Yuan1z0825/nature-skills

nature-skills collects reusable research skills for AI scholars worldwide, with an emphasis on real problem solving, verifiable workflows, and research outputs that can be used directly.

Table of Contents

• 1. Project Founder and Operations
  • 1.1 Founder Introduction
  • 1.2 Knowledge Planet
  • 1.3 Repository Store
  • 1.4 Business Cooperation
• 2. Nature Skills Core Developers
• 3. Project Philosophy and Community
• 4. Quick Start
• 5. Installation
  • 5.1 npx skills Installation
  • 5.2 Claude Code Installation
  • 5.3 Codex Installation
  • 5.4 Chatbox Installation and Usage
  • 5.5 Other Agent Scenarios
• 6. Skill Index
• 7. Contribution and Development
• 8. Star History

1. Project Founder and Operations

1.1 Founder Introduction

Hello, I am Yizhe Yuan, the founder of nature-skills. Thank you for following this project. We have published many video tutorials on Douyin; search by topic name to find them, and I sincerely hope they help with real research work.

1.2 Knowledge Planet

Knowledge Planet name: Nature Skills and the Philosophy Behind It.

1.3 Nature AI GPT / Claude Top-ups and Ready-Made Accounts

We carefully screen service channels and provide legitimate top-up channels and support. Visit Nature AI Recharge Card Store:

Nature AI Recharge Card Store https://apiciyuan.top/

WeChat Support Scan to add

1.4 Business Cooperation

For business cooperation, please contact us at natureskills2026@outlook.com.

2. Nature Skills Core Developers

3. Project Philosophy and Community

3.1 Some Personal Views

• Recently, I noticed that the Nature Skills design has drawn attention from Google DeepMind and has been referenced by them. They drew on its citation system, script ideas, and skill-design philosophy to launch Science Skills. To be honest, this makes me pleased: when leading international AI institutions begin to draw inspiration from our work, it means original ideas from Chinese developers are being seen by the world. This is not a feeling of loss from being copied, but a sign of Chinese strength taking root in open source and naturally growing outward.
• The focus of our skill design has never been to require every user to fully master the whole philosophy. The point is that the philosophy itself can be understood and reused by machines. If you want to create a new skill or adapt this system to your own field, you can give the Nature Skills GitHub repository to Codex and let it learn the design pattern, then help you create or modify a skill. This is how ideas become operational rather than remaining oral explanations.
• The real value of Nature Skills may not be limited to any individual module. It may be that the project has quietly opened a new door: many people realize for the first time that Codex or other agents can operate a local computer for research. I have been fortunate to witness and accompany many people through this shift in research workflow. When they say, "so research can be done this way," that break in understanding and liberation of thought matters more to me than the skills themselves. This is not merely the success of a tool, but the beginning of a new way of thinking spreading among people.
• In practice, almost every useful tool can be distilled into a standardized process, and standardized processes can be packaged as reusable skills.

3.2 Tutorials and Community

Follow Douyin for video tutorials

Agent Research Community

Yuan's personal WeChat

4. Quick Start

After installation, you can give the agent a paper, paragraph, reviewer letter, or task description directly. These prompts are ready to copy:

If you are unsure which skill to use, describe the task naturally. If you already know the skill name, explicitly say "use nature-reader" or "use nature-response" in the prompt.

5. Installation

nature-skills is a collection of reusable skill packages organized around SKILL.md. Each top-level skill directory under skills/ is an installable unit, such as nature-*; nature-shared is an installable support package read by other skills.

