How to use the nature-writing agent skill: install and draft

What an agent skill is and what nature-writing makes the agent do

An agent skill is a folder of instructions and reference files that an AI agent loads when a request matches the skill's description. The agent does not gain new tools from it; it gets better guidance for a particular kind of task.

nature-writing covers drafting or restructuring scientific manuscript arguments, sections and initial-submission materials from evidence the author supplies. The skill detects the paper type, target section, language and journal (for example Nature or Nature Communications), then loads matching guidance on argument structure, abstracts, introductions, results, discussion and submission materials.

The skill is told to flag missing evidence with placeholders rather than invent content, and to list gaps under a heading for missing inputs.

• Titles, abstracts, introductions, results narratives, discussions, conclusions and significance paragraphs
• Claim-evidence organisation from figures and data
• Turning Chinese research notes into English manuscript paragraphs (language setting zh-to-en)
• Reordering Results or Discussion at section level, not just sentence polishing
• Initial-submission materials such as cover letter, title page, highlights, author contributions, data and code availability, and a submission-readiness checklist

Install the nature-writing skill from the ZIP

Download the ZIP of the skill from this page. The skill lives in the folder skills/nature-writing of the Yuan1z0825/nature-skills project, and its licence is Apache-2.0.

For a personal install, unzip the folder into ~/.claude/skills/ so the skill sits at ~/.claude/skills/nature-writing. To use it only inside one project, unzip it into that project's .claude/skills/ folder instead.

If you use the Claude apps, you can instead upload the skill under Settings → Capabilities → Skills. Use whichever route matches how you work; the skill content is the same.

• Personal install: ~/.claude/skills/nature-writing
• Project-only install: the project's .claude/skills/nature-writing
• Desktop or web app: Settings → Capabilities → Skills, then upload

Check that the skill folder is complete

After unzipping, the skill folder should contain SKILL.md at its top level, along with manifest.yaml, the README files, the references folder and the agents folder. SKILL.md is the entry point the agent reads first.

The skill's router reads manifest.yaml to learn its axes (task, paper_type, section, language, journal) and the file each value maps to. If those files are missing, the routing steps cannot run as written.

• SKILL.md (entry point and routing protocol)
• manifest.yaml (axes and file mapping)
• references/ (abstract, article architecture, Chinese author workflow, conclusion and example templates)
• README_EN.md and README.md (English and Chinese notes)

Trigger the skill with a natural request

You do not need special syntax. Describe the manuscript task, say what you have, and name the target journal if you have one. The description triggers on drafting, restructuring, paragraph compression and first-submission material requests, including Chinese phrases such as 论文写作, 章节起草 and 首次投稿材料.

Before drafting, the agent states the axis values it detected in one short line, such as the section and journal it chose. You can correct these at once. The skill then continues unless a necessary decision is still open.

• "Draft the Introduction for a Nature Communications paper from the notes below. The gap is in my pasted summary."
• "Restructure the Results section so each subsection advances one claim. Here are my figure descriptions."
• "Turn my Chinese notes on the degradation experiments into English Results paragraphs."
• "Build a first-submission package outline and readiness checklist for this Nature Machine Intelligence manuscript."
• "Compress the main text of this draft; move supporting analyses to the SI."

Give the skill the evidence it needs

The skill can only structure what you give it. Supply your claims, the key results or figure descriptions, the main methods in brief, and any limits you already know about. Where something essential is absent, expect a placeholder in the output and a list of missing inputs to resolve.

The skill does not verify results. Every quantitative statement and causal claim in the output must still be checked against your data before submission.

• Your core claims and the question they answer
• Key results or figure descriptions with the numbers you trust
• Target section, paper type and journal if known
• Any Chinese notes, which the skill can translate into English structure

Know the boundaries and the safety notes

nature-writing handles initial submission materials prepared before peer review. Language-only edits to existing prose belong to the separate nature-polishing skill, and post-decision reviewer responses and rebuttals belong to the separate nature-response skill. The listing for this skill states that it runs no scripts and uses no network access, and it lists no required dependencies.

Because the skill runs no scripts and makes no network calls, the risk to your machine is limited to the instruction files you install. Review them as you would any document you add to your agent's configuration.

• Runs scripts: no
• Uses network: no
• Requires: nothing listed
• Use nature-polishing for language-only edits to existing text
• Use nature-response after a decision, for revision cover letters and rebuttals