Polish, Review, and Illustrate Papers With nature-skills' 22 Research Skills

A manuscript paragraph that reads fine in your head still has unit inconsistencies, claim drift, and prose that no Nature editor would pass. nature-skills is a roughly 50k-star collection of reusable research skills for AI scholars, organized around SKILL.md: each top-level directory under skills/ is an installable unit, and the README's own emphasis is on verifiable workflows and research outputs you can use directly.

Why This Skill Matters

The Skill Index lists 22 triggerable skills, each tagged Stable, Beta, or Draft. The four Stable ones are nature-figure (a submission-grade Python or R scientific-figure workflow for Nature and high-impact journals), nature-polishing (polish, restructure, or translate academic prose into Nature-style English, with manuscript-wide terminology, unit, precision, and claim-drift checks), nature-ref-verifier (cross-check references across sources and flag author, title, year, volume, issue, and page inconsistencies), and nature-literature-pipeline (multi-source retrieval, six-axis scoring, deep-reading delivery, and local archiving). Twelve more are Beta — among them nature-reader for figure-aware Chinese-English Markdown readers, nature-response for mutually blind reviewer responses and cover letters, and nature-academic-search for multi-source literature search and strict other-citation audits — and six are Draft, including nature-writing and nature-reviewer.

The README's status table defines those labels: Draft means rules are defined but not yet tested on real cases; Beta means tested on examples with possible edge-case issues; Stable means validated on real academic content. A 23rd directory, nature-shared, is a support package read by other skills and is not counted in the index.

Installation

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

npx skills add Yuan1z0825/nature-skills --list

Install the skills your current task needs. 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

Omit --global for a project-local install. The README's install badge names Claude Code, Codex, OpenClaw, OpenCode, and Hermes; for Claude Code specifically, it documents keeping a stable clone and pointing a subagent or slash-command wrapper at the real skills/*/SKILL.md, or running scripts/autoupdate-skills.sh to copy the skills into ~/.claude/skills/. For the complete collection on Codex, it documents scripts/update-codex-skills.sh --pull from a clone, which syncs every top-level skill directory and verifies the copies with diff.

Real Workflow: Polish a Paragraph Into Nature-Style English

Install nature-polishing plus nature-shared, start a new session, and paste the README's own quick-start prompt with your paragraph:

把这段中文改写成 Nature 风格英文，保持学术含义不变。

The skill's purpose line is specific about what comes back: polished, restructured, or translated prose in Nature-style English, with manuscript-wide checks for terminology, units, numerical precision, and claim drift. The README's Chatbox example adds the shape of a fuller request — preserve academic meaning, numerical values, and limits on conclusions, then list the main changes and explain the reasons.

Real Workflow: Simulate Pre-Submission Review

With nature-reviewer installed, ask for a reviewer-perspective assessment:

从 Nature 审稿人视角评估这篇稿件，给出三份互盲 reviewer reports；全部定稿后再综合。

The index describes the output as three mutually blind reviewer reports with tiered Major/Minor comments and manuscript-internal consistency checks. Note the status: nature-reviewer is Draft, so by the README's own definition its rules have not yet been tested on real cases. When the revision email arrives, nature-response (Beta) parses it and produces separate mutually blind reviewer responses, a cover letter, a red-marked manuscript, LaTeX templates, and a revision-package consistency check.

Tips

• Pass the frontmatter name shown by --list to --skill: the nature-proposal-writer directory is currently listed as researchwrite.
• npx skills manages skill files only. Python, R, browser, and MCP runtime dependencies need separate setup — nature-academic-search requires PUBMED_EMAIL, and nature-paper-to-patent has its own requirements.txt.
• Keep the full directory structure when copying by hand: many skills depend on references/, static/, manifest.yaml, scripts, assets, or nature-shared. Do not copy only SKILL.md.
• If you are unsure which skill applies, describe the task naturally; if you know the name, say "use nature-reader" in the prompt.
• Update a global install with npx skills update --global --yes, or one skill with npx skills update nature-reader --global --yes.

When Not to Use This

Six of the 22 skills are Draft — nature-writing, nature-reviewer, nature-data, nature-statistics, nature-experiment-log, and nature-social-science — and the README defines Draft as untested on real cases, so treat their output as a starting point rather than a verdict. Several deliverables are Chinese by design: nature-paper2ppt generates Chinese PPTX decks, nature-paper-trans produces a Chinese image-based PDF, and nature-paper-to-patent drafts Chinese invention patents — if your deliverable is English-only, check each skill's purpose line first. And a single grammar fix does not need a 22-skill collection; a general-purpose polish is lighter.

See the leaderboard for more skills.