Skip to content

1235357/research-english-academic-paper-writing-guide

v4.7.0MIT

Full 13-lecture research and academic-writing reconstruction, both complete independent versions, 19 execution playbooks and evidence-integrity workflows.

GitHubOpen Website

Research & English Academic Paper Writing Guide — Agent Skill

基于中科院信息工程研究所于静老师公开课程 《科研与英文学术论文写作指南》 （截至 13 讲）构建的 Agent Skill 与全课程 Markdown 课堂重构。

An Agent Skill (plus a full-course Markdown reconstruction) built from Yu Jing's public 13-lecture course Guide to Scientific Research and English Academic Paper Writing (Institute of Information Engineering, Chinese Academy of Sciences). Course homepage: https://mmlab-iie.github.io/course/

v4.0.0 merge note: this package now merges in a second, independently-authored skill built by a different person from the same course material. This is an organic splice, by explicit design — not a hand-merged rewrite, and not a separate cross-reference file: the companion skill's own SKILL.md body is appended in full to the end of this package's SKILL.md ("Version 2" section), and the companion skill's own 13 lecture reconstructions are each appended in full to the end of this package's corresponding references/lectures/*.md file. The redundancy is intentional — two independent full passes over the same course, physically present together, give an agent more complete coverage than either alone. This package's trigger is also now always-on (fires in every conversation, not only paper-writing ones) — see SKILL.md's frontmatter description and references/provenance.md → "Merge history" for the full rationale and everything else the merge did.

这个 skill 能做什么 / What it does

装上之后，Agent 在以下任务中会自动套用课程的方法论（问题链、A类论文标准、 问题-方法-实验互相呼应、简洁严谨的英文规范等）：

  • 论文各部分写作与修改：立意 / 标题 / 摘要 / 引言 / 相关工作 / 方法 / 实验 / 结论 / 致谢 / 参考文献
  • 研究动机图与模型框架图的设计、评审与迭代，含占位符强制政策 （禁止用 TikZ 临时糊图顶替、必须维护配套的图注/手绘说明文档）、 参考图检索方法论与 AI 生成图痕迹自查清单
  • 表格类型与排版设计（定位表/数据集特征表/主实验表/跨协议对比表/消融表 该怎么选、怎么排版，列数爆炸怎么收敛，附录与正文该怎么取舍； benchmark 核心表该在"方法广度"还是"数据集广度"上做文章）
  • 学术英文规范：简洁与严谨、术语、符号、图表、引用、学术道德
  • "快速读者"原则：标题/摘要/图表本身能否在不读正文的情况下讲清楚一个 可信、完整的故事——审稿人真实阅读行为的调研依据与自查清单
  • AI 检测分数的正确应对方式（不是调统计特征去规避检测，是理解检测器 对非母语写作者的已证实偏见、且分数永远不能改变事实或结论）
  • 创新性研究点的发掘与三级审计（真实需求 → 科学问题 → 解决方法）
  • 找论文与文献调研（三阶段工作流）
  • 导师/审稿反馈证据化修订（claim ledger、表内引用、可比性契约、 figure evidence card、源码/PDF/结果同步与匿名发布安全）
  • 投稿冲刺阶段纪律（冻结基线只读、并发编辑、收据体系、署名诚信、 投稿表单一致性、按栏而非按页量测版面）
  • 按讲原封复现课堂内容（每讲一个文件，内含两份独立作者各自完成的 逐页 slides 文字化 + 课堂讲解复现，前后并行呈现、互为校验）
  • 本 skill 始终在线（每次对话都会加载，不论话题是否与论文写作相关）

仓库结构 / Layout

├── README.md
├── docs/
│   └── course-full-reconstruction.md   # 13 讲全课程课堂重构·整合版(每讲含双版本, ~1.3MB, 纯 Markdown)
└── skills/
    └── research-english-academic-paper-writing-guide/
        ├── SKILL.md                    # 入口：Version 1(核心教义+任务路由表)+ Version 2(附属分支全文，拼接于后)
        ├── agents/openai.yaml          # Codex 界面元数据（可选）
        ├── scripts/                    # 可选辅助脚本（Python3 标准库，无三方依赖）
        │   ├── check_title.py          #   标题红旗检查
        │   ├── new_research_point_canvas.py  # 研究点画布生成
        │   └── measure_prose_rhythm.py #   句长节奏/单调性诊断（无检测器目标区间）
        └── references/                 # 渐进式加载的三层知识
            ├── course-full-reconstruction.md  # skill 内置完整课程重构（每讲含双版本）
            ├── quick-reference.md      #   全课程速查规则
            ├── lecture-index.md        #   13 讲索引（官方视频 + PPT 链接）
            ├── provenance.md           #   溯源、更正记录、官方 PDF 校验和与合并历史
            ├── playbooks/              #   19 个任务剧本（工作流+清单+模板+反模式）
            └── lectures/               #   13 讲完整课堂重构：每个文件 = Version 1(主版本) + Version 2(附属分支独立复现)拼接

设计要点：SKILL.md 是严格入口协议 + 附属分支全文合并（Version 1 + Version 2）； 完整课程重构已经复制进 skill 的 references/course-full-reconstruction.md， Agent 对规则有疑问时必须优先查阅它。任务执行摘要仍按需放在 references/playbooks/ 与 references/quick-reference.md。

安装 / Install

本 skill 遵循开放的 Agent Skills 格式（一个含 SKILL.md 的文件夹），可用于 任何支持该格式的产品。将 skills/research-english-academic-paper-writing-guide/ 整个文件夹复制到对应产品的 skills 目录即可：

This skill follows the open Agent Skills folder format: the installable unit is skills/research-english-academic-paper-writing-guide/, which contains the required SKILL.md file plus optional references, metadata, and helper scripts. Copy that folder into the skills directory used by your agent runtime.

