paper-reviewer
Skill661 starsChanged 39 days agoMITView on GitHub
Referee-side skill for Nature Portfolio manuscripts: write a referee report, inventory every ask and sub-ask in received comments, grade drafted replies, and test reply readability. Check methodology, statistics, reporting, reproducibility and data/code availability. Preserve referee wording, numbering and order; require one reply per ask with a direct opening answer and a manuscript location or explicit no-change rationale. Use for review this manuscript, what will reviewers attack, turn comments into a checklist, answer each comment, did I miss a point, would this reply satisfy the reviewer, 审稿、写审稿意见、模拟审稿、拆解审稿意见、逐个回复、逐条对照、有没有漏回、审稿人会不会接受、以审稿人视角检查回复、回复够不够直白. Use rebuttal-response to choose response stances, calibrate claims and assemble or audit the whole letter.
8 other projects use a version of this.
Adopted by
Most recently taken up by
- whoisraibolt/paper-skillstheir own version, first seen Oct 13, 2026
- Ingar30/reviewertheir own version, first seen Oct 9, 2026
- darker2016/WorkbuddySkillGroups4DSHtheir own version, first seen Aug 14, 2026
- Chanw-research/claude-code-paper-writingtheir own version, first seen May 10, 2026
- yipng05-max/-skillstheir own version, first seen Apr 2, 2026
- tcztzy/skillstheir own version, first seen Apr 2, 2026
Other versions8 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 8
- yipng05-max/-skills★ 302same name
- Ingar30/reviewer★ 229same name
- Chanw-research/claude-code-paper-writing★ 42same name
- darker2016/WorkbuddySkillGroups4DSH★ 6same name
- astoreyai/claude-skills★ 3same name
- tcztzy/skills★ 2same name
- JieyuZhao/claude-skills★ 1same name
- whoisraibolt/paper-skills★ 1same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository26 other files it gives its agents.
- anti-defensive-writingSkill
- citation-verifierSkill
- claim-source-verificationSkill
- data-availabilitySkill
- draft-marker-disciplineSkill
- figure-plannerSkill
- manuscript-optimizerSkill
- paper-bootstrapSkill
- paper-workflowSkill
- rebuttal-responseSkill
- results-section-revisionSkill
- review-article-architectureSkill
- scientific-prose-styleSkill
- scientific-writingSkill
- stats-reporting-auditSkill
- submission-auditSkill
- write-scientific-manuscriptSkill
- figure-styleSkill
- nature-figureSkill
- academic-presentationsSkill
- conference-paper-writingSkill
- reference-audit-guideSkill
- academic-researcherSkill
- paper-analyzerSkill
- results-analysisSkill
- nature-portfolio-playbookSkill
The file
We split this skill into the rules it gives an agent
11 rules. 0 are common, 9 are distinctive, and 2 are very specific to this project.
0
Common
Most of the 6 edited versions we read keep these.
9
Distinctive
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 34
The reply uses the referee's own words: If they wrote data leak, the reply says data leak.
- Distinctive: line 35
Never coin a replacement term, a framework name, or a category label that the referee would have to learn in order to read your answer.
- Distinctive: line 60
Read review_checklist.md for the stage-by-stage criteria, the referee fields Nature Portfolio actually asks about, the report shape, and …
- Distinctive: line 68
Read comment_splitting.md.
- Very specific: line 83
Numbering is the referee's: Never renumber, never reorder, never regroup by theme.
- Distinctive: line 95
Read reply_blocks.md. The block is fixed:
- Distinctive: line 160
Never rename what the referee named.
- Distinctive: line 164
Every number carries value, unit or metric, direction of goodness, and comparator.
- Distinctive: line 168
No reply ends without a location, or an explicit statement that nothing changed and why.
- Distinctive: line 191
Every figure, panel, table, and line number they named appears by name, including the ones needing no change.
- Distinctive: line 198
Run the checker for the mechanical half: verbatim quoting, order, one reply per ask, locations, blanket phrases, bloat.
“Common” is measured: kept by at least 6 in 10 of the 6 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Referee Reports and the Replies That Answer Them
Two jobs, one standpoint. Writing a referee report and answering one are the same competence seen from two sides: you can only prove you answered a question if you can reconstruct the question that was asked.
Pick the mode before starting.
| Mode | Input | Output |
report | a manuscript | a referee report, or a prediction of what referees will attack |
inventory | a referee report | every ask and sub-ask, in the referee's own words and order |
reply | a report plus a manuscript | one reply per ask, plus the coverage audit |
grade | a drafted reply | what an impatient referee would send back |
The contract
A referee report is a list of questions. A reply is a list of answers. The job is to make the two lists line up so exactly that a referee can verify it by scrolling their own report alongside your letter.
Same count. Same identifiers. Same order. Same words. Depth matched to weight.
