Claude Skill Paper Ecosystem
shimo4228/claude-skill-paper-ecosystem
Escrita e revisão acadêmica com orquestração de 5 subagendes especializados
Ver no GitHubBundle de skill + agents para escrita e revisão de artigos acadêmicos (preprints, position papers, journals). Orquestra um skill de redação, um de revisão e cinco subagendes especializados (fidelidade de fonte, consistência vocabular, clareza, formatação de citações) que trabalham em paralelo para garantir qualidade de publicação.
Instalação
npx convez-ai shimo4228/claude-skill-paper-ecosystem1 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
- Redigir e revisar preprints para arXiv/SSRN com verificação de fidelidade de fonte
- Garantir consistência vocabular e formatação de citações em artigos acadêmicos
- Automatizar loop completo de escrita→revisão paralela para papers de pesquisa
Veredito

Arquitetura técnica sólida: primeiro bundle a empacotar skills + agents juntos via install.sh. Reduz fricção de setup multi-componente, mas requer Claude Code e entendimento da hierarquia de orquestração. Ideal para pesquisadores que precisam rigor editorial automatizado.
README
A tradução para o português deste README ainda está na fila.
claude-skill-paper-ecosystem
An Agent Skill bundle for writing and reviewing academic papers — position papers, preprints, and journal-style articles aimed at SSRN / arXiv / Zenodo / journal venues. It is the academic counterpart to claude-skill-writing-ecosystem (human-facing blog/essay) and llms-txt-writer (AI-facing docs).
Unlike a single-skill repo, this bundles an orchestrator skill + a draft skill + five reviewer subagents so the whole write→review loop installs as one unit.
| Component | Kind | Role |
|---|---|---|
paper-ecosystem | skill | Orchestrator. Holds the canonical rules (Source Fidelity, Vocabulary Consistency, Academic Voice, Reader Clarity, Citation Format) and the role-boundary map. |
paper-writing | skill | Draft procedure. Title → outline → section drafting → abstract → references, with claim↔cite 1:1 mapping enforced. Inherits the orchestrator's rules. |
paper-reviewer | agent | Argument flow, claim sharpness, evidence-claim alignment, section structure. |
source-fidelity-checker | agent | Reads each cited primary source directly; flags drift between paper claim and source. |
vocabulary-consistency-checker | agent | Term-definition consistency and sub-classification discipline. |
clarity-reviewer | agent | First-contact reader clarity (coined-term budget, title-axis carry-through, no insider context). |
citation-formatter | agent | In-text ↔ reference 1:1 mapping, format consistency, DOI/arXiv validity. Final gate. |
Why bundle agents? The five reviewer agents read their canonical rules from the
paper-ecosystemskill (~/.claude/skills/paper-ecosystem/SKILL.md). The skill and its agents must be installed together — that is what this repo packages. Agent Skills are an open cross-tool standard, but Claude Code subagents are Claude-Code-specific, so this bundle targets Claude Code.
Install
This repo bundles both skills and the agents they depend on.
Option A — one command (recommended)
git clone https://github.com/shimo4228/claude-skill-paper-ecosystem
cd claude-skill-paper-ecosystem
./install.sh
Copies every skills/* into ~/.claude/skills/, every agents/*.md into ~/.claude/agents/, and runs uv sync for any skill that declares Python dependencies. Existing files are backed up to *.bak-<timestamp> before being replaced (use --force to skip backups, --dry-run to preview).
Option B — manual (full control)
# Skills
cp -r skills/paper-ecosystem ~/.claude/skills/paper-ecosystem
cp -r skills/paper-writing ~/.claude/skills/paper-writing
# Agents (required — the reviewers read the skill's canonical rules)
cp agents/*.md ~/.claude/agents/
These skills are documentation-only; no uv sync step is required.
SkillsMP
/skills add shimo4228/claude-skill-paper-ecosystem
Caveat: SkillsMP installs
skills/only — it does not installagents/. After/skills add, copy the agents manually:cp agents/*.md ~/.claude/agents/(or just use Option A). Without the agents, the orchestrator has no reviewers to delegate to.
How It Works
- Draft with
paper-writing— establish the primary-source list first, then draft section by section, holding a claim↔cite 1:1 mapping. - Review in parallel — run
paper-reviewer,source-fidelity-checker,vocabulary-consistency-checker, andclarity-revieweras orthogonal lenses after a section or full draft. - Final gate — run
citation-formatterlast, once content review has settled, to verify the reference apparatus and DOI/arXiv validity. - Deposit is a separate stage (not in this repo) — see
release-doifor repo-level Zenodo deposits; apaper-depositskill covers paper-level deposits.
When to Use
- Writing or reviewing a position paper / preprint / journal-style article (SSRN / arXiv / Zenodo / journal).
Do not use for:
- Human-facing blog posts / essays / newsletters →
claude-skill-writing-ecosystem - AI-facing docs (
llms.txt/ FAQ / glossary) →llms-txt-writer
Requirements
- Claude Code (or another agent product that supports the Agent Skills standard and subagents)
- No runtime dependencies (documentation-only skills)
Related skills (siblings)
claude-skill-writing-ecosystem— human-facing writing & review (also bundles its agents)llms-txt-writer— AI-facing documentsreadme-writer— human-facing READMEs
About this skill
This bundle is a component of the Authorship Strategy research line (DOI 10.5281/zenodo.20263316) maintained by @shimo4228. It operationalizes the academic-publishing surface of the program's diffusion tactics: harness-neutral papers deposited as citable records.
Frontmatter note: each SKILL.md carries two harness extensions beyond the Agent Skills spec — user-invocable (Claude Code slash-invocation) and origin (author provenance tracking). Both are tolerated as extra fields by the standard.
License
MIT
日本語
学術論文(position paper / preprint / journal 向け)を書く・レビューするための Agent Skill バンドルです。人間向け blog/essay の writing-ecosystem、AI 向け doc の llms-txt-writer に対する学術版にあたります。
単一スキルの repo と違い、orchestrator skill + draft skill + 5 つの reviewer subagent を一括同梱し、write→review ループ全体が 1 単位で入ります。
paper-ecosystem(skill)— orchestrator。canonical rules(Source Fidelity / Vocabulary Consistency / Academic Voice / Reader Clarity / Citation Format)と役割境界の正本。paper-writing(skill)— draft 手順。claim↔cite の 1:1 mapping を強制。- reviewer agents 5 種 — paper-reviewer / source-fidelity-checker / vocabulary-consistency-checker / clarity-reviewer / citation-formatter。
なぜ agent を同梱するか: 5 つの reviewer agent は canonical rules を paper-ecosystem skill から読みます。skill と agent はセットで入れる必要があり、それを本 repo が梱包します。Agent Skills はオープンな cross-tool 標準ですが、Claude Code の subagent は Claude Code 固有なので、本バンドルは Claude Code 向けです。
インストールは ./install.sh(推奨)か手動 cp。SkillsMP は agents/ を入れないので、その場合は cp agents/*.md ~/.claude/agents/ を実行してください。
詳細は各 SKILL.md を参照してください。