Nature Skills – Open-Source AI Research Skill Library, Focused on the Full Paper Workflow

1. What is Nature Skills

Nature Skills is an open-source skill library designed for AI scholars and researchers worldwide, focusing on assisting with the full workflow of writing and submitting papers at the Nature / CNS level. Based on a routing-style SKILL.md architecture, this tool includes 18 independent skills that cover every stage from drafting the initial manuscript, polishing and translating, simulating peer review, responding to revisions, to generating charts, managing references, and converting research into patents. After installing via multiple agents such as Codex or Claude Code, users can describe tasks in natural language, and the system will automatically match and load the corresponding rules, outputting ready-to-submit results such as text, charts, and PPTs. All of its rules are derived from the official guidelines of Nature / CNS journals and published papers, ensuring a high degree of consistency with the writing style and review standards of top-tier journals.

Technical Positioning and Domain: Nature Skills resides at the intersection of natural language processing and AI-assisted research tools, specifically focusing on academic writing automation and optimization of research workflows. It stands out from similar tools by employing a "skill routing" mechanism rather than a single AI chat interface. This approach breaks down paper writing into independently callable and combinable skill modules, each of which is written based on the official journal guidelines, not on the aesthetic preferences of general-purpose language models. This design enhances rule transparency and output reliability in academic writing scenarios.

Development Background: This project was initiated and continuously maintained by open-source developer Yuan1z0825, driven by the repetitive labor and formatting challenges faced by researchers during the paper writing and submission process. Existing AI writing tools often lack deep adherence to specific journal guidelines, and their outputs require extensive manual proofreading. Nature Skills aims to address this by encoding the writing standards, review criteria, and chart requirements of Nature / CNS journals as explicit, executable rules through open-source collaboration, enabling AI agents to directly generate submission-ready outputs and significantly improving research productivity.

Core Value: Nature Skills tackles two major pain points in academic writing: "rule opacity" and "tool fragmentation." Traditionally, researchers have had to use separate tools for writing assistance, reference management, chart creation, and peer review simulation, with limited interoperability between them. Nature Skills integrates all these stages into a single agent skill library, allowing users to simply describe their tasks and have the system automatically route to the appropriate skill and generate the final output. More importantly, each rule is annotated with its source and applicable boundaries, enabling users to trace and verify the rules, thus avoiding the uncertainties associated with black-box models.

Technical Features: Its core technical advantage lies in the routing-based loading architecture, which dynamically loads the required rule segments based on the task content, paper sections, and target journal, thereby avoiding unnecessary resource consumption from the entire skill library. Additionally, the rules are entirely derived from the official guidelines of Nature / CNS journals and published papers, rather than general writing advice, ensuring that the output style aligns closely with the requirements of top-tier journals. Furthermore, Nature Skills supports compatibility with multiple agents (such as Codex, Claude Code, OpenClaw, etc.) and allows the community to freely contribute new skills, offering excellent scalability.

2. Key Features

• Paper Writing: Draft manuscripts in the style of Nature by chapter, supporting the full workflow from research plan to final submission. The system automatically generates sections such as abstracts, introductions, methods, and discussions based on the user's research data and figures, reconstructing the argumentation logic. The polishing and translation feature optimizes the language of manuscripts for non-native authors while maintaining academic rigor.

• Review Revisions: Simulate three blinded Nature reviewers and generate structured review reports with Major/Minor categorized feedback. After receiving a revision email from the journal, the system can automatically parse the review comments and independently generate point-by-point responses, cover letters, and highlighted sections in the revised manuscript for each reviewer, significantly shortening the revision cycle.

• Figure and Table Statistics: Generate scientific figures and tables that meet Nature submission requirements, supporting output via Python/R scripts. The system can also review statistical reports to ensure key indicators such as p-values, replication counts, and effect sizes are compliant, guaranteeing that data presentation meets the journal's statistical standards.

• Literature Management: Full-chain management covering search, citation verification, full-text download, and automated literature recommendations. The system automatically checks if the citation format matches the target journal's requirements and pushes the latest relevant literature based on the user's research topic.

• In-depth Reading and Reporting: Generate bilingual (Chinese/English) readers, structured in-depth reading cards, and PPTs directly suitable for group meetings. After a user uploads a paper, the system automatically extracts core contributions, methodological processes, and experimental results, generating multilingual interpretations to facilitate cross-team communication.

