peer-review
Prepare evidence-bounded, constructive peer-review drafts and structured manuscript assessments. Use for authorized review of scientific manuscripts, protocols, preprints, or research proposals; reporting-guideline selection; claim–evidence checks; methods, statistics, reproducibility, ethics, figure/table, and citation critique; or revision-response planning.
Install this skill
Works with Claude Code, Codex, Cursor, Gemini CLI, GitHub Copilot, OpenCode, and 70+ other harnesses.
npx skills add KalarisLabs/research-agent-skills --skill peer-reviewuvx research-agent-skills install peer-reviewDocumentation & Instructions
Peer Review
Support an accountable human reviewer with a rigorous, fair, actionable assessment. Treat every unpublished submission and review as confidential.
Mandatory safety boundary
Before reading or analyzing unpublished content:
- Confirm the user is authorized by the publisher, editor, author, or other material owner.
- Check the target venue’s review, confidentiality, co-review, retention, and AI/tool policies.
- Record conflicts, competence limits, requested scope, and specialist-review needs.
- Default to local-only processing.
If authorization is unclear, do not inspect or quote the manuscript. Ask for confirmation or use only the bundled local CLIs, whose reports do not echo manuscript text.
Never:
- Send unpublished manuscript, supplement, review, or editorial text to an external service without specific publisher/author authorization and venue permission
- Upload confidential content to a public model, search engine, citation service, grammar tool, plagiarism checker, or image service
- Reuse content for training, benchmarking, product improvement, or unrelated research
- Read broad environment state,
.envfiles, API keys, or credentials - Call a network, LLM, or image API from bundled tools
- Invoke another skill or a PDF/image pipeline automatically
- Impersonate an assigned reviewer, editor, journal, funder, or author
- Fabricate manuscript details, review findings, citations, analyses, experiments, reproduction, or an editorial outcome
- Announce a decision that belongs to an editor or panel
Delete local copies and derivatives when policy requires; otherwise retain only what the controlling policy authorizes. Record deletion or retention without copying confidential content into the record.
Read references/ethical_review_practice.md before handling confidential material.
Human accountability
Label generated text as a working draft. The accountable human must:
- Read the complete authorized submission and relevant supplements
- Verify every factual statement, calculation, citation, and manuscript location
- Resolve conflicts and disclose assistance as required
- Rewrite comments in their own expert judgment
- Submit through the authorized channel
Automated coverage, consistency, or lint results are not peer review and do not establish manuscript merit.
Intake gate
Copy and complete assets/review_intake_template.json, then run:
python3 scripts/validate_review_intake.py completed-intake.json
Proceed only when status is READY_FOR_LOCAL_REVIEW.
The validator blocks:
- Undocumented authorization
- Missing human accountability
- Unassessed or unresolved conflicts
- Unknown review model or unchecked venue policy
- Unauthorized AI assistance
- External service use
- Data reuse
- Missing deletion/retention planning
It validates declarations, not their truth.
Review workflow
1. Establish scope and available evidence
Record:
- Submission type and stage
- Review question and requested focus
- Target venue and review model
- Materials actually available: manuscript, supplements, protocol, registration, analysis plan, data/code statement, prior decision, or response letter
- Competence areas and limits
- Missing material that prevents assessment
Do not infer absent content. Use “not reported” or “not available for review.”
2. Orient without deciding
Create a short neutral map:
- Research question
- Population or system
- Design and unit
- Intervention, exposure, test, or model
- Comparator/reference
- Outcomes and timing
- Principal claims
Do not write an acceptance/rejection recommendation. Identify what evidence would be needed to evaluate each claim.
3. Select reporting guidance
Copy assets/study_profile_template.json and run:
python3 scripts/select_reporting_guidelines.py local-profile.json
For checklist coverage:
python3 scripts/select_reporting_guidelines.py \
local-profile.json \
--coverage local-coverage.csv
Use the current base guideline, explanation/elaboration, applicable extensions, and target venue policy. See references/reporting_standards.md.
Critical distinction: reporting completeness is not design quality, risk of bias, validity, or merit. Never convert missing items into an automatic score or publication judgment.
4. Map claims to evidence
Prioritize central, causal, mechanistic, safety, diagnostic, prediction, and generalization claims.
For each claim, record:
- Location and claim ID
- Supporting result, figure, table, analysis, or citation IDs
- Direction, magnitude, population, outcome, timepoint, and uncertainty alignment
- Limitation or alternative explanation
- Bounded requested action
Run:
python3 scripts/validate_claim_evidence.py local-claim-matrix.csv
Start from assets/claim_evidence_matrix_template.csv. The report emits IDs and counts, not claim text.