5.1 npx skills Installation

Install Node.js 18 or later first. The CLI does not need to be installed globally. List the skill names available in this repository:

npx skills add Yuan1z0825/nature-skills --list

Start with the skills needed for the current task and include their shared dependencies. Omit --global for project-local installation; add it when the skill should be available across projects. For example:

npx skills add Yuan1z0825/nature-skills --agent codex --skill nature-figure --yes --copy

When installing nature-reader, nature-paper2ppt, nature-polishing, or nature-writing alone, select the shared support package as well:

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

If you need the complete collection, install every skill globally for Codex. This selection includes nature-shared:

npx skills add Yuan1z0825/nature-skills --global --agent codex --skill '*' --yes --copy

To install all skills for every agent supported by the CLI:

npx skills add Yuan1z0825/nature-skills --all

Verify the global Codex installation and update it later:

npx skills list --global --agent codex --json
npx skills update --global --yes

Update one skill, or update only the current project's skills:

npx skills update nature-reader --global --yes
npx skills update --project --yes

Pass the frontmatter name displayed by --list to --skill; for example, the nature-proposal-writer directory is currently listed as researchwrite. npx skills manages skill files only. Optional Python, R, browser, and MCP runtime dependencies still need the separate setup described below.

5.2 Claude Code Installation

Claude Code cannot use scripts/update-codex-skills.sh directly because that script only syncs skills into Codex's ~/.codex/skills/. For Claude Code, keep a stable local clone and create a subagent or slash command wrapper that points to the real skills/*/SKILL.md. This preserves the skill directory structure and lets the workflow keep using references/, static/, manifest.yaml, scripts, assets, and skills/nature-shared/.

If Claude Code is not installed yet:

npm install -g @anthropic-ai/claude-code
claude

Clone the repository to a stable path:

mkdir -p ~/ai-skills
cd ~/ai-skills
git clone https://github.com/Yuan1z0825/nature-skills.git

Recommended method: create a Claude Code subagent wrapper for the skills you use often. Example for nature-reader:

mkdir -p ~/.claude/agents
cat > ~/.claude/agents/nature-reader.md <<'EOF'
---
name: nature-reader
description: Use for Chinese-English paper reading, figure-aware translation, and source-grounded paper notes.
---

When invoked, first read `~/ai-skills/nature-skills/skills/nature-reader/SKILL.md` and follow it as the governing workflow.
Read supporting files from `~/ai-skills/nature-skills/skills/nature-reader/` and `~/ai-skills/nature-skills/skills/nature-shared/` only when needed.
Do not replace this skill with a generic paper-reading response.
EOF

Then start a new Claude Code session and ask for the subagent explicitly:

Use the nature-reader subagent to turn this paper into a Chinese-English Markdown reader.

If you prefer a slash command, create a command wrapper instead:

mkdir -p ~/.claude/commands
cat > ~/.claude/commands/nature-reader.md <<'EOF'
Read `~/ai-skills/nature-skills/skills/nature-reader/SKILL.md` first and follow it strictly.
Read directly needed supporting files from `~/ai-skills/nature-skills/skills/nature-reader/` and `~/ai-skills/nature-skills/skills/nature-shared/`.

$ARGUMENTS
EOF

Use it inside Claude Code:

/nature-reader Turn this paper into a full Chinese-English side-by-side Markdown reader.

To install other skills, replace nature-reader with the target directory name, such as nature-polishing, nature-writing, nature-reviewer, nature-response, or nature-figure. To update later:

cd ~/ai-skills/nature-skills
git pull

As long as the wrapper still points to this stable clone path, no repeated file copy is needed.

Auto-Update (Optional)

If you want Claude Code to pull upstream updates automatically on every session start, use scripts/autoupdate-skills.sh together with a SessionStart hook.

This approach copies the skills straight into ~/.claude/skills/ (Claude Code auto-discovers that directory and loads each skill by its directory name) instead of using the wrappers above. Pick whichever one you prefer.

Keep a dedicated stable clone (used only to sync skills — don't make dev commits inside it):

mkdir -p ~/ai-skills
git clone https://github.com/Yuan1z0825/nature-skills.git ~/ai-skills/nature-skills

Install once, copying the skills into Claude Code's skills directory:

~/ai-skills/nature-skills/scripts/autoupdate-skills.sh --force

Then add a SessionStart hook to ~/.claude/settings.json (merge this entry into an existing hooks block rather than replacing it):

{
  "hooks": {
    "SessionStart": [
      {
        "hooks": [
          {
            "type": "command",
            "command": "$HOME/ai-skills/nature-skills/scripts/autoupdate-skills.sh",
            "async": true,
            "timeout": 120
          }
        ]
      }
    ]
  }
}

async: true runs it in the background so it never blocks startup. The script is safe to run constantly: it skips the network if it already checked within the last hour, silently skips when offline or the pull fails (exit 0, never stalling a session), re-syncs when the upstream HEAD changes or the installed copy drifts, and refuses to fast-forward a clone that has uncommitted changes. New skills take effect on the next session (the current one already loaded its skills). Each destination has a separate log at ~/.local/state/nature-skills/<destination-id>/autoupdate.log.

