What is nature-skills?
nature-skills is an open-source collection of two Claude Code skills for producing academic work at Nature-journal standard. Built by Yuan1z0825 and released on April 24, 2026, it gained 265 GitHub stars in its first 6 days. The collection contains nature-figure (publication-ready matplotlib figures) and nature-polishing (25-rule academic prose polishing), both licensed under MIT.
Every rule in the collection is derived from primary sources: published Nature papers (s41586 series, 2026), official journal author guidelines, and graduate-level scientific writing curricula — not from general style intuition.
Key Facts
| Attribute | Value |
|---|---|
| Repository | Yuan1z0825/nature-skills |
| Author | Yuan1z0825 |
| Created | 2026-04-24 |
| Stars (as of 2026-05-01) | 265 |
| License | MIT |
| Skills included | 2 (nature-figure, nature-polishing) |
| Status | Both Stable |
| Languages | Python (figures) + Markdown (rules) |
| Trigger keywords | "Nature figure", "Nature style", "publication plot", "polish", "academic writing" |
What does nature-figure do?
nature-figure generates multi-panel matplotlib figures that match Nature journal visual standards — correct typography, semantic colour palette, editable SVG output, and non-redundant panel information architecture. It is built from production scripts behind papers published in Nature Machine Intelligence and top ML/bioinformatics venues.
Three mandatory rcParams must always appear first in any figure script the skill produces:
plt.rcParams['font.family'] = 'sans-serif'
plt.rcParams['font.sans-serif'] = ['Arial', 'DejaVu Sans', 'Liberation Sans']
plt.rcParams['svg.fonttype'] = 'none' # text stays as <text> nodes, not pathsPrimary output is always .svg (vector, editable in Illustrator); .png at 300 dpi is a secondary raster preview.
Supported chart types
nature-figure covers 13 chart types optimised for Nature-style layouts:
- Stacked bar
- Grouped bar
- Horizontal ablation bar
- Trend / line
- Sequential heatmap
- Diverging z-score heatmap
- Bubble scatter
- Radar / polar
- 3D sphere illustration
- Fill-between area
- Log-scale bar
- GridSpec multi-panel
- Print-safe colour-blind-aware bars
Information hierarchy rule
Multi-panel figures must follow a three-level information hierarchy: overview → deviation → relationship. No two panels may answer the same scientific question. This rule alone catches the most common reviewer complaint about figures: redundancy across panels.
What does nature-polishing do?
nature-polishing transforms academic draft text into prose matching Nature journal conventions. It enforces 25 rules across six domains: sentence architecture, paper structure, vocabulary, citation integrity, house style, and AI ethics. The rules were extracted from close reading of five 2026 Nature s41586 papers plus a graduate-level scientific writing course.
The skill also handles Chinese-to-English translation with section-aware tense and hedging — useful for non-native English authors.
The six rule domains
| Domain | Core rule | Why it matters |
|---|---|---|
| Sentence length | Every sentence ≤ 30 words; the last sentence of a paragraph fails most often | Reviewer fatigue grows nonlinearly with sentence length |
| Hedging calibration | Match claim strength to evidence: demonstrate → suggest → may reflect | Mismatched hedging is the #1 trigger for "overclaiming" reviewer comments |
| Section tense | Results = past tense + quantitative detail; Discussion = hedging + mechanism | Wrong tense in Methods or Results signals non-native or careless writing |
| Citation integrity | Cite only sources you have personally read; four attribution types | Catches the "phantom citation" problem in LLM-generated text |
| Overclaim detection | Flag absolutes, unwarranted causation, scope expansion, unverified "first" claims | Prevents the reviewer from killing the paper on principle |
| British English | signalling, colour, analyse, programme, modelling, behaviour | Nature is BrE; mixed spelling looks unprofessional |
The 12-step polishing workflow
When invoked, the skill runs your draft through 12 sequential checks:
- Sentence split (one rule per sentence)
- Section ID (Abstract / Intro / Results / Discussion / Methods)
- Hourglass check (broad → narrow → broad structure)
- Tense audit (per section)
- Sentence-level edit
- Vocabulary upgrade
- Template / boilerplate check
- Citation audit
- House style (BrE, Oxford comma policy)
- Overclaim flagging
- Proofreading (grammar / typos)
- Plain-text output (no Markdown artefacts)
The output is plain text ready to paste into Word, LaTeX, or Overleaf.