• Support for Commercialization: Handle data availability statements and experimental log records, and support converting papers into patent drafts. This feature is especially useful for research teams with commercialization needs, enabling the rapid extraction of patentable technical solutions from academic achievements.

3. How to Use

1. Environment Requirements: Ensure that Node.js version 18+ is installed locally. It is recommended to use the LTS version for stability. The operating system supports Windows, macOS, and Linux. You need to have compatible AI Agent clients such as Codex, Claude Code, or OpenClaw pre-installed.

2. View Available Skills: Run the command npx skills add Yuan1z0825/nature-skills --list in the terminal. The system will list the names and functional descriptions of all 18 skills. Users can choose to install all or some of the skills based on their needs.

3. Install the Skill Package Globally: Execute npx skills add Yuan1z0825/nature-skills --global --agent codex --skill '**' --yes --copy. This command will deploy all the skills into the Codex Agent and enable the automatic copy output feature. If using Claude Code or another Agent, replace --agent codex with the corresponding name.

4. Restart the Agent Session: After installation, you must restart the Agent client for the skills to take effect. Some Agents may require explicitly loading the skill package in the configuration file; refer to the respective Agent's documentation for details.

5. Describe the Research Task: Directly describe your requirements in natural language within the Agent's conversation interface, such as "Polish this abstract into a Nature style" or "Simulate three reviewers evaluating my manuscript." The system will automatically parse the task type and match the corresponding skill.

6. Obtain Ready-to-Use Outputs: The Agent will load the relevant skill based on routing rules and output ready-to-paste text, Python/R chart code, PPT files, or cover letters. Users can directly use these outputs for manuscript submission or group meeting presentations.

Notes: Ensure a stable internet connection during the first installation, as the npx command will automatically download the latest version. If permission errors occur during installation, try adding sudo before the command (for Linux/macOS) or running the terminal as an administrator (for Windows). It is recommended to regularly execute the update command to obtain the latest skill rules.

4. Pros and Cons Analysis

5. Comparative Analysis with Similar Tools

Selection Recommendations: For research teams submitting to Nature or its sub-journals, Nature Skills is the optimal choice, as its rules are directly written based on the official guidelines of these journals, ensuring high consistency with writing styles and review standards, and it is completely free. If the team prioritizes cloud collaboration and low learning curve, Paperguide offers a more user-friendly SaaS experience, but it requires payment and lacks rule transparency. For general writing assistance, ChatGPT can handle various writing styles, but users must provide journal guidelines themselves and repeatedly adjust Prompts, resulting in lower efficiency. SciSpace, on the other hand, is better suited for literature research and quick comprehension, but is not appropriate for full paper writing and submission processes.

6. Editor's Summary

Nature Skills stands out with its distinctive approach to technological innovation: it no longer treats AI-assisted scientific writing as a general conversational task, but instead breaks down the paper writing process into independently executable skill modules through a routing-based SKILL.md architecture. Each module is explicitly encoded according to the official guidelines from Nature / CNS. This "rule-driven + Agent routing" design philosophy offers significant advantages over traditional end-to-end large model dialogues in terms of output controllability, explainability, and domain adaptability. Its transparency in rule definitions is particularly commendable — each rule is annotated with its source and applicable boundaries, allowing users to trace back to specific journal guidelines or published papers. This level of transparency is extremely rare in academic writing tools.

In terms of practical value, Nature Skills covers the entire workflow from drafting a paper to resubmitting after review, including peer review simulation, figure and table statistical checks, literature management, and patent conversion — all high-frequency needs. It can significantly reduce the time researchers spend on repetitive tasks such as formatting adjustments, language polishing, and response to reviewers. For researchers preparing to submit to Nature series journals, this tool can serve as an efficient assistant, helping ensure that the manuscript meets the editorial expectations in terms of format and style.

In terms of target users, Nature Skills is most suitable for researchers with a programming background (familiar with command line and Node.js), especially those who frequently submit to Nature / CNS series journals. For users unfamiliar with command line operations, there may be a certain learning curve, but the project documentation and community support can help reduce the difficulty of getting started. Additionally, the tool's reliance on AI Agents means users also need to be proficient in using the corresponding Agents.