The alignment has to be visible on the page, not asserted in a summary. Three consequences, and every rule in this skill follows from them:
- The referee's comment is reproduced verbatim, immediately above its reply. Not paraphrased, not summarised, not trimmed. Nature Methods names rephrasing a referee's comment as a manipulative tactic, and Nature Chemistry asks authors to list the referees' individual points.
- The reply uses the referee's own words. If they wrote
data leak, the reply saysdata leak. Never coin a replacement term, a framework name, or a category label that the referee would have to learn in order to read your answer. Plain language comes from glossing their term, never from inventing a better one. - The referee is impatient. Nature Neuroscience puts it as
keeping in mind that referees are busy. The first sentence carries the answer. Reply length is the minimum that settles the point. Over-answering a light comment is as much a defect as under-answering a heavy one, and Nature Chemistry warns specifically that a long-winded essay style makes a revision harder to follow.
What this skill does not own
rebuttal-responsedecides what to say: the stance on each ask, the strongest claim the evidence supports, honest handling of negative results, and keeping letter, manuscript, and Supplementary Information consistent. Build the inventory here, draft there, bring the draft back here for the impatient-referee pass.submission-auditruns a single referee-style rejection pass over a manuscript nobody has reviewed yet. Use it before submitting; use this skill to simulate a referee beforehand and to work real reports afterwards.stats-reporting-auditowns statistical substance: the independent unit behindn, replication, multiple-comparison correction, figure-legend statistics. When an ask is statistical, take the substance from there and only the reply shape from here.nature-portfolio-playbookowns venue and article-type choice. This skill assumes the venue is fixed.
Mode report
Read review_checklist.md for the stage-by-stage criteria, the referee fields Nature Portfolio actually asks about, the report shape, and the rules for writing comments that can be answered one at a time. common_issues.md is the catalogue of what to look for; reporting_standards.md covers guideline compliance and the portfolio-wide Reporting Summary.
Mode inventory
Read comment_splitting.md. The rule in one line:
Split wherever a different action would be needed to satisfy the referee.
Two clauses are one ask only if the same action settles both. The test: could the referee write in
round 2, about this span alone, the authors did not do this?
Split on any enumerator the referee supplied, any request verb, any question, any sentence asserting a defect as fact, any separate target object, and any change of deliverable. Subtract a lead sentence that the sub-items enumerate, a request restated in different words, and the sentences that merely justify a request. An ask naming a set expands to one entry per element in the ledger, which is what stops a half-done set from slipping through. In the letter those entries may share one reply, provided it names every element and says what happened to each, including the ones needing no change. A range reply that names none of them reads the same whether you fixed four panels or two.
Numbering is the referee's. Never renumber, never reorder, never regroup by theme. Preserve gaps: if
they commented on panels a) and c) and skipped b), so does the reply. Preserve their typos inside
the quote. If the referee numbered nothing, do not invent a scheme that looks like theirs; split at
their paragraph breaks, use visibly author-added labels such as R1-P1, and say once that you added
them.
Record the result in notes/review-ledger.md. It is a working artifact and it never enters the letter:
a referee should not have to consult a table to find the reply to their own point, and should never be
handed a table in which the authors have graded their comments.
Mode reply
Read reply_blocks.md. The block is fixed:
[the referee's comment, verbatim, complete]
[answer sentence]
[evidence, if the band allows it]
[location]What that looks like on the page. The referee's own labels, their own order, one reply each, and the answer first in every one:
## Reviewer #2
Reviewer #2 raised 5 numbered comments, 6 points under "Figures:", and 12 points under
"Minor comments:". Each is quoted below in your original order with its own reply.
### Fig. 2a
> a) y-axis legend (FoE) is missing.
Added. The y axis of Fig. 2a now reads "FoE (fraction of enriched features, higher is better)".
(Fig. 2a)
### Fig. 2c
> c) The used dataset is not mentioned in the figure legend. Is it BBBC022?
Yes, it is BBBC022. The legend now names it. (Fig. 2 legend)
### Fig. 2d
> d) Classification loss decrease makes MAP and FoE worse?
Yes, beyond epoch 40. Lower classification loss past that point reflects fitting plate-level batch
signal, which MAP and FoE penalise. (page 7; Supplementary Fig. 7b)Two panels the referee did not raise get no block, and the gap is left open. A referee scrolling their
own report sees their a), c), d) against your a), c), d).
Two dials
Severity, meaning your judgment of scientific risk, sets how much work happens. The referee's own framing sets how many words the reply gets.
| The referee's framing | Ceiling |
under a minor heading, or one sentence requesting one edit | 40 words |
| a numbered substantive comment | 120 words |
| an objection to validity, controls, statistics, or the central claim | 300 words |
When the two disagree, do the work and keep the reply short. A serious problem filed as a one-line minor comment still gets the full experiment and the manuscript rewrite. The reply is still two sentences, because the fix lives in the manuscript and the reply points at it.
