Nature Skills: A Skill Pack That Turns Research Agents Into Publication-Focused Assistants

What Problem This Skill Pack Solves

Research agents like Codex and Claude Code are good at following instructions, but they are not born with domain-specific procedures. A generic prompt like 'summarize this paper' produces an ad hoc result. Nature Skills addresses that by providing a collection of reusable skill packages, each organized around a SKILL.md file. These packages encode workflows for academic writing, literature review, scientific figures, and reproducible outputs. The intended user is a researcher or research engineer who wants an agent to produce something close to a Nature-style manuscript or a submission-ready figure, not just a rough draft. The README gives example prompts such as 'turn this PDF into a Chinese-English Markdown reader' or 'draft a Nature-style abstract and introduction from these results.' The project assumes you already have an agent runtime; it does not provide one.

How the Skills Are Structured and Installed

Each top-level directory under skills/ is an installable unit, named with a nature- prefix. A shared package, nature-shared, holds supporting references that other skills read, and it is not meant to be triggered independently. The installation path goes through the npx skills CLI, which requires Node.js 18 or newer. You first list available skills with 'npx skills add Yuan1z0825/nature-skills --list'. To install everything globally for Codex, you run 'npx skills add Yuan1z0825/nature-skills --global --agent codex --skill '*' --yes --copy'. For a single skill, you omit --global and specify the skill name. The README notes that the frontmatter name may differ from the directory name; for example, nature-proposal-writer currently displays as researchwrite. That mismatch is a real usability trap: you must read the listing output, not guess from the directory name.

Claude Code Setup Requires a Local Clone, Not a Script

Claude Code cannot use the same update script that syncs skills to Codex's ~/.codex/skills/ directory. The README explicitly says that script is Codex-only. Instead, the recommended approach is to clone the repository to a stable path, such as ~/ai-skills/nature-skills, and then create subagent or slash command wrappers that point to the actual SKILL.md files. The README provides a full example for nature-reader, where you write a Markdown file in ~/.claude/agents/ that instructs the agent to read the governing SKILL.md and supporting files only when needed. Updates are then handled by a simple 'git pull' in that clone directory. This design keeps the skill directory structure intact, which matters because skills reference references/, static/, manifest.yaml, and scripts. It also means you are responsible for keeping the clone updated; the optional autoupdate script exists but is not described in detail in the README.

What the Skills Actually Cover

The README lists several skills by name: nature-reader, nature-paper2ppt, nature-polishing, nature-writing, nature-reviewer, nature-response, and nature-figure. Each maps to a concrete task. Nature-reader handles Chinese-English paper reading with figure-aware translation. Nature-paper2ppt generates a Chinese seminar presentation from a paper, preserving key figures and source annotations. Nature-polishing rewrites Chinese paragraphs into Nature-style English. Nature-writing drafts abstracts, introductions, and discussions. Nature-reviewer simulates peer review by producing three blinded reviewer reports from a Nature reviewer perspective. Nature-response writes point-by-point replies to review comments. Nature-figure generates submission-ready scientific figures or schematic drafts. The prompts in the README are direct and usable, but the README does not show the internal content of any SKILL.md file. So you cannot verify from this README how deep these procedures go or how well they handle edge cases.

A Genuine Limitation: It Is a Set of Instructions, Not a Tool

The most important constraint is that nature-skills is not a standalone application. It has no runtime, no GUI, and no built-in execution engine. You must have an agent that can read SKILL.md files and follow them. The README warns that Python, R, browser, or MCP dependencies must be configured separately. This means a skill like nature-figure may require specific plotting libraries that are not installed by the skill itself. If your agent environment lacks those dependencies, the skill will fail or produce incomplete output. Another limitation is the maintenance state: the repository has no recent releases listed, and the README contains promotional sections about the founder, a paid knowledge planet, and a recharge card website. That mix of commercial promotion and technical documentation can make it hard to judge the project's maturity. You should treat the skills as starting points, not as verified, production-grade workflows.

The Real Alternative: Write Your Own SKILL.md Files

The direct alternative is to write your own SKILL.md files tailored to your specific research workflow. The nature-skills README itself suggests this: you can give the GitHub URL to Codex and it will learn the design philosophy to create new skills. That is a meaningful difference. With nature-skills, you get a pre-packaged set of procedures that may not match your exact field, journal requirements, or figure style. Writing your own skill gives you full control over the steps, references, and scripts. The trade-off is effort: you must define the workflow, test it, and maintain it. Nature-skills saves you that initial design work, but you inherit someone else's assumptions. If your work involves niche methods or specific journal formats, a custom skill will likely serve you better than adapting a generic one.

Maintenance and License Considerations

The project is licensed under Apache-2.0, which allows you to use, modify, and redistribute the skills, including for commercial purposes, as long as you preserve the license notice. That is a permissive license, so there is no obligation to share your modifications. Maintenance is a real concern. The README shows an active community with a knowledge planet and video tutorials, but it gives no release history. The only update mechanism described is 'npx skills update' for Codex and 'git pull' for Claude Code clones. That means you are responsible for tracking upstream changes. If the maintainers stop updating, the skills may drift from current agent capabilities or library versions. Before adopting, you should check the repository's commit activity and the last modification date of the specific skills you plan to use.

Editorial conclusion

Adopt Nature Skills if you are a researcher or engineer who already uses Codex or Claude Code and wants ready-made, publication-oriented procedures for reading papers, drafting manuscripts, and generating figures. Do not adopt it if you need a standalone tool with its own runtime, because it is only a set of instruction files that require an external agent. Before installing, verify that the specific skills you need are maintained and that their references and scripts match your local environment, since the repository has no recent release history and the README mixes promotional content with technical instructions.

Official sources

1. Official README
2. Project repository