Looking ahead, the open-source architecture and modular design of Nature Skills provide a solid foundation for community collaboration. As more researchers contribute skill rules tailored to different journals (such as Science, Cell, etc.), its coverage is expected to expand further. Moreover, its design philosophy has been adopted by Google DeepMind, which launched Science Skills, indicating that this routing-based skill library approach has already gained recognition within top-tier research institutions. In the future, if it can reduce its dependency on local environments (e.g., by providing Docker images or cloud deployment options), it has the potential to attract a much broader user base.

7. Application Scenarios

• Nature-style paper draft creation: After researchers complete their experiments and data collection, they can input core figures and key results into the Agent. The system automatically generates sections such as the abstract, introduction, methods, results, and discussion in the style of Nature, reconstructing the argumentation logic. Users only need to conduct a scientific review of the content to quickly obtain a draft that conforms to the journal's formatting standards.

• Pre-submission peer review simulation: Before formally submitting a manuscript, users can request the Agent to simulate three blinded Nature reviewers. The system will generate review reports from each reviewer, evaluating the manuscript on dimensions such as innovation, methodological rigor, data completeness, and argumentation logic. The system provides Major/Minor categorized feedback, helping authors identify weaknesses in their manuscript in advance and make targeted revisions.

• Point-by-point response to revision comments: After receiving a revision email from a journal, users can paste the review comments into the Agent. The system automatically parses each reviewer's questions and independently generates point-by-point responses, a cover letter, and a highlighted list of all changes made in the revised manuscript. This scenario significantly reduces the manual workload during the revision phase.

• Manuscript-level research figures and statistical review: Users provide raw data and analysis requirements, and the Agent generates Python/R plotting scripts that meet Nature's layout standards. At the same time, the system automatically checks whether p-values are fully labeled, whether the sample size meets requirements, and whether effect sizes are reported in the statistical report, ensuring that the figures and statistical sections comply with the journal's guidelines.

• Full lifecycle literature management and in-depth reading reports: Users only need to provide the paper title or DOI, and the system automatically retrieves the full text, verifies citation information, generates bilingual (Chinese-English) reader versions and structured in-depth reading cards, and converts them into a Chinese conference presentation PPT with one click. For research groups that regularly conduct literature reports, this feature can significantly improve efficiency.

8. FAQ

Q: Does Nature Skills support Chinese paper writing?
A: The current rules are primarily designed for writing English Nature/CNS papers. However, the polishing and translation feature supports translating Chinese abstracts into English, and the in-depth reading report function can generate bilingual (Chinese-English) reader documents. For submissions to purely Chinese journals, it is recommended to use this tool in conjunction with other writing tools.

Q: What should I do if the terminal shows that the "npx skills add" command is not found during installation?
A: Please confirm that Node.js 18+ is correctly installed and that the npm version is 9+. You can check the versions by running node -v and npm -v in the terminal. If the command is still unavailable, try installing npx globally with npm install -g npx.

Q: Can it run on a computer without a GPU?
A: Nature Skills itself does not perform model inference; it relies on the user-installed Agent (such as Codex, Claude Code) to call the underlying large model. Therefore, a local GPU is not required. However, the Agent client may need an internet connection to access cloud-based models.

Q: Can the skill rules be customized and modified?
A: Yes. All skills are stored locally in the form of SKILL.md files, and users can directly edit these files to adjust the rules, add new journal guidelines, or modify the output format. After making changes, simply restart the Agent for the updates to take effect.

Q: What are the advantages of Nature Skills compared to other scientific writing tools?
A: The main advantages are rule transparency (each rule is labeled with its source and applicable boundaries), a full-process closed loop (covering everything from writing to revisions in one place), and routing-based loading (rules are loaded on demand to avoid redundancy). In addition, it is completely open-source and free, allowing the community to freely contribute and expand upon it.

Q: Does it support Agents other than Codex and Claude Code?
A: Currently, the official support includes Codex, Claude Code, and OpenClaw. For other clients that are compatible with the Open Agent protocol, it is theoretically possible to adapt them by modifying the configuration, but users will need to test this themselves.

9. Project Links

• GitHub Repository: https://github.com/Yuan1z0825/nature-skills (Official main repository, containing complete source code, installation guide, and skill documentation)