Revisão Acadêmica com Dupla Perspectiva
ChanMeng666/academic-paper-review-skill
Pareceres acadêmicos rigorosos: editorial + técnico, com gráficos e PDFs polidos
Ver no GitHubUma skill de Claude Code que transforma manuscritos em pareceres de revisão por pares rigorosos, aplicando dupla lente (editorial + técnica) e exportando em Markdown, Typst e PDF. Inclui biblioteca reutilizável de gráficos CeTZ e templates de diagramas Mermaid para evidências visuais de qualidade profissional.
Instalação
npx convez-ai ChanMeng666/academic-paper-review-skill1 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
- Gerar pareceres de revisão por pares para periódicos e conferências
- Auditar manuscritos com rigor editorial e técnico simultâneos
- Recompor e verificar estatísticas de papers antes de rejeição/aceitação
Veredito

Skill especializada e bem documentada para revisar papers com rigor acadêmico real — dupla lente editorial+técnica, gráficos de qualidade e PDFs polidos. Recomendado para pesquisadores, editores e revisores que precisam automatizar pareceres consistentes e profissionais.
README
A tradução para o português deste README ainda está na fila.
Academic Paper Review — a Claude Code skill
A reusable Claude Code skill that turns manuscripts into a credible peer review and ships it as Markdown, a Typst source, and/or a typeset PDF.
It encodes a workflow refined on real reviews: read each paper through two lenses at once — editorial (consistency, statistics, citations, writing) and technical/domain (methods vs. established practice; missing baselines, gameable metrics, eval bugs) — verify load-bearing claims against primary sources, recompute the paper's statistics, and produce a deliverable in the right register for the audience.
What it produces
- Reviewer letter — author-facing, narrative prose (~2–3 paragraphs per paper), the way a human referee actually writes. No severity badges, boxes, or checklists (that scaffolding is what makes a review read as machine-generated).
- Audit report — internal/triage, severity-tiered findings, tables, optional data charts.
Either renders to a polished PDF via bundled Typst templates.
Figures are first-class. A reusable CeTZ chart library (grouped-bars, log-bars,
diverging-bars, number-line, stacked-bars) draws publication-quality vector charts at compile
time, and Mermaid diagram templates (with a shared palette + batch render script) cover
structure/architecture figures — so a review can be as evidence-rich as the audience warrants.
Repository layout
chanmeng-academic-paper-review/
├── README.md ← you are here
└── skills/
└── chanmeng-academic-paper-review/
├── SKILL.md ← the skill (entry point)
├── references/
│ ├── review-dimensions.md ← both lenses' checklists + RAG/IR domain knowledge
│ ├── output-formats.md ← letter vs audit: structure, voice, skeletons
│ └── typst-rendering.md ← md→typst→pdf pipeline, chart gallery, gotchas
├── scripts/
│ └── render-mermaid.sh ← batch-render all .mmd → high-res PNG
└── assets/
├── README.md ← figure toolkit guide
├── review-letter.typ ← narrative letter template (configurable reviewer block)
├── audit-report.typ ← structured report template (severity tiers)
├── cetz-charts.typ ← reusable CeTZ chart library (5 chart types)
├── mermaid/ ← reusable diagram templates (shared palette)
└── figures/ ← rendered diagram PNGs
Install
Any agent, with the skills CLI
Installs into Claude Code, Cursor, Codex, GitHub Copilot, Gemini CLI and the other agents the CLI supports. Listed on skills.sh.
npx skills add ChanMeng666/chanmeng-academic-paper-review
As a Claude Code plugin (recommended)
/plugin install chanmeng-academic-paper-review --marketplace ChanMeng666/chanmeng-academic-paper-review
On Claude Code older than 2.1.275, add the marketplace first:
/plugin marketplace add ChanMeng666/chanmeng-academic-paper-review
/plugin install chanmeng-academic-paper-review@chanmeng-academic-paper-review
Manual install
Clone the repo, then make the inner skill folder discoverable to Claude Code, which looks for
personal skills in ~/.claude/skills/<name>/. A symlink is preferred so edits stay live.
git clone https://github.com/ChanMeng666/chanmeng-academic-paper-review.git
cd chanmeng-academic-paper-review
macOS / Linux:
ln -s "$(pwd)/skills/chanmeng-academic-paper-review" ~/.claude/skills/chanmeng-academic-paper-review
Windows (PowerShell, run as Administrator for the symlink):
New-Item -ItemType SymbolicLink `
-Path "$env:USERPROFILE\.claude\skills\chanmeng-academic-paper-review" `
-Target "$(Resolve-Path .\skills\chanmeng-academic-paper-review)"
# …or just copy if you'd rather not symlink:
Copy-Item -Recurse ".\skills\chanmeng-academic-paper-review" "$env:USERPROFILE\.claude\skills\chanmeng-academic-paper-review"
To scope it to a single project instead, place (or link) the folder under that project's
.claude/skills/ directory. Restart Claude Code (or start a new session) so it picks up the skill.
Dependencies
- Typst CLI — for PDF output (
typst compile). The bundledtypstskill can install/help if needed. - Mermaid CLI (
mmdc) — optional, only for flow diagrams in figure-rich variants. - The Typst templates fetch CeTZ (
@preview/cetz) automatically on first compile (needs network once).
Use it
Just ask Claude Code in natural language — the skill triggers on review/audit/referee requests:
"Review the three papers in
./papers/and give me a reviewer letter as a PDF.""Audit this manuscript before I submit — check the stats and citations, flag any strawman baselines."
"Critique the methodology of
draft.mdfrom a RAG perspective and write up the findings."
Claude will read the papers (never editing them), apply both lenses, verify the load-bearing claims, and produce the deliverable you asked for.
Personalize the reviewer
Open assets/review-letter.typ and edit the CONFIG block at the top (name, affiliation, the
optional "About the reviewer" note and links). Those values flow into the letterhead, signature, and
footer.
Why two lenses
An editorial pass alone yields a proofread; a domain opinion alone yields hallway chatter. Authors
trust a review that demonstrably did both — which is why references/review-dimensions.md pairs a
general editorial checklist with concrete domain critique (with a deep RAG/IR/dense-retrieval section:
ANN baselines, MMR, relevance vs. diversity metrics, the bge query-prefix bug, and more).
Contributing
Issues and pull requests are welcome — new chart types for cetz-charts.typ, additional Mermaid
templates, sharper domain checklists for fields beyond RAG/IR, or refinements to the review voice.
License
MIT © Chan Meng