Paper Review - Avaliação Simulada de Artigos Acadêmicos
academicatstool-netizen/Cat_paper_review
Revisão acadêmica simulada com padrões metodológicos contextualizados
Ver no GitHubSkill que executa revisão por pares simulada em 5 perspectivas independentes (significância, rigor, evidência, argumento, clareza) com validação metodológica específica por tipo de paper (STROBE/PRISMA/CONSORT/COREQ/Toulmin). Retorna carta de referee com parecer escalonado e veredito numérico. Funciona nativamente em Claude, ChatGPT e DeepSeek.
Instalação
npx convez-ai academicatstool-netizen/Cat_paper_review1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Avaliar rascunhos de artigos de pesquisa antes de submissão
- Validar propostas de estudo contra critérios NIH ou equivalentes
- Revisar estrutura argumentativa de ensaios e trabalhos acadêmicos
- Calibrar qualidade de redação científica em diferentes estágios
Veredito

Ferramenta especializada e bem-fundamentada para acadêmicos que buscam feedback estruturado antes de submissão. A cross-examination entre revisores diferentes reduz feedback genérico, e a adaptação a padrões metodológicos reais (STROBE, PRISMA, etc.) é rara e valiosa. Recomendado para pesquisadores em estágios intermediários que entendem limitações de IA na revisão crítica profunda.
README
A tradução para o português deste README ainda está na fila.
English · 中文
🐱 Paper Review
Run your own draft through a six-phase simulated peer-review panel and get back a referee letter plus a scored verdict — calibrated to how finished your draft really is.
🪶 This is the lite, open-source edition of AcademiCats — now in free open beta
The full product at academicats.com is an AI research workbench that takes you from finding papers through reading, writing, and self-review. This skill is a free, self-contained slice of the review workflow you can run on your own AI — Claude, ChatGPT, or DeepSeek.
✨ What it does
🧪 A whole panel's scrutiny, from one model — your draft is examined through five independent review lenses (significance, rigor, evidence, argument, clarity) that then challenge each other in a rebuttal round before a moderator reconciles them. That cross-examination is what kills plausible-but-wrong feedback and sharpens what's real.
🔬 Judged by the right standard — the methodology critique adapts to what you actually wrote: an observational study is checked against STROBE, a systematic review against PRISMA + AMSTAR-2, a trial against CONSORT, a qualitative study against COREQ, a proposal against NIH criteria, an essay against the Toulmin model of argument — not one generic checklist for everything.
📊 A verdict with a number — you get a referee letter (verdict, strengths, issues to fix, optional improvements) and a scorecard: each dimension scored 0–10, an overall, and a recommendation — Accept / Minor / Major Revision / Reject.
🎚️ Calibrated to your stage — tell it whether it's a final, working, sketch, or student draft, and it won't judge a class essay like a journal submission.
🎬 Demo
"Peer-review my draft — it's a working draft of a short research proposal. Is it ready?"
You said it's a working draft — so it reviews straight away (no forms; it notes the stage it used so you can correct it). The result is always the same shape — a referee letter, a scorecard, then a one-line summary — no matter which model runs the skill:
Verdict
A clear, relevant proposal — but as a working draft it's a promising skeleton, not yet a viable plan. The key fix: it rests on lines of code per hour, a discredited productivity proxy that would invalidate the result no matter how the experiment runs.
Strengths
- The two-group design is the right shape for the question.
- Clear, readable prose.
Issues to Fix
Replace the sole metric — lines-of-code rewards verbosity and ignores correctness; add a validated outcome (task completion, tests passed, time-on-task). Ground it in prior work — there are no citations anywhere; add a short related-work paragraph.
| Significance | Rigor | Evidence | Argument | Clarity | Overall | Verdict |
|---|---|---|---|---|---|---|
| 6 | 3 | 3 | 6 | 6 | 4.8/10 | Major Revision |
A relevant question undermined by a confirmation-seeking design; not ready until the metric and controls are fixed. Appraised as a research proposal against NIH criteria (Significance · Innovation · Approach) · 5 lenses cross-checked.
🚀 Get started — pick your platform
Pick whichever AI you use; each setup takes under a minute.
🖥️ Claude Code — runs locally, triggers itself
mkdir -p ~/.claude/skills
git clone https://github.com/academicatstool-netizen/Cat_paper_review.git ~/.claude/skills/paper-review
Restart Claude Code, then just ask — "review my draft, it's a working draft of a research proposal …"
🌐 Claude (web / desktop app) — download paper-review.skill, then upload it under Settings → Capabilities → Skills. Ask in any chat.
🤖 ChatGPT — open PORTABLE_PROMPT.md and paste it into a Custom GPT's Instructions (or just send it as your first message). Then paste your draft.
💬 DeepSeek / any other model — paste PORTABLE_PROMPT.md as the system prompt (or first message). Then paste your draft.
✅ No web access needed — Paper Review is pure reasoning, so it runs fully on any model, online or off. It reviews the draft you paste; it never needs to look anything up.
💙 What's inside
Six phases — intake → 5 lenses → rebuttal round → moderator → editor letter → scorecard — distilled from the same review method behind the full product.
| Paper Review (this skill) | AcademiCats full product → | |
|---|---|---|
| ⚡ Speed | minutes (one model pass) | faster — streamed, parallel panel |
| Six-phase panel + scored verdict | ✅ | ✅ |
| Standards by paper type (STROBE / PRISMA / CONSORT / COREQ / Toulmin) | ✅ | ✅ |
| Stage-aware calibration | ✅ | ✅ |
| Find & read the literature it asks for | bring your own | ✅ end-to-end |
| Saved drafts & revision history | — | ✅ |
| Polished web & mobile app | — | ✅ |
🐱 The AcademiCats skill family
Four open skills that chain into one research workflow — install any or all:
- 🧭 Find Angles — turn a topic into research directions
- 🔍 Paper Search — find & read papers
- ✍️ Synthesis Lab — write grounded papers from your sources
- 🧪 Paper Review (you are here) — peer-review your own draft
Install all at once — clone any one repo, then run bash install.sh.
🙋 FAQ
- What are all these files? Use just one path above — a git clone (Claude Code), the
.skillfile (Claude web/desktop), orPORTABLE_PROMPT.md(ChatGPT/DeepSeek).SKILL.mdandreferences/are internals your assistant loads for you — no need to open them. - It didn't trigger? Restart Claude Code after installing, and phrase your message as a task — "review my draft …".
- Tell it the draft stage. Say it's a "working draft", "final", "rough sketch", or "student essay" so it grades you fairly — a class essay shouldn't be judged like a journal submission.
- Which model? Any strong model works — Claude Sonnet/Opus, GPT‑4o/o‑series, or DeepSeek‑V3/R1 give the best results.
- Private & free? It runs on your own AI — no account, nothing sent to us.
Want the whole research workflow?
→ academicats.com ←
🚀 The full product is in open beta — free to try right now.
Made with 💙 by the AcademiCats team · MIT License