The destination and check interval are both configurable:

# Defaults to ~/.claude/skills; use --dest for another location, e.g. Codex:
~/ai-skills/nature-skills/scripts/autoupdate-skills.sh --dest ~/.codex/skills
# Check the network at most once per hour; still verify the local install each run:
~/ai-skills/nature-skills/scripts/autoupdate-skills.sh --throttle 3600

5.3 Codex Installation

Use the repository script when installing or updating the complete skill collection. It syncs every top-level skill directory under skills/ and verifies the copied contents with diff. It does not overwrite unrelated Codex skills. For a selected subset, use the npx skills installation above.

git clone https://github.com/Yuan1z0825/nature-skills.git
cd nature-skills
scripts/update-codex-skills.sh --pull

If you already have a clone:

cd nature-skills
scripts/update-codex-skills.sh --pull

Verify that the current Codex installation matches this checkout:

scripts/update-codex-skills.sh --check

If you use this script for long-term updates and want to remove directories that were previously managed by this installer but no longer exist upstream:

scripts/update-codex-skills.sh --pull --prune

--prune only removes directories recorded by this installer. It will not guess or delete unrelated skills.

You can also ask Codex to install the repository for you:

Install Codex skills from this repository:
https://github.com/Yuan1z0825/nature-skills.git

Clone the repository, run scripts/update-codex-skills.sh --pull, and then run
scripts/update-codex-skills.sh --check to verify the installation. Keep complete
skill directories under skills/; do not copy only SKILL.md.

To install only one skill, specify the skill name:

Install only nature-reader from this repository:
https://github.com/Yuan1z0825/nature-skills.git

If the skill needs shared files, install skills/nature-shared as well.

Key rule: keep the full directory structure. Many skills depend on references/, static/, manifest.yaml, scripts, assets, or shared files.

The installer does not install Python dependencies automatically. Install them only when you need the corresponding scripts or MCP services:

python -m pip install -r skills/nature-paper-to-patent/requirements.txt
python -m pip install -r skills/nature-paper-to-patent/scripts/disclosure/requirements-cnipa.txt  # optional CNIPA published-patent search
python -m pip install -r skills/nature-academic-search/mcp-server/requirements.txt

If you enable nature-paper-to-patent CNIPA published-patent search, also run python -m playwright install chromium.

nature-academic-search also requires PUBMED_EMAIL. Optional Scopus, ScienceDirect, and other provider credentials should be configured locally and must not be committed to the repository.

After installation, start a new Codex session and describe your task naturally, for example:

Turn this paper into a full Chinese-English side-by-side Markdown reader.

Create a Chinese PPT deck from this paper.

For OpenClaw, OpenCode, Hermes, and other open-source agent frameworks, see the OpenClaw / OpenCode / Hermes integration guide.

Auto-Update (Optional)

Codex supports a global SessionStart hook. With a dedicated clone, it can check for updates whenever a Codex session starts or resumes and sync new versions into ~/.codex/skills/.

