Guide · Agent skill
How to use the Peer Review Assistant agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
The peer-review agent skill is a set of instructions, templates and small local scripts that helps an agent support a human reviewer. It organises a scientific manuscript review: it maps the research question, checks whether claims match the cited evidence, suggests reporting guidelines, and reviews methods, statistics and reproducibility. This guide explains what the skill does, how to install it, how to ask for it in plain language, and the safety limits you need to respect before using it on unpublished work.
Peer Review AssistantK-Dense-AI/scientific-agent-skills · MIT Download the skill Quick start
- 1Download the ZIP. Download the ZIP for skills/peer-review from the directory page.
- 2Unzip into a skills folder. Unzip it into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder. Alternatively, upload it under Settings → Capabilities → Skills in the Claude apps.
- 3Confirm authorisation. Before using it on unpublished material, confirm you are authorised by the publisher, editor or author and check the venue's policies.
- 4Complete the intake file. Copy assets/review_intake_template.json, fill it in and run the validator until the status reads READY_FOR_LOCAL_REVIEW.
- 5Ask in plain language. Describe the task, such as checking claim–evidence alignment or selecting a reporting guideline, and name the manuscript section.
- 6Verify and write your own comments. Check every point against the manuscript, resolve conflicts, and rewrite the comments in your own expert judgement.
What the peer-review skill makes the agent do
The skill is written for an accountable human reviewer. It prepares an evidence-bounded, constructive draft and a structured assessment. It covers authorised review of manuscripts, protocols, preprints and research proposals, along with reporting-guideline selection, claim–evidence checks, methods, statistics, reproducibility, ethics, figure and table, and citation critique, and revision-response planning.
Everything the agent produces is labelled as a working draft. The skill does not make an accept or reject recommendation, and it does not give a quality score. It states that checklist coverage is not the same as merit.
- Maps the research question, population, design, comparator, outcomes and principal claims without deciding
- Selects reporting guidance and checks checklist coverage
- Lists each claim against the evidence that supports it
- Works through methods and statistics in a fixed order
- Reviews reproducibility and transparency items
- Drafts a review scaffold with placeholders for the reviewer to complete
Install the peer-review skill
Download the ZIP for skills/peer-review from the directory page. Then choose one of two routes. For a personal install, unzip it into ~/.claude/skills/. For a single project, unzip it into the project's .claude/skills/ folder. You can also upload the skill in the Claude apps under Settings → Capabilities → Skills.
The bundled scripts use only the Python 3.11+ standard library, so you need Python 3.11 or newer to run them. The skill metadata states that no API keys or accounts are required.
- Personal install: unzip into ~/.claude/skills/
- Project install: unzip into the project's .claude/skills/
- Claude apps: Settings → Capabilities → Skills, then upload
~/.claude/skills/How to trigger the skill with natural requests
You do not need a special command. Describe the task in plain language and mention the manuscript or the kind of help you want. The skill's description covers requests about reporting guidelines, claim–evidence checks, statistics, reproducibility, ethics, figures and tables, citations and revision-response planning.
Before the agent reads unpublished material, it will ask you to confirm your authorisation. Have that answer ready, along with the venue's review and confidentiality policy.
- "Help me check whether the claims in this assigned manuscript match its evidence."
- "Which reporting guideline fits this randomised trial report?"
- "Draft a structured methods and statistics review for this preprint I've been assigned."
- "Plan responses to the reviewer comments in this revision letter."
Safety notes: authorisation, scripts and network
The skill sets a mandatory safety boundary. Before reading unpublished content, the reviewer must be authorised by the publisher, editor, author or other material owner, and the venue's confidentiality and AI-tool policies must be checked. If authorisation is unclear, the agent will not inspect or quote the manuscript. It will only use the intake validator or synthetic or public examples until the question is resolved.
The bundled scripts run locally and make no network calls, according to the skill. Running a script locally does not make the surrounding chat session local, though. If manuscript text is pasted into a hosted assistant, treat that as external processing for policy purposes. The skill is also explicit that it must not read .env files, API keys or credentials, and must not send confidential text to public tools.
- Requires Python 3.11+ standard library for the bundled scripts
- Runs scripts: yes. Uses network: no, per the skill metadata
- Do not use it on confidential manuscripts without documented authorisation and venue permission
- Local processing of the manuscript also requires authorisation
Run the intake gate and the bundled checks
Before manuscript analysis, the skill asks you to copy and complete assets/review_intake_template.json and run the intake validator. Analysis proceeds only when the status is READY_FOR_LOCAL_REVIEW. The validator checks declarations, not whether they are true, so the reviewer remains responsible for accuracy.
The skill also includes scripts for reporting-guideline selection, claim–evidence validation and statistics and reproducibility audits. Each one takes a local JSON or CSV file that you start from the matching template in assets/.
- Intake: copy assets/review_intake_template.json and validate it
- Reporting guidance: start from assets/study_profile_template.json
- Claim–evidence: start from assets/claim_evidence_matrix_template.csv
- Statistics and reproducibility: start from assets/statistical_reproducibility_template.json
python3 scripts/validate_review_intake.py completed-intake.jsonTips for getting useful output
Give the agent only the material you are authorised to use, and name the sections or figures you care about. The skill asks for locations such as section, page, line, figure, table or claim ID. Specific locations make the draft easier to check.
Treat every statement in the output as unverified until you have checked it against the manuscript. Where the material does not contain something, the skill writes "not reported" rather than guessing, and you should keep that wording when it applies.
- Name the section, figure or claim ID you want reviewed
- Ask for a neutral summary before asking for critique
- Ask for specialist review when a central method is outside your competence
- Record any conflicts and competence limits before starting
What people use it for
- Organising a checklist-based draft for a manuscript you have been formally assigned
- Creating consistent review templates for a journal or conference
- Teaching students to critique methods, statistics and claim–evidence alignment using public or synthetic papers
- Choosing the right reporting guideline for a study report before reviewing it
- Planning point-by-point responses to reviewer comments on a revision
Frequently asked questions
Is the peer-review agent skill free?
The skill is published under the MIT licence, and the skill metadata states that no API keys or accounts are required. Check the directory page for any other terms before you rely on it.
What do I need to run the peer-review skill?
The bundled scripts need Python 3.11 or newer, using only the standard library. The skill makes no network calls, and it needs no API keys or accounts according to its metadata.
How do I install the peer-review skill?
Download the ZIP from the directory page and unzip it into ~/.claude/skills/ for personal use or into a project's .claude/skills/ folder. In the Claude apps you can instead upload it under Settings → Capabilities → Skills.
Can I use it on a confidential manuscript?
Only with documented authorisation from the publisher, editor or author and the venue's permission. The skill says that pasting unpublished text into a hosted assistant counts as external processing, so check the policy first.
Does the skill decide whether a paper should be accepted or rejected?
No. The skill explicitly forbids accept or reject recommendations and quality scores. It treats checklist coverage as not equal to merit, and the human reviewer makes the final judgement.
How is this different from a general AI writing assistant?
This skill is built around review workflow: an intake gate, reporting-guideline checks, claim–evidence matrices and statistics audits, with local scripts. It is designed to keep the reviewer accountable, whereas a general assistant has no such review structure.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Use this skill if you are a formally assigned reviewer, an editor building consistent templates, or a student learning to critique methods and claim–evidence alignment, and you can work with public or authorised material. Pick something else if you need an automatic accept or reject decision, a quality score, or a tool for confidential manuscripts you are not authorised to share, and get specialist input for high-stakes or clinical work.
Download the skill