Plain language
The rules that matter most. The full set of fourteen, each with a failing example and its rewrite, is in reply_blocks.md.
- The first sentence is the answer, and reads correctly for someone who has not re-read the comment.
- First sentence under 25 words, one main clause.
- First sentence uses the referee's own nouns.
- Never rename what the referee named. If their term is loose, keep it as the head word and attach the correction in the same sentence.
- Any term they did not use gets a plain gloss, six words or fewer, on first use in that block.
- Thanks is at most a clause, never a sentence of its own.
- Every number carries value, unit or metric, direction of goodness, and comparator.
- One idea per sentence, 30 words maximum, one hedge, and no pronoun subject reaching back across a sentence boundary. No em dashes.
- Say the unwelcome thing first, and never open on filler such as
As the referee correctly points out. - No reply ends without a location, or an explicit statement that nothing changed and why.
The five answers
Plain descriptions for your own use; they never appear as labels in the letter. We did it, past-tense
action plus the outcome. We did a smaller version, what you did and what you did not, in that order.
It is already in the paper, the location first, then a change that makes it findable anyway. We
disagree. We cannot do it, the refusal and the concrete reason in one sentence.
Disagreement is the one that goes wrong. State the fact, not the stance: There is no data leak in these
experiments, never We respectfully disagree. Attack the premise, never the reader. Lead the evidence
with a result that would have come out the other way if the referee were right. Concede the true
fraction in its own sentence. Change the manuscript anyway, because the concern arose from text that
permitted it: make that text precise (a definition, a clearer design statement, a narrowed claim),
not defensive. Reassurance and the letter's argument stay in the letter.
Mode grade
- The two-line test. Referee's ask on one line, your first sentence below it, read only those two. Does the main verb match, or is it explicitly refused? Does the object they named appear? Was your first sentence already true before the revision? If a reader who does not know the paper cannot say whether line 2 answers line 1, the reply fails.
- The count check. Asks equal reply blocks. Every identifier appears once, in order, gaps preserved. Every figure, panel, table, and line number they named appears by name, including the ones needing no change. No block's whole answer is a cross-reference. Every block ends with a location.
- The highlighter check. Strike through each clause of the report once you find the sentence that settles it. Anything unstruck is a missing reply.
- Landing. The change a reply promises must exist in the manuscript. A reply that lives only in the letter is the most commonly caught failure in re-review.
Run the checker for the mechanical half: verbatim quoting, order, one reply per ask, locations, blanket phrases, bloat.
python3 ~/.agents/skills/paper-reviewer/scripts/check_coverage.py \
--reviewer reviews/round1.txt --letter response/round1.md \
[--ledger notes/review-ledger.md] [--manuscript manuscript.md] [--strict] [--json]
# Claude Code (global install): replace ~/.agents/skills with ~/.claude/skills
# Claude Code (project-local install): replace ~/.agents/skills with .claude/skillsIt cannot judge whether a reply is responsive. A clean run is not a good letter.
Nature Portfolio notes
Sources for everything in this section are listed at the end of review_checklist.md. Policy changes, so check the journal page.
- Assume the reply will be published unless you have checked the specific journal page and confirmed otherwise. Nature has published referee reports and author responses for all new research submissions since June 2025; Nature Communications since November 2022; other titles vary, some by opt-in. Every referee sees every other referee's report and your rebuttal in any case.
- The overview belongs in the cover letter to the editor, not at the head of each referee reply. The cover letter is not published and is the place for anything the referees should not see.
- The editor adjudicates. Decisions are not a vote. When referees conflict, answer each in their own block, state the conflict once as fact, and let the editor resolve it. Do not use one referee against another.
- Editorial requirements are not referee comments. A decision letter can impose things no referee asked for. They get their own blocks.
- Referee report fields differ across the portfolio, and reports need not follow the listed order, so never assume a received report is numbered. Some journals expose a separate code-assessment section, so a code comment may arrive outside the main remarks block and still needs its own reply.
Related files
| File | Open when |
| review_checklist.md | writing a referee report, or predicting what referees will attack |
| comment_splitting.md | turning a received report into an inventory, or carrying it across rounds |
| reply_blocks.md | drafting or grading any individual reply |
| common_issues.md | naming a methodological or statistical problem, or classifying a received comment |
| reporting_standards.md | a guideline, checklist, or Reporting Summary is at issue |
scripts/check_coverage.py | before sending, for the mechanical coverage audit |
Provenance: the referee-report half is adapted from inno-paper-reviewer. The reply-alignment layer is
original to this repository, distilled from published Nature Portfolio guidance on responding to
referees and from transparent peer review files.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