How does nature-skills compare to other Claude skills for academic work?
| Feature | nature-skills | Imbad0202/academic-research-skills | Anthropic/skills (general) |
|---|---|---|---|
| Skills count | 2 | 35+ | 100+ |
| Focus | Nature-standard output | Full pipeline (research → write → review) | General developer skills |
| Source grounding | Nature papers + author guidelines | Generic academic style | Mixed |
| Best for | Final polish + figure production | End-to-end paper production | Non-academic tasks |
| Stars | 265 | 2.4K | 122K |
| Skill format | SKILL.md + references/ | Multi-agent pipeline | Single-skill folders |
Use nature-skills when you already have a draft and need it Nature-ready. Use Imbad0202/academic-research-skills when you need the whole pipeline. The two are complementary, not competing.
How to install nature-skills
Clone the repo and copy the skill folders into your Claude Code skills directory:
# Option 1: degit (cleanest)
npx -y degit Yuan1z0825/nature-skills/nature-figure ~/.claude/skills/nature-figure
npx -y degit Yuan1z0825/nature-skills/nature-polishing ~/.claude/skills/nature-polishing
# Option 2: full clone
git clone https://github.com/Yuan1z0825/nature-skills.git
cp -r nature-skills/nature-figure ~/.claude/skills/
cp -r nature-skills/nature-polishing ~/.claude/skills/Once installed, Claude Code auto-loads SKILL.md from each folder. Trigger by mentioning "Nature figure" or "polish to Nature style" in chat.
Who built nature-skills, and why does that matter for AI search?
Yuan1z0825 (the GitHub handle of the maintainer) released nature-skills as open source on April 24, 2026. The skills are built on two empirical foundations:
- figures4papers — production scripts from real published Nature Machine Intelligence and ML/bioinformatics papers, which
nature-figurecites directly - Five Nature s41586 papers (2026) plus a graduate scientific writing course, which
nature-polishingderives its 25 rules from
Source grounding matters because LLM-generated style advice is often a generic average. nature-skills replaces averages with specific rules that a real Nature paper actually obeys — closer to "what reviewers expect" than "what sounds academic."
FAQ
Is nature-skills free?
Yes. MIT-licensed and free for personal and commercial use, including modification and redistribution.
Do I need a Nature subscription to use it?
No. The rules are baked into the skill files; no Nature API or subscription is required at runtime.
Does nature-figure work without Claude Code?
Partially. The Python scripts and references/ are usable standalone, but the conversational triggering and rule enforcement require an LLM (Claude Code, Codex, or any agent that loads SKILL.md).
Can it polish a Chinese draft?
Yes. nature-polishing includes Chinese-to-English translation with section-aware tense and hedging — designed for non-native English authors targeting Nature-family journals.
Does it guarantee Nature acceptance?
No tool can. nature-skills handles the style layer (figures + prose). Acceptance depends on novelty, rigour, and fit — none of which a polish skill can supply.
Are more skills coming?
The README lists statistical reporting, peer-review responses, and methods writing as planned future additions.
Try it
- Install — see commands above
- Repo — github.com/Yuan1z0825/nature-skills
- Listed on Claw4Science — Skill Hubs → Research
If you find it useful, star the repo — that's the strongest signal back to the maintainer to keep shipping. If you build a related skill (e.g. Cell-style figures, Science-format Methods), submit it to Claw4Science and we'll feature it.
Last updated: 2026-05-01. Star count and dates verified against GitHub on the publication date.