How to use the Nature Reviewer Response Drafter agent skill

What is an agent skill and what does this one do?

An agent skill is a folder of instructions that an AI agent loads when a task matches its description. This skill is for the revision stage of a journal submission. It tells the agent how to parse the editor's letter, classify each reviewer comment, and produce a response package.

According to the skill's description, the agent parses the editor's letter, sorts each reviewer comment by the response it needs, and drafts point-by-point replies, cover letters, and marked-up manuscript changes. It also audits existing drafts. The skill is written to avoid inventing experiments, citations or line numbers, and it flags anything the author must supply.

• Parses the editor letter and reviewer reports into numbered items (editor instructions as E.1, reviewer comments as R1.1, R2.1).
• Drafts a standalone reply for each reviewer, with a privacy-filtered version for mutually blind reviewers.
• Drafts a revision cover letter and maps each claimed change to a manuscript location or a placeholder.
• Audits an existing response draft for tone, completeness and consistency with the manuscript.
• Marks anything the author must supply as AUTHOR_INPUT_NEEDED.

What the skill does not do

The skill is not for writing a first submission or simulating peer review of your own paper. The source names two other skills for those jobs: nature-writing for initial-submission materials and nature-reviewer for simulated peer review.

The skill also forbids fabricating or promising new experiments or evidence. If a reviewer asks for work you have not done, the agent should flag it rather than write a reply that claims it was done. You still have to verify every claim before submitting to a journal.

Install the nature-response skill

Download the ZIP of skills/nature-response from the directory page and unzip it. Then place the folder in one of these locations, depending on how you use your agent.

For a personal setup, put the unzipped folder in ~/.claude/skills/. For a single project, put it in the project's .claude/skills/ folder. If you use the Claude apps instead, upload the skill under Settings → Capabilities → Skills.

• Personal: ~/.claude/skills/nature-response/
• Project only: .claude/skills/nature-response/ inside your project folder
• Claude apps: Settings → Capabilities → Skills, then upload the skill

What you need before you start

The skill needs the text you want it to work with. You must supply the pasted editor letter, the reviewer reports, and the manuscript text or excerpts. Without the editor letter, the agent can still help with a local wording edit or a comment-level triage, but it cannot build a complete revision package.

If you want .tex outputs, you need a LaTeX toolchain. The skill's file list also refers to scripts/check_package_consistency.py, which is not among the listed files, so do not rely on that script being present.

• Editor letter, pasted in full
• Reviewer reports, pasted in full
• Manuscript text or excerpts for the sections the reviewers mention
• Optional: a LaTeX toolchain for .tex outputs
• Your own notes on any new analyses, data or citations you have actually added

How to trigger the skill with natural requests

You do not need a special command. Describe the task in plain language and paste the material. The skill's description lists triggers such as 审稿意见回复, 逐点回复, 返修信, rebuttals and edits to existing response drafts, so requests like the ones below should match.

The skill also works from Chinese requests. If you write in Chinese, it also produces a 中文核对 block.

• "Here is the editor's decision and two reviewer reports. Draft a point-by-point response for a major revision."
• "Write the revision cover letter for this minor revision. Here are the editor's notes."
• "Audit my existing response draft for tone and whether every promise matches the manuscript."
• "Give me a tracker of all reviewer comments with the action needed for each."
• "Revise only this reply to Reviewer 2 so it is shorter and more polite."

Examples of what you get back

The source includes three synthetic examples. In a minor revision, the agent produces a response strategy summary with a decision type, task mode, package readiness and major risks, and it marks missing line numbers instead of inventing them. In a major revision with missing evidence, the agent may repeat details you supplied, such as a dataset ID or figure number, but it does not invent validation results, sample sizes, p-values or confidence intervals.

When reviewers conflict, for example one asking for a stronger causal claim and another asking for softer wording, the agent surfaces the conflict only in an internal strategy summary and prioritises the editor's instruction and the study design.

Safety notes: scripts and network

According to the listing, this skill runs no scripts and uses no network access. Its inputs are the text you paste. Still, review the output before sending anything to a journal, because the skill can only draft from what you give it.

The one exception to watch is LaTeX packages. The skill's workflow says to run scripts/check_package_consistency.py after each draft and manuscript edit for LaTeX packages, but that file is not in the listed files, so that check may not be available.

• Runs scripts: no
• Uses network: no
• Any manuscript edit invalidates earlier quotes and page references, so re-check the package before delivery.

Tips for better revision responses

Give the skill the full decision letter first, so it can identify the decision type. The skill will not infer whether a revision is major or minor from the tone or number of comments, and it asks if the decision is unclear.

Tell the agent about every change you actually made, with page or section references if you have them. If you have no line numbers, say so, and the agent will use placeholders.

• Paste the editor's letter and reviewer reports unedited.
• Give the agent the real list of changes, including any new analyses.
• Ask for an audit after you edit the manuscript, not only before.
• Read every claim in the draft against your own data before you submit.