5. Review methods and statistics
Assess in this order:
- Question and target quantity
- Design and unit of inference
- Sampling, allocation, controls, masking, and timing
- Sample-size or precision rationale
- Inclusion, exclusion, attrition, and missingness
- Analysis–design alignment and assumptions
- Multiplicity and prespecification
- Effect estimates, uncertainty, denominators, and harms
- Interpretation, causality, and generalizability
Use references/common_issues.md and references/statistical_reproducibility.md.
For a structured local audit:
python3 scripts/audit_statistics_reproducibility.py \
local-statistics-reproducibility.json
Start from assets/statistical_reproducibility_template.json. Request specialist review when a central method exceeds competence; do not hide uncertainty behind a generic critique.
6. Review reproducibility and transparency
Check, as applicable:
- Protocol, registration, amendments, and analysis-plan consistency
- Data provenance, exclusions, transformations, and accession IDs
- Software, package, model, and parameter versions
- Code, environment, seeds, run instructions, and tests
- Data, code, materials, and model availability or justified restrictions
- Domain metadata standards
Do not claim reproduction unless authorized inputs were actually run with documented commands, environment, and outputs.
7. Review ethics and integrity
Check applicable approvals, consent, welfare, privacy, community governance, funding, sponsor role, conflicts, authorship/contribution, registration, biosafety, and dual-use concerns.
Describe observable evidence and uncertainty. Do not accuse authors or investigate them. Route credible concerns through the confidential editor channel under venue policy.
8. Review figures, tables, and citations
For figures and tables, assess:
- Consistency with text and supplements
- Denominators, units, axes, scales, uncertainty, and legends
- Accessible encoding and sufficient context
- Image acquisition/processing disclosure and source-data policy
This skill has no image-generation or PDF-conversion workflow. Use only user-authorized local artifacts and tools.
For Pandoc-style citations such as [@ref-id]:
python3 scripts/audit_citations.py local-manuscript.md local-references.csv
Start from assets/citation_references_template.csv. This checks key consistency and identifier format only; it does not verify that a source exists or supports a claim.
9. Draft actionable comments
Generate a private scaffold only after intake passes:
python3 scripts/generate_review_scaffold.py \
completed-intake.json \
-o private-review.md
Every major/minor comment should include:
- Location
- Observation
- Evidence or criterion
- Why it matters
- Requested action
Prioritize:
- Claim–evidence alignment
- Methods and statistical validity
- Reproducibility and transparency
- Ethics and participant/animal protection
- Reporting needed for appraisal
- Figures, tables, limitations, and citations
Requests for new work must be necessary to support a central claim and proportionate to scope. Offer narrowing, clarification, sensitivity analysis, correction, or limitation language when that is sufficient.
10. Keep channels separate
Comments to authors contain the scientific review, strengths, major/minor comments, and limitations.
Confidential comments to editor contain only policy-appropriate conflicts, competence limits, assistance disclosure, specialist requests, or substantiated integrity/process concerns that require a separate route.
Do not place ordinary criticism only in confidential notes. Do not reveal reviewer identity under an anonymized process.
11. Lint and finalize
python3 scripts/lint_review.py private-review.md
The linter checks channel separation, unresolved placeholders, a narrow abusive-language lexicon, role/decision phrases, and required actionability fields. It emits line numbers and rule IDs, not review text. Human tone and scientific review remain mandatory.
Before handoff:
- Verify all locations and evidence.
- Remove unsupported or speculative criticism.
- Confirm professional, non-abusive language.
- State review limits and specialist needs.
- Disclose permitted assistance.
- Remove all placeholders.
- Ensure no invented citation, experiment, reanalysis, or outcome.
- Follow the documented deletion/retention rule.
Local tool index
scripts/validate_review_intake.py— scope, authorization, conflicts, policy, handlingscripts/select_reporting_guidelines.py— dated selector and non-scoring coverage auditscripts/validate_claim_evidence.py— claim/evidence alignment matrixscripts/audit_statistics_reproducibility.py— methods/statistics/reproducibility checklistscripts/audit_citations.py— local citation/reference consistencyscripts/generate_review_scaffold.py— separated private Markdown scaffoldscripts/lint_review.py— tone, channel, and actionability lint
Full schemas and exit codes: references/tool_reference.md.
