Best-Paper-Writing-Skill

# Best-Paper-Writing-Skill

Writes a draft of a research paper in the style of award winners, from a form or a pasted abstract: abstract, introduction, methods, experiments and figure plans. Works in Claude and Codex.

Made by gulucaptain on GitHub. These notes are Mindket's, written from its README; the project's own files and license are in its repository.

## About

What it is: Best-Paper-Writing-Skill produces a draft of a research paper that follows the writing habits of award-winning papers. You give it your research through a form or a pasted abstract, and it writes the abstract, introduction, methods and experiments, and plans the figures. One skill works in both Claude and Codex.

Who it suits: It suits researchers who write conference papers and want a structured first draft instead of a blank page. Its writing rules come from 60 award-winning papers at CVPR, ICCV and NeurIPS.

```table
Industry | Who uses it | How they use it | Example
Academic research | PhD students and early-career researchers | Turn a one-line summary of the work into a draft abstract and introduction | Pasting an abstract and asking for the other sections
Research labs | Teams preparing a conference submission | Fill the intake form and get methods, experiments and a claim-and-evidence table | Writing into the CVPR author kit template
Paper revision | Authors with an existing abstract | Get the abstract split into fields, with a table of edits and the reason for each | Accepting or rejecting each suggested change
```

You can skip it if you write in areas the corpus lacks. The maintainer names NLP, robotics and systems as missing.

Where to use it:

- Claude Code: run the install script for claude, or copy the skill folder into ~/.claude/skills/
- Claude.ai and Cowork: zip the skill folder and upload it on the Skills page
- Codex: run the install script for codex, or copy the folder into ~/.agents/skills/, and call it with $bestpaper-writing
- A browser: the intake form opens offline from an HTML file
- LaTeX: the draft can be written into a template such as the CVPR author kit

```table
Use case | Who | What they get back
Draft from a form | Authors | Title, abstract, introduction, methods and experiments as Markdown, or written into a LaTeX template
Draft from an abstract | Authors | A full draft plus a table of changes to the abstract, with the original sentence, the new one and the reason
Figure plan | Authors | A table of what each figure carries, where it goes and what material it needs, plus drawn overviews
Author report | Authors | A report.md with core claims, a claim-and-evidence table, a to-do list and points for the Limitations section
```

How to set it up: You need a few things in place first.

- Claude Code, Claude.ai or Cowork, or Codex
- git, to clone the repository
- Python 3, for the helper scripts

1. Clone the repository from GitHub and move into its folder.
2. Run the install script for your agent: claude, codex or all.
3. Add the --project option to install into the current project folder instead.
4. Restart your agent if the skill does not show up.

After cloning, move into the folder and pick the install line for your agent:

```sh
cd Best-Paper-Writing-Skill
./install.sh claude
./install.sh codex
./install.sh all
```

How to use it: There are two ways to start. The first is the intake form, an HTML file you open in a browser. It works offline, saves a draft as you type and can export or import JSON.

Only five items are required: a one-line description of the work, the task, when existing methods fail, the method's parts and the headline result. More detail gives a better draft. Send the exported JSON to your agent and ask it to write the title, abstract, introduction, methods and experiments with the skill.

The second way is to paste a full abstract. The skill splits it back into fields and tells you which are present, vague or missing. It asks about the gaps once, then writes the draft. Your numbers and terms stay as you wrote them.

```table
Tool | What it does
intake_form.html | An offline form for your research details, with autosave and JSON export and import
Built-in draft checker | Compares a draft with corpus statistics and flags AI tone, dashes, stacked aphorisms and invented numbers
check_figures.py | Checks caption length, first sentences, references, vector figures and placeholders
report.md | The author report with core claims, a claim-and-evidence table, a to-do list and points for Limitations
```

Example: You fill in the five required items for a method paper and ask for the abstract and introduction. You get a draft in which each component is explained as a need, then a mechanism, then a property. A report.md comes with it. Numbers, baselines and citations you did not give stay as [TODO].

For newcomers and experts: If you are new, open the intake form and fill only the five required items. Run the skill and answer its one round of questions. Each [TODO] in the draft marks something you still have to supply.

If you know the field, the method matters. The rules come from sentence-level labelling and statistics of 60 winning papers, and each rule has an upper and a lower bound instead of pulling toward the average. Papers are routed into eight types, and each type has its own argument route. The rules were fitted on 55 papers, and five papers from 2024 were held out for testing only.

```flow
Fill the form or paste an abstract
One round of questions on gaps
Judge the paper type
Draft from exemplar passages
Run the built-in checks
Draft and report.md
```

Good to know:

- License: the original content, meaning notes, templates and scripts, is CC BY 4.0
- Exemplar quotes are short teaching excerpts that stay with their authors, so do not copy them into a new paper
- The repository holds no paper PDFs or full text
- The corpus has 60 papers from 2021 to 2025 at CVPR, ICCV and NeurIPS
- This is version v1.2, with figure planning added in v1.1
- The workflow and templates build on an MIT-licensed project by Master-cai
- The README compares how often sentences end in a participle phrase, a habit of generic AI drafts

```chart
Title: Sentences ending in a participle phrase (approximate share)
Generic AI drafts: 32%
Award-winning papers: 8%
```

## On GitHub, 7 November 2026

- Stars: 280
- Forks: 17
- Watching: 4
- Open issues and pull requests: 1
- Language: TeX
- License: see the repository
- Topics: academic-writing, agent-skills, claude-skills, codex, cvpr, eccv, iccv, iclr, icml, nips, paper-writing

## Links

- Code: https://github.com/gulucaptain/Best-Paper-Writing-Skill
- Releases: https://github.com/gulucaptain/Best-Paper-Writing-Skill/releases
- Issues: https://github.com/gulucaptain/Best-Paper-Writing-Skill/issues
- Star history: https://www.star-history.com/#gulucaptain/Best-Paper-Writing-Skill&Date

## Get the code

```
git clone https://github.com/gulucaptain/Best-Paper-Writing-Skill.git
```

Then follow the setup in its README. Nothing in this kit runs by itself.

README.md  7 KB