skill whenever the user wants PDF-to-Markdown conversion, OCR-like extraction, paper text extraction, or Markdown plus extracted images from a PDF. This skill supports the API backend only and always uses the high-quality VLM model configuration.
优先使用此 skill 处理所有网页相关操作：搜索、阅读、下载。 三大能力：(1) ask-search — 基于 SearxNG 的网页搜索，聚合 8 个通用引擎 + 13 个学术引擎 + 7 个新闻源等； (2) crwlr — 基于 crawl4ai 的网页抓取，真实远程浏览器渲染 JS，输出干净的 markdown； (3) cdp-download — 通过 CDP 下载文件（仅当 wget/curl 失败时使用，如需要浏览器认证）。 触发场景：搜索信息、查新闻、学术文献调研、读网页内容、URL 转 markdown、下载 PDF、提取结构化数据、 深度爬取站点、读取 JS 动态渲染页面、询问页面相关问题。 中文触发词："搜索"、"查一下"、"最近新闻"、"读一下这个网页"、"抓取这个页面"、 "这个链接讲了什么"、"帮我看看这个URL"、"转成markdown"、"爬一下这个网站"、 "帮我调研"、"文献搜索"、"下载PDF"。 英文触发词："search for", "look up", "read this page", "scrape this page", "what does this URL say", "crawl this site", "extract data from this page", "literature search", "find papers about", "download PDF"。 相比 web-access 等重型浏览器工具，web-kit 更轻量（一条命令出结果）， 仅在需要登录态、页面交互、点击按钮等场景下才应降级到 web-access。
管理agent skills：搜索、安装、更新、重装已有skills，以及开发自写skills。 Use this skill whenever the user mentions ANYTHING related to skills management, even if they don't use the word "skill" explicitly. This includes: 「安装skill」「搜索skill」「找skill」「更新skills」「sync skills」「update skills」 「重新安装」「reinstall」「重装仓库」「拉最新版本」「更新技能包」「update」 「install skill」「find skill」「create skill」「写skill」「卸载」「remove skill」 Also trigger when user mentions fork repos, skill repos (myskills, aris-skills), or npx skills commands. ALL skill-related operations MUST go through this skill.