Nature Review Studio

SkillFiles & storage

Multi-perspective manuscript review and point-by-point response generation, distilled from 1287 Nature Peer-Review-File PDFs (2025–2026). Review and respond each produce exactly two synchronized deliverables: one formal DOCX and one clean Markdown file with the same base name. Output language follows the user's prompt. Also trigger on 多视角审稿、高水平审稿、Nature 风格审稿、 审稿回复、逐条回复、修订程度、修改任务表、蒸馏审稿语料、审稿知识库更新。

Available today. Use it from your connected AI after setup.

Connect ahel once, and every AI you use reads what you have installed.

Then ask your AI: use the Nature Review Studio skill

What this skill tells your AI

The instructions your AI receives, as published by mumdark/nature-review-studio in skill/SKILL.md and read by ahel’s review.

V1.4.0: full 1287-case knowledge base, simplified Word styling, and a locked two-file output contract.

Interfaces

InterfacePurposeUser-facing outputs
Nature_ReviewStudio.reviewGenerate 2–5 independent reviewer reports from a manuscriptone .docx + one same-stem .md
Nature_ReviewStudio.respondGenerate editor letter, point-by-point responses, and revision tasksone .docx + one same-stem .md
Nature_ReviewStudio.updateIngest, anonymize, distill, index, benchmark, and version new PRF materialupdate report and internal knowledge artifacts

Locked output contract

For review and respond, create exactly two synchronized user-facing files:

  • build/output/<entry>_<case>_<YYYYMMDD>.docx: formal Word deliverable.
  • build/output/<entry>_<case>_<YYYYMMDD>.md: clean Markdown companion.
  • The two files must use the same language, section order, reviewer numbering, concerns, consensus content, and revision task table.
  • Do not create separate preamble, compliance, claim–evidence, coverage, adversarial-check, divergence, severity-change, or editor-decision files.
  • Markdown output is mandatory. Do not expose or use a --no-md mode.

Formal content order

  1. Editor letter and revision-level banner.
  2. Each reviewer in Overall assessment followed by numbered Major concerns and Minor concerns.
  3. Cross-review consensus, without divergence or editor-decision hints.
  4. Revision task table.

Internal compliance checks, source coverage, claim–evidence mapping, and adversarial review remain silent and are never inserted into either deliverable.

Language

Use the language of the user's request unless the user explicitly requests a different output language. Chinese requests return Chinese DOCX and Markdown; English requests return English DOCX and Markdown.

Rendering

Use nature_open_peer_review/build/pipeline/render_review_docx.py. It writes the DOCX and the mandatory same-stem Markdown companion in one invocation. Report both absolute paths to the user after successful generation.

Word styling follows the current V1.4 renderer: system-default fonts, black text, no forced hyperlink decoration, and bold limited to headings, table headers, confidence lines, task IDs, and evidence anchors.

Knowledge use

  • review: retrieve concern axes, method checks, severity priors, reviewer-set patterns, and structurally similar historical cases.
  • respond: retrieve response strategies, action statuses, resolution patterns, and second-round follow-up risks.
  • Never copy historical reviewer sentences verbatim or expose case identities.

Read the references for workflow details and silent quality-control rules.

Signals

GitHub stars
137
Forks
11
Last commit
Aug 2026
Advanced
Catalog kind
skill
Gateway key
nature-review-studio
Source
github.com/mumdark/nature-review-studio