Create the dedicated clone and perform the initial sync:

mkdir -p ~/.codex
git clone https://github.com/Yuan1z0825/nature-skills.git ~/.codex/.nature-skills-src
~/.codex/.nature-skills-src/scripts/autoupdate-skills.sh \
  --dest ~/.codex/skills --force

Then create or merge ~/.codex/hooks.json:

{
  "hooks": {
    "SessionStart": [
      {
        "matcher": "startup|resume",
        "hooks": [
          {
            "type": "command",
            "command": "/bin/bash \"$HOME/.codex/.nature-skills-src/scripts/autoupdate-skills.sh\" --dest \"$HOME/.codex/skills\"",
            "timeout": 120,
            "statusMessage": "Checking Nature Skills updates"
          }
        ]
      }
    ]
  }
}

If hooks.json already contains other hooks, merge the SessionStart entry instead of replacing the file. After enabling or changing the hook, run /hooks in Codex to review and trust it. Codex currently runs command hooks synchronously, so this setup relies on the script's built-in 1-hour throttle, 60-second network guard, and offline-safe exit to avoid repeated network checks or blocking startup when an update cannot be fetched. Even when upstream is unchanged, the script verifies the destination and repairs detected drift.

Each destination has a separate log at ~/.local/state/nature-skills/<destination-id>/autoupdate.log. Newly fetched skills normally take full effect in the next session.

5.4 Chatbox Installation and Usage

Chatbox desktop provides a graphical Skills manager. Use a version with a Settings → Skills entry; this section applies to the desktop app.

Install from GitHub

1. Open Settings → Skills in Chatbox and click Install from GitHub.
2. Paste https://github.com/Yuan1z0825/nature-skills and click Scan.
3. Select the skills you need and click Install Selected. For a first try, choose nature-polishing and nature-shared; use Select all if you want the full collection.
4. Check the result under Installed Skills and make sure the skills you want to use are enabled.

nature-shared is a shared support package. Install it alongside skills that reference it, including nature-polishing, nature-writing, nature-response, nature-reader, and nature-paper2ppt; do not invoke it as a standalone task. Installing one skill does not automatically install other skill dependencies. The list may show frontmatter names: for example, nature-proposal-writer appears as researchwrite.

Start using the skills

Create a conversation, select a model that supports tool calling, and enable Agent Mode. Paste your text into the conversation and name the skill explicitly. For example:

Use nature-polishing to rewrite the following Chinese paragraph in Nature-style English.
Preserve its academic meaning, numerical values, and limits on conclusions.
List the main changes and explain the reasons:

[Paste your manuscript paragraph here]

For paper reading, install nature-reader and nature-shared, provide the paper file or an accessible local path, and ask:

Use nature-reader to turn this paper into a Chinese-English Markdown reader
with aligned figures and text. Preserve equations and source anchors,
and save the deliverables to my specified output directory.

Check the skill-loading and tool-execution records in the conversation to confirm that the intended skill was loaded. If it does not trigger, check the skill toggle, Agent Mode, and whether the selected model supports tool calling.

Runtime dependencies and updates

Installing skills adds instructions and supporting files. Configure Python/R, PDF/PPTX tools, browsers, and MCP services according to each skill's documentation. For local files or scripts, grant the required directory and command access when Chatbox prompts you. External services such as image generation require their own credentials. A successful skill installation does not mean these external capabilities are configured.

Use Check Update in a skill's action menu and follow the prompts to update it; also check nature-shared when it is a dependency. If GitHub scanning or downloading fails, download and extract this repository through Code → Download ZIP, click Open Skills Folder in Chatbox, and place the complete skill directories and nature-shared alongside one another in that folder. Refresh the skill list and enable them. Preserve references/, static/, scripts, and assets; do not copy only SKILL.md. For manual copies, match the directory name to the name in SKILL.md (for example, name the nature-proposal-writer directory researchwrite). Manually copied skills require manual updates.

5.5 Other Agent Scenarios

For OpenClaw, OpenCode, and Hermes, see the dedicated integration guide.

For other agents, keep a stable repository clone and create a lightweight subagent, slash command, or custom prompt wrapper that points to the real skills/*/SKILL.md files. Preserve skills/nature-shared/.

For manual or other-agent use:

1. Copy complete skill directories into your prompt library or project.
2. Preserve SKILL.md, manifest.yaml, static/, references/, scripts, assets, and required skills/nature-shared/ files.
3. If the target agent has its own format requirements, adjust the frontmatter and body structure.

6. Skill Index

The current skills/ directory contains the following triggerable skills. skills/nature-shared/ is shared content and is not counted in the skill index. Click a skill name or the "Details" link to open its dedicated documentation page.

