Yuan1z0825/nature-skills

Project Overview

nature-skills collects reusable skills packaged around SKILL.md files for academic research tasks. The repository targets researchers who work with coding agents such as Claude Code, Codex, OpenClaw, OpenCode, Hermes or Chatbox and want structured assistance with paper reading, polishing, drafting, reviewing, citation management, statistical reporting, figure preparation, slide generation and revision responses. Each skill lives in its own directory under skills/ and can be installed independently; skills/nature-shared holds shared support files and is not counted as a standalone skill.

The skill index lists 23 skills, with status labels Draft, Beta and Stable indicating how far each workflow has been tested. Examples include nature-polishing for Nature-style English editing, nature-figure for publication-quality plots in Python or R, nature-reader for bilingual Markdown paper readers, nature-response for point-by-point reviewer replies, and nature-paper2ppt for Chinese journal-club decks. The project also provides a web version with paper polishing and figure functions at natureskills.cn.

Quick Start

Install one skill with the npx skills workflow. Node.js 18 or later is required requirements. First list the available skill names, then install only what you need. Installing nature-reader, nature-paper2ppt, nature-polishing or nature-writing requires the shared support package nature-shared.

npx skills add Yuan1z0825/nature-skills --list
npx skills add Yuan1z0825/nature-skills --global --agent codex \
  --skill nature-reader --skill nature-shared --yes --copy

After installation, start a new agent session and describe the task in natural language. For example: 把这篇论文做成中英文对照的完整 Markdown reader。 You can also name the skill explicitly: 请使用 nature-polishing，把下面这段中文改写为 Nature 风格英文。 The README provides a table of ready-to-copy prompts for reading papers, generating a journal-club PPT, polishing a paragraph, drafting an abstract, simulating peer review and creating figures prompt examples.

For Claude Code, the README recommends keeping a local clone and pointing a subagent or slash command at the real skills/*/SKILL.md path rather than copying individual files Claude Code setup. Codex users can run scripts/update-codex-skills.sh --pull to synchronise the full skill set and --check to verify the installation Codex setup. A graphical installation path through Chatbox is also documented Chatbox setup.

Runtime and Dependencies

Installing a skill copies instructions and supporting files, but does not install runtime dependencies. Python packages needed by individual skills must be installed separately, for example python -m pip install -r skills/nature-paper-to-patent/requirements.txt or the MCP server requirements for nature-academic-search dependency notes. The patent-disclosure search feature additionally requires python -m playwright install chromium. The nature-academic-search MCP service requires a separate PUBMED_EMAIL configuration, and optional Scopus or ScienceDirect providers need locally configured credentials that should not be committed to the repository.

Some skills depend on external services. nature-figure can generate figure drafts through OpenRouter GPT Image 2, which requires corresponding service credentials. The README notes that skills involving local files or scripts need directory and command permissions granted through the host application. Installation alone does not configure these external capabilities.

Usage Constraints

Skill selection uses the frontmatter names shown by npx skills --list; for example, the directory nature-proposal-writer currently appears as researchwrite name note. The README warns against copying only SKILL.md because many skills rely on references/, static/, manifest.yaml, scripts and assets. The project is licensed under Apache-2.0 license.

The README also includes commercial sections for GPT/Claude top-up cards, a Knowledge Planet community and business contact details commercial section. These are separate from the skill installation and usage flow described above and are not required to use the repository.

HotRepo's take

The repository recorded 167 new stars in the Python daily list on both 2026-11-03 and 2026-11-04, with a best position of 17 on that language list. These two snapshots show sustained daily-list visibility during that period, but there is no all-language daily rank or longer time series to support a broader trend claim. The project's value for readers depends on whether they already use a coding agent and need a packaged academic workflow; the skill index and the documented installation paths give a concrete starting point without requiring the commercial services mentioned in the README.