Nature Reviewer Skill for Claude Code: Mock Peer Review

Makes your assistant review a manuscript the way three Nature referees would, for researchers who want a hard look before submitting.

Installs: 14.3K; on skills.sh
Stars: 51.7K; on GitHub
Files: 20; besides SKILL.md
Updated: Nov 2026; last repo commit

nature-reviewer makes Claude simulate a Nature-style peer review from the referee side. It returns three mutually blind reviewer reports and one synthesis written after the reports are frozen. Each concern has an ID, a pointer to the claim, a pointer to the evidence and a Major or Minor label. A Major concern is marked Blocking only if it stops the manuscript from making its central case.

It triggers on requests such as mock review, pre-submission self-review or reviewer-style assessment. It is not for writing a rebuttal; that goes to the nature-response skill. It runs each reviewer in a separate context where it can, so a full review uses several passes. It also reads shared files from the nature-shared skill, so install both. The repo README lists it as a Draft.

Good for

Researchers stress-testing a manuscript before submission
PhD students and postdocs who want likely referee objections early
Lab leads checking claims, controls and evidence chains

Install nature-reviewer

Claude Code

Copy this prompt.

Install the nature-reviewer skill in Claude Code. Run this command:

npx skills add Yuan1z0825/nature-skills --skill nature-reviewer --skill nature-shared

Then check that it loaded and tell me how to use it.

2
Paste it into Claude Code. It shows the change and asks before running it.

Requires Node.js.

Install nature-shared too: the reviewer's related-files table points at nature-shared/core/consistency-sweep.md, and the README calls nature-shared a shared support package. The README shows this --skill pattern with nature-reader, so these commands are our adaptation of it. For Claude Code the README also describes keeping a clone and adding a subagent or slash command that points at the skill folder, or copying skills into ~/.claude/skills.

Built from the package data we hold. Not yet tested by us. Source.

What's in the skill

When it loads

Provide evidence-grounded mock peer review of scientific manuscripts or excerpts, covering significance, validity, and major/minor concerns. Use for 模拟审稿、投稿前自审、审稿人视角评估; not author rebuttal drafting.

Sections in SKILL.md

Default stance
Accepted inputs
Workflow
Output format
Red lines
Related files
Source hierarchy

Key facts

Author: Yuan1z0825 (nature-skills)
Repository: yuan1z0825/nature-skills
Installs on skills.sh: 14,313 as of 2026-11-02
GitHub stars: 51,696 for the whole yuan1z0825/nature-skills repo
License: Apache-2.0 (repo README badge)
Last repo commit: 2026-11-01
What is inside: 21 files including SKILL.md: 10 reference guides, 5 files under tests (4 Markdown and 1 Python test, test_reviewer_instruction_contracts.py), 2 READMEs, a changelog, manifest.yaml and agents/openai.yaml. SKILL.md does not call the Python file. It links to a file in the sibling nature-shared skill.
Status in README: Draft
Depends on: nature-shared, a support package in the same repo

What it includes

Grounded only in local rules and your manuscript facts
Scores five axes: originality, scientific importance, interdisciplinary readership, technical soundness and readability for nonspecialists. It must not claim the editor's decision or that the paper fits Nature.
Try: “Give me three mutually blind Nature-style reviews of this abstract and results section, then a synthesis.”
Read · Default stance
Full draft, abstract, excerpts, figure legends, author notes
Notes can be in Chinese or English. With partial material it does a bounded review and marks the boundary.
Read · Accepted inputs
3 blind reports, each in its own context
Each reviewer gets only the same manuscript packet and its own emphasis brief. If contexts cannot be isolated, the skill says mutual blindness cannot be guaranteed.
Read · Isolated reviewers
Stable IDs, claim pointers and evidence pointers
Concerns use IDs like R1-M1 and R1-m1. Missing locations are marked instead of invented, and no concern quota is imposed.
Read · Concern ledger
A separate pass over frozen reports and the manuscript
Checks arithmetic, metric bounds, aggregation levels, prose-table order, duplicate displays, odd spread, provenance and reproducibility. Findings are not fed back into reviewers.
Read · Forensic consistency audit
Consensus, differences, forensic findings, revision checklist
Labels something consensus only if at least two reports raised it independently.
Read · Cross-review synthesis
Checks isolation, severity, evidence anchoring and non-invention
Bans invented reviewer identities, experiments, citations and figure details, and bans rewriting frozen reports to reduce overlap.
QA and red lines
Source basis, axes, taxonomy, domain gates, report layout
Loaded on demand. Domain gates cover chemistry, engineering, materials, atmospheric science, climate-ecology, hydrology and remote sensing.
Read · references/ (10 files)

From the official docs

Permissions

Low risk

The skill bundles no scripts, but a full review uses several separate runs and reads your unpublished manuscript.

What it can do

Read your data
Create and change things
Delete things
Read-only mode

Watch out for

Unpublished work goes to your AI provider
Review your provider's data policy before pasting a confidential manuscript. github.com
A full review runs several passes
Three reviewers, an audit and a synthesis run separately, which uses more of your usage allowance than a single reply. github.com

Security advisories

No published security advisories as of Nov 2, 2026.

Technical details

Turning on read-only	No read-only mode is described.
Talks to	Your manuscript is sent to your AI provider as part of the conversation
Sign-in and keys	No key or account is mentioned in SKILL.md.
Data retention	Not described in SKILL.md.
Scopes	Claude reads the manuscript text and files you give it The folder includes agents/openai.yaml and a manifest.yaml with display and loading settings Its related-files table links to nature-shared/core/consistency-sweep.md; nature-shared also holds a check_consistency.py script, which the reviewer SKILL.md does not mention

Questions

What does the nature-reviewer skill do?

It makes Claude produce a mock peer review of your manuscript: three blind referee reports with Major and Minor concerns, then one synthesis. It works only from what you give it and its own review rules.

How do I install nature-reviewer?

Use the install box at the top of this page. It installs nature-reviewer and nature-shared together, because the reviewer reads files from nature-shared.

Does it write my rebuttal?

No. The skill says to route rebuttal drafting to nature-response, which is a different skill in the same repo.

Is the result a real editor decision?

No. The skill must not present its output as an editorial decision, and must not say that the paper belongs in Nature as settled fact.

Can it really keep the three reviews independent?

Only if your setup can run separate contexts or subagents. If it cannot, the skill says to generate one report per invocation or state that mutual blindness cannot be guaranteed.

Is it safe?

The skill file tells Claude to read your manuscript and write reports; its folder holds Markdown, YAML and one Python test file that SKILL.md does not call. Your manuscript text goes to your AI provider like any other prompt, so share only what you are allowed to share. The output is review feedback, not a publication decision.

Information

updated Nov 2, 2026

Repository: yuan1z0825/nature-skills
Folder: skills/nature-reviewer

yuan1z0825/nature-skills skills.sh

Sources

Docs, issue trackers and discussions this page is built from.

Show all 6

Yuan1z0825/nature-skills README · github.com
nature-reviewer README_EN.md · github.com
skills CLI commands · github.com
nature-reviewer SKILL.md · github.com
nature-shared SKILL.md · github.com
nature-reviewer SKILL.md · github.com