7. Contribution and Development

7.1 Shared Design Principles

1. Prefer primary sources: rules should be grounded in published Nature content, official journal guidance, or explicit local sources rather than generic taste.
2. Make rules explicit: explain the reason behind each rule instead of giving unsupported assertions.
3. Respect section and task context: writing, figures, citations, and responses depend on the manuscript section and task.
4. Output first: every skill should produce something directly usable, such as paste-ready text, .svg, .pptx, .docx, or concrete instructions.
5. Keep skills extensible: each skill should be self-contained, and adding a new skill should not require modifying existing skills.

7.2 Repository Layout

skills/
├── nature-shared/              # keep this when skills reference ../nature-shared
├── nature-<topic>/
│   ├── README.md
│   ├── README_EN.md
│   ├── SKILL.md
│   ├── manifest.yaml     # present in router-style skills
│   ├── static/           # present in router-style skills
│   └── references/...
└── nature-proposal-writer/
    ├── README.md
    ├── README_EN.md
    ├── SKILL.md
    ├── scripts/...
    ├── templates/...
    └── references/...

7.3 Adding a Skill

When adding a skill to this repository, follow this process.

1. Create the skill directory

skills/nature-<topic>/

2. Add required files

3. Write the Chinese and English READMEs

Every new skill must include both README.md and README_EN.md. The README is a human-facing entry page, not a duplicate of SKILL.md and not an installation manual. Its job is to help users decide within 30 seconds whether the skill fits their task, what they need to provide, what it will output, and where its boundaries are.

Basic rules:

• The Chinese and English README files must be one-to-one mirrors: same heading count, same order, and same information points. Do not let the English page become a separate template.
• Start with the skill name, language switch link, and one positioning sentence.
• Use the base structure below by default; add optional sections only when they are genuinely needed.
• Do not repeat repository installation instructions, author bios, changelogs, development history, full file trees, or long internal implementation details inside a single skill README.
• Put complex rules, API parameters, long tutorials, script explanations, and template indexes in references/, static/, scripts/, or SKILL.md. Keep only navigational pointers in the README.
• If the skill has visual assets, a small preview table is fine; do not turn the README into a large gallery or long technical manual.

Chinese README base structure:

# `nature-<topic>` 技能

[English](README_EN.md)

一句话说明这个技能的定位、主要任务和使用边界。

## 适合用它做什么
## 典型请求
## 你需要提供
## 产出
## 边界
## 相关技能

The English README must mirror it as:

# `nature-<topic>` Skill

[中文说明](README.md)

One sentence describing the skill's role, main task, and usage boundary.

## What To Use It For
## Typical Requests
## What You Need To Provide
## Outputs
## Boundaries
## Related Skills

Optional sections must be inserted in both languages and in the same order. Common optional sections:

Before submitting, run at least these README checks:

python scripts/validate-readmes.py
git diff --check
for d in skills/nature-*; do
  [ -f "$d/README.md" ] && [ -f "$d/README_EN.md" ] || continue
  rg -q '^\[English\]\(README_EN\.md\)$' "$d/README.md"
  rg -q '^\[中文说明\]\(README\.md\)$' "$d/README_EN.md"
  test "$(rg -c '^## ' "$d/README.md")" = "$(rg -c '^## ' "$d/README_EN.md")"
done

4. Record a usage tutorial

When submitting a PR, please also record a short usage tutorial explaining what problem the skill solves, how to trigger it, what inputs it needs, and what outputs it produces. Add the video, screencast link, or public tutorial URL to the PR description.

5. Configure SKILL.md frontmatter

---
name: nature-<topic>
description: >-
  One sentence explaining what the skill does, when it should trigger, primary
  outputs, and core use cases.
---

6. Update the skill index

After adding a skill, update the Skill Index table:

| [`nature-<topic>`](skills/nature-<topic>/README_EN.md) | Draft / Stable | One-sentence purpose | Trigger terms | [Details](skills/nature-<topic>/README_EN.md) |

7. Set the status label

8. Star History