Academic Paper Writing Skill

Evidence-first academic AI workflow that reduces hallucinated citations, fake evidence, unsupported claims, and integrity drift before prose generation. 证据优先的学术写作 skill：写论文、综述、rebuttal 和毕设前先核验证据，降低虚构引用、虚构结果和无证据 claim 风险。

简体中文 · English

快速开始 · Demo · 产物 · 为什么需要 · 工作流 · 完整性规则

🔥 最新定位

这个仓库不是一段“学术写作提示词”，也不是帮你把论文说得更漂亮的包装器。它是一套可安装的 Agent Skill：把论文、文献综述、研究写作、实验规划、rebuttal 和毕业设计拆成可验证的模块，让 agent 在写作前先处理来源、证据、claim、实验和完整性风险。

它面向的是 evidence-first academic AI assistance：先建立可审查的证据包，再写正文、rebuttal 或答辩材料。本科论文 / 毕业设计也可以使用这套证据优先流程，作为兼容场景保留。

适用于 Codex、Claude Code，以及遵循 Agent Skills 目录结构的其他 agent。Claude Code 可以用同一个 SKILL.md，但安装路径和直接调用方式不同，见下方 Codex / Claude Code。

⭐ Star this if you...

• need academic AI help but do not want fake citations, fake SOTA, or unsupported claims;
• want literature matrices, claim ledgers, experiment matrices, and integrity checks before prose;
• support thesis / paper workflows where unknown requirements stay unknown until verified.

🚫 Not for

• ghostwriting papers without evidence review;
• inventing citations, SOTA, experiments, school rules, or advisor requirements;
• replacing human decisions on authorship, ethics, submission, or school compliance.

✨ 为什么需要

学术写作最容易出问题的地方不是措辞，而是证据链断裂：论文没核验、SOTA 被夸大、实验结果被提前写成已完成、学校模板或导师要求被凭空补全。

Academic Paper Writing Skill 的原则是先建立可审查的中间产物，再进入最终 prose。它默认使用 primary sources、claim-to-evidence mapping 和完整性检查，降低虚构引用、虚构结果、过度承诺和毕业要求误判的风险。

🎬 Demo: Evidence-first workflow

Input:

使用 $academic-paper-writing-skill 帮我把一个 RAG-based academic assistant 题目整理成毕业论文开题范围和证据包，不要编造论文或学校要求。

Clone 后可以先跑一个离线 demo workspace：

python scripts/demo_academic_workflow.py --mode undergraduate_thesis --out demo_workspace

The skill first creates:

1. evidence/requirement_discovery_log.md
2. evidence/scope_ladder.md
3. evidence/graduation_evidence_map.csv
4. claim_ledger.csv
5. integrity_checklist.md

Only then does it draft proposal prose, related work, or rebuttal text. A fuller walkthrough is in examples/undergraduate-thesis-proposal-demo/README.md.

See filled output samples:

• examples/outputs/rag-evaluation-literature-matrix.sample.csv
• examples/outputs/rag-evaluation-claim-ledger.sample.csv
• examples/outputs/rag-evaluation-novelty-check.sample.md
• examples/outputs/undergraduate-thesis-evidence-map.sample.csv
• examples/outputs/sample-source-note.md
• examples/generated-demo-workspace/README_NEXT_STEPS.md
• examples/outputs/evidence-status-summary.generated.md
• examples/outputs/pre-prose-check.generated.md
• examples/outputs/claim-blocker-report.generated.md

Machine-check the integrity boundary:

python scripts/summarize_evidence_status.py demo_workspace
python scripts/pre_prose_check.py demo_workspace --expect-block
python scripts/validate_demo_workspace.py demo_workspace --mode undergraduate_thesis

📦 What you get

🧾 为什么不直接让 AI 写论文？

👨‍💻 适用场景

🎯 能力亮点

📦 快速开始

Codex 用户

mkdir -p ~/.agents/skills
git clone https://github.com/xcl2005/academic-paper-writing-skill.git ~/.agents/skills/academic-paper-writing-skill

Windows PowerShell:

New-Item -ItemType Directory -Force -Path "$HOME\.agents\skills"
git clone https://github.com/xcl2005/academic-paper-writing-skill.git "$HOME\.agents\skills\academic-paper-writing-skill"

调用方式：

使用 $academic-paper-writing-skill 帮我做这篇论文的文献矩阵和 novelty 检查。

Claude Code 用户

mkdir -p ~/.claude/skills
git clone https://github.com/xcl2005/academic-paper-writing-skill.git ~/.claude/skills/academic-paper-writing-skill

项目级安装：

mkdir -p .claude/skills
git clone https://github.com/xcl2005/academic-paper-writing-skill.git .claude/skills/academic-paper-writing-skill

直接调用：

/academic-paper-writing-skill 帮我把毕业设计题目拆成开题报告范围和证据清单

初始化工作区

# 创建研究论文工作区
python scripts/init_project.py --out paper_workspace --type research_paper

# 创建本科论文 / 毕业设计工作区
python scripts/init_project.py --out thesis_workspace --type undergraduate_thesis

# 验证 skill 结构
python scripts/validate_skill.py

# 生成可检查的 demo workspace
python scripts/demo_academic_workflow.py --mode undergraduate_thesis --out demo_workspace

🔁 Codex / Claude Code

Claude.ai / Claude API 通常需要把 skill 作为自定义 skill 上传或通过 Skills API 注册；GitHub clone 路径主要面向本地 Claude Code 与 Codex。

🧭 工作流

Topic / Draft
  -> Source Verification
  -> Literature Matrix
  -> Claim Ledger
  -> Experiment / Evidence Matrix
  -> Integrity Check
  -> Final Draft / Rebuttal

控制流程

模块路由

🗂️ 产物

Markdown、YAML 和 CSV 是规范工作格式；Word、PDF、Excel、slides 是导出格式。

🛡️ 完整性规则

Integrity-first by default

This skill blocks or qualifies final prose when:

• citations are unverified;

• novelty claims are not checked against prior work;

• planned experiments are written as completed results;

• local thesis requirements are missing but treated as confirmed;

• strong claims are not mapped to sources, experiments, implementation evidence, or official requirements.

• 不编造论文。

• 不编造 SOTA。

• 不编造结果。

• 不编造学校、导师、rubric、模板、答辩或工作量要求。

• 优先使用 primary sources。

• 强 claim 必须映射到证据。

• 投稿、署名、伦理、学校合规必须由人复核。

🛠️ 质量检查

python scripts/validate_skill.py
python scripts/validate_readme_quality.py

📁 仓库结构

.
|-- SKILL.md
|-- skill_manifest.yaml
|-- .github/
|-- docs/
|-- modules/
|-- templates/
|-- schemas/
|-- scripts/
|-- examples/
|-- README.md
`-- README_EN.md

🔎 搜索关键词

Academic writing AI, research paper workflow, literature review matrix, thesis writing assistant, graduation project, rebuttal assistant, claim evidence mapping, research integrity, Codex skills, Claude Code skills, Agent Skills.

📄 许可证

MIT