References and assets
references/ethical_review_practice.md— COPE/ICMJE duties, confidentiality, AI, channelsreferences/reporting_standards.md— current major guidelines and verified domain standardsreferences/statistical_reproducibility.md— methods, statistics, and reproducibility reviewreferences/common_issues.md— contextual issue patterns and constructive responsesreferences/security_validation.md— baseline remediation and local scan resultsassets/source_ledger.csv— authoritative sources verified 2026-07-23assets/reporting_guidelines.json— local selector catalogassets/review_scaffold_template.md— private structured draft
The source ledger is dated. Recheck live primary sources and the target venue policy for a later review, without exposing confidential manuscript text in search queries.
Agent operating procedure
- Check the environment. Collect the user's draft, data, figures and target venue before writing anything.
- Pin down the inputs. Confirm formats, identifiers and parameters from the data or the user. Ask rather than guess any value that changes the result.
- Run a small version first. Draft one section or paragraph and confirm tone, structure and terminology with the user.
- Execute the full task using the instructions and references above.
- Validate the result. Every claim traces to the user's data or a verified source; numbers match tables and figures; citations are real (
citation-verification). - Report. State what was run (versions, commands, parameters), what was checked, and what is still uncertain.
| If this happens | Do this |
|---|---|
| A claim has no supporting data or citation | Flag it in the text (e.g. [citation needed]) instead of inventing support. |
| A function, flag or endpoint in these instructions is missing in the installed version | Check the installed version's own documentation (help(), --help, official docs), adapt, and tell the user. Never invent an API. |
| A required input, identifier or parameter is ambiguous | Ask the user, or state the assumption explicitly before running. |
Integrity rules
- Never fabricate results, parameters, identifiers, citations or statistics. If something cannot be run or verified, say so plainly.
- The user is the author: never add results, references or claims they did not provide or approve.
- Treat version-specific details here as possibly outdated: confirm them against the official documentation for the installed version.
- Ask before actions that cost money, consume shared GPUs or cloud quota, touch personal or patient data, or cannot be undone.
Related skills
scientific-writing: Draft, revise, and audit scientific manuscripts or reports with explicit evidence provenance, reporting-guideline coverage, authorship acco…reproducibility-statement: Prepare the reproducibility, transparency and open-science parts of a paper, including data and code availability statements, reproducibili…cell-press: Prepare manuscripts for Cell Press journals (Cell, Molecular Cell, Neuron, Immunity, Cell Reports, Cell Systems, iScience, Cell Metabolism,…
Trigger Prompts & Evaluation Corpus
These evaluation prompts test that the agent correctly selects peer-review for relevant tasks and leaves it out for near misses.
Should trigger (3)
- should trigger
“I've been invited to review a Nature Communications manuscript on single-cell atlas integration. Can you help me draft my referee report?”
expects Confirms authorization and the venue's confidentiality and AI/tool policies, and runs the intake gate before reading the manuscript. Processes locally only. Labels output as a working draft that the human reviewer must verify and rewrite. Does not give an accept/reject recommendation.
- should trigger
“I'm a reviewer for BMJ Open and have the editor's invitation. The paper is a stratified cluster RCT of a school-based vaccination reminder (manuscript_v1.docx and supplement S1.pdf are saved locally). The abstract says the intervention 'reduced HPV vaccine hesitancy by 30%', but the primary outcome in the registry entry looks like uptake. Help me work out which reporting guideline applies and whether the claims match the evidence. I don't want anything uploaded anywhere.”
expects Completes the intake and authorization gate, then keeps all processing local. Selects CONSORT with the cluster extension and treats checklist coverage as non-scoring. Builds a claim–evidence matrix that flags the outcome and registration mismatch, with bounded requested actions. Does not claim anything has been reproduced, and does not decide for the editor.
- should trigger
“Our grant resubmission got three reviewer reports back and I have to write the rebuttal. Reviewer 2 says our mixed-effects model ignores clustering by site and wants a sensitivity analysis. Help me plan how to respond to each comment point by point.”
expects Plans a revision-response with a point-by-point structure. Assesses each critique against the evidence and offers correction, clarification, or a sensitivity analysis where appropriate. Does not fabricate new analyses or results, and keeps the tone constructive and professional.
Near misses (2)
- near miss
“Can you rewrite the Discussion section of my own manuscript so it flows better and tighten the abstract to 250 words? It's about CRISPR screens in T cells.”
expects Recognizes this as authoring and editing the user's own manuscript, not peer review. Directs to scientific-writing, which covers revising manuscript sections with evidence provenance and consistency checks.
- near miss
“Write me a Python function that computes Cohen's kappa between two raters for my annotation data.”
expects Writes the kappa code directly. No peer-review workflow, intake gate, or manuscript review is needed.