MIT 授權的跨代理論文學習外掛，透過一條 npx 命令以 Agent Skills 形式接入 DeepSeek Harness，提供自適應精讀教材、快速摘要與 localhost:5815 本機 Web 閱讀器。
DSH 整合
相容
作者聲明
安全稽核
未稽核
最後核實
2026-08-21
授權條款
MIT
01它能幫你完成什麼？
Study a research paper deeply inside DeepSeek Harness
Adaptive learning package — README, detailed summary, key insights, Q&A, code demonstrations with Jupyter notebooks, and extracted figures — stored under ~/claude-papers/ with a global search index and an auto-launched web viewer
Researchers, students and engineers running DeepSeek Harness who want papers turned into structured, searchable study material
Screen research papers quickly before committing to a deep study
quick-summary.md with a concise 300–500 word overview, while the PDF, complete paper.txt and metadata stay in the shared paper library
Readers triaging many papers in DeepSeek Harness who need a fast overview without running the full study workflow
Browse and search the studied paper library from DeepSeek Harness
Interactive Nuxt.js web viewer at http://localhost:5815 with KaTeX math rendering, code demonstrations and full-library search
Learners who want to revisit generated study materials and search their paper collection in a local web UI
02如何將外掛接入 DeepSeek Harness？
先決條件
- Node.js 20.19.x, or 22.12.0 and higher
- poppler-utils for PDF image extraction (macOS: brew install poppler; Ubuntu/Debian: sudo apt-get install poppler-utils; Arch: sudo pacman -S poppler)
- DeepSeek Harness installed as the agent host
安裝步驟
- 01
Run `npx --yes @zlzliqing/claude-paper@latest install --target deepseek-harness` — the installer copies the plugin runtime to the user data directory and installs the generated compatibility Skills
$ npx --yes @zlzliqing/claude-paper@latest install --target deepseek-harness
- 02
Restart DeepSeek Harness after installation or upgrade
驗證整合成功
作者未說明
03DSH 整合程度與能力邊界
Installed into DeepSeek Harness as Agent Skills via the npx installer (generated discovery adapters); invoke by asking the agent or loading the claude-paper-summary / claude-paper-webui Skills
Automatic PDF parsing & figure extraction
local PDF path or arXiv PDF/abstract URL→title, authors, abstract, links and complete paper text (paper.txt), meta.json with a 50k preview, detected code-repository links, and extracted figures
writes paper.pdf, paper.txt, meta.json and images/ into ~/claude-papers/papers/{paper-slug}/Adaptive learning material generation
parsed paper with difficulty-level and paper-type assessment→README, summary.md, insights.md, qa.md, plus method/mental-model/reflection notes when the paper complexity warrants them
writes Markdown study materials into the paper's library folder and updates the global search index (index.json)Code demonstrations
papers with detected GitHub / arXiv / CodeOcean code links→clean reference implementations (core-demo.py) and interactive Jupyter notebooks (concept-demo.ipynb), with original code integration when available
writes code demos into the paper library's code/ folderInteractive web viewer
the studied paper library at ~/claude-papers/→Nuxt.js web interface at http://localhost:5815 with KaTeX math rendering, material browsing, code demos and library search
launches a local web server on port 5815
04適合誰？何時不該用？
適合
- Researchers, students and engineers running DeepSeek Harness who want papers turned into structured, searchable study material
- Readers triaging many papers in DeepSeek Harness who need a fast overview without running the full study workflow
- Learners who want to revisit generated study materials and search their paper collection in a local web UI
不適合
- PDF image extraction depends on poppler-utils, which must be installed separately through the system package manager (brew / apt / pacman).
05相容性、維護與安全提醒
- Requires Node.js 20.19.x or 22.12.0 and higher; older Node versions cannot run the npx installer or the plugin.
- PDF image extraction depends on poppler-utils, which must be installed separately through the system package manager (brew / apt / pacman).
- DeepSeek Harness must be restarted after installation or upgrade before the newly installed Skills become available.
MIT · actively maintained (last push 2026-08-14)
06常見問題
如何把 Claude Paper 安裝到 DeepSeek Harness？
執行 `npx --yes @zlzliqing/claude-paper@latest install --target deepseek-harness`。安裝程式會把外掛執行階段複製到使用者資料目錄並安裝產生的相容 Skills，之後重新啟動 DeepSeek Harness 即可。預設的 `all` 目標會一次涵蓋 Claude Code、Codex、OpenCode 和 DeepSeek Harness 四個代理。
它是 DeepSeek Harness 原生外掛還是 MCP 伺服器？
不是 MCP，也無需複製儲存庫：Claude Paper 透過 Agent Skills 運作。README 明確說明不需另外安裝或設定 MCP 伺服器，安裝程式會為 DeepSeek Harness 產生探索與呼叫轉接器。
需要什麼前置條件？
Node.js 20.19.x 或 22.12.0 以上版本；PDF 圖片擷取還需要 poppler-utils（macOS 用 brew install poppler，Ubuntu/Debian 用 sudo apt-get install poppler-utils，Arch 用 sudo pacman -S poppler）。
在 DeepSeek Harness 裡怎麼使用？
直接請代理研讀論文即可，支援本機路徑或 arXiv 連結（如 https://arxiv.org/abs/1706.03762）；也可以要求快速論文摘要，或明確載入 `claude-paper-summary` / `claude-paper-webui` Skill。
我的資料會存在哪裡？
全部儲存在本機：論文與產生的教材依 ~/claude-papers/papers/{paper-slug}/ 組織，並維護全域 index.json 搜尋索引；Web 閱讀器執行在 http://localhost:5815。
07相關的 DSH 工作流程
dsh-web
作者 zhu1090093659
DeepSeek Harness（dsh）外掛聚合生態，萬物皆外掛，通過創意工坊分發。
petdex
作者 crafter-station
面向 Codex、Claude Code、DeepSeek Harness 等智慧體的動效寵物公共相簿。
j-space-cognition-suite-v3.7
作者 tiger3807861189
J-Space Cognition Suite V3.7 - AI cognitive-enhancement Skills based on Anthropic's J-space global workspace research. | 嗶哩嗶哩：Tiger380 (UID 3494375382321675) — https://space.bilibili.com/3494375382321675
dsh-deep-whale
作者 small-tailqwq
Whale Girl skin series for DeepSeek Harness. 適用於 DeepSeek Harness 的，鯨魚娘系列皮膚。
08資料與來源
A research-paper learning plugin for **Claude Code, Codex, OpenCode, and DeepSeek Harness**.
npx --yes @zlzliqing/claude-paper@latest install --target deepseek-harness
**Agent Adapters** - Generated discovery and invocation wrappers for Codex, OpenCode, and DeepSeek Harness
此頁面根據專案公開文件、儲存庫中繼資料與 DSH Plugins 的結構化解析所產生；最後核實於 2026-08-21。發現錯誤？提交更正。
最佳 DeepSeek Harness 外掛
從全目錄挑出的 12 個值得優先安裝的外掛，涵蓋各個分類。