平台安装位置
Claude Code（CLI / VS Code / JetBrains / 桌面端 / web）~/.claude/skills/（个人）或项目内 .claude/skills/
OpenAI Codex~/.codex/skills/（含 agents/openai.yaml 界面元数据）
ChatGPT（Business/Enterprise 网页端，支持 skills 的场景）按产品入口上传 skill 文件夹（保持文件夹结构）
OpenClaw / Trae / Qoder / CodeBuddy 等复制到各自的 skills 目录（Agent Skills 兼容格式）
skills CLInpx skills add 1235357/research-english-academic-paper-writing-guide@research-english-academic-paper-writing-guide

兼容性保证：SKILL.md frontmatter 仅保留必需的 name/description 字段；全部相对路径；ASCII 文件名；无符号链接；无网络依赖；脚本为可选项 （不执行脚本时清单同样可人工套用）。

Compatibility notes: SKILL.md keeps only the required name and description frontmatter fields; all runtime paths are relative; installable filenames are ASCII; no symlinks, raw PDFs, raw transcripts, or network calls are required; helper scripts use only the Python 3 standard library.

全课程整合文档 / Full-course reconstruction

docs/course-full-reconstruction.md 将 13 讲课堂重构原封收录为单个 Markdown 文档：每讲含逐页幻灯片文字复现、 视觉版式文字描述（不嵌图）、结合 transcript 的课堂讲解复现、方法论提炼与 可执行清单，另附全课程通用框架、目录、溯源更正记录与官方 PDF 校验和。 同一份 Agent 优化后的全文也发布在 skill 内部： skills/research-english-academic-paper-writing-guide/references/course-full-reconstruction.md。

The integrated Markdown file keeps all 13 sessions in one document. Each session includes slide-by-slide text reconstruction, textual visual/layout descriptions instead of embedded images, classroom narration reconstructed from transcripts, distilled methodology, and actionable checklists. The same agent-optimized full reconstruction is included inside the installable skill package at references/course-full-reconstruction.md.

校验 / Verification

发布前或安装前可运行：

Before publishing or installing, run:

python3 scripts/validate_package.py
python3 scripts/test_review_feedback_evidence_playbook.py
python3 scripts/test_professor_feedback_operationalization.py
python3 scripts/test_dataset_census_experiment_evidence.py
python3 skills/research-english-academic-paper-writing-guide/scripts/check_title.py \
  "A Novel Framework for Visual Question Answering"
python3 skills/research-english-academic-paper-writing-guide/scripts/new_research_point_canvas.py

scripts/validate_package.py checks required files, skill frontmatter, lecture/playbook coverage, local Markdown links, raw asset leakage, likely credential leakage, and Python syntax without network access. The three feedback regression scripts parse the playbook's heading hierarchy, workflow order, section-local sentence/paragraph rules, and exact ledger/table schemas. They also reject explicit contradiction patterns. Each script runs in-memory keyword-bag, section-relocation, and negation-reversal mutations; a mutation that passes makes the regression script fail. These checks validate the documented evidence contract. They do not validate a manuscript's actual citations, experiments, PDF, archive contents, or venue compliance.

版权与致谢 / Copyright & credits

  • 课程内容版权归 于静老师（中科院信息工程研究所）所有；课程素材致谢 丁阳、庄佳敏、林鑫杰、唐源民、屈详颜、李一立。

  • 本仓库为学习用途的派生笔记：官方 slides PDF 与原始 transcript 不在 本仓库中再分发，官方 PDF 请从课程主页下载（references/lecture-index.md 含逐讲链接；references/provenance.md 含 SHA-256 校验和）。

  • v4.0.0 起，本仓库合并了另一位作者基于同一课程材料独立整理的 skill 版本； 合并明细见 references/provenance.md → "Merge history"。

  • 公开传播或商业使用前，请先征得课程作者许可，并保留课程署名。

  • 仓库原创包装、脚本与元数据的授权见 LICENSE；课程归属、 溯源与再分发边界见 NOTICE；版本记录见 CHANGELOG.md。

  • Repository-original packaging, scripts, and metadata are covered by LICENSE. Course attribution, provenance, and redistribution boundaries are documented in NOTICE. Release history is in CHANGELOG.md.

视频课程 / Videos

全部课程视频见于老师 B 站主页：https://space.bilibili.com/301285406， 逐讲链接见 references/lecture-index.md。

4.7.0：Skill／插件发布与完整安装

原有内容完整保留。本次新增安装和平台兼容说明、正式发布附件、标准Skill ZIP、Agent Plugins／Claude插件清单、社区npx与离线字节校验安装器。发布层版本与旧正文里的历史版本号分别记录。