Best Paper Writing Skill
A skill that drafts an ML paper after the patterns of CVPR, ICCV and NeurIPS best papers and draws its figures and schemes.
- 359 ★ on the repository
- Last commit
Minimal access
Why this level
- The skill writes text and draws figures, with helper scripts checking the draft locally.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/best-paper-writing-skillDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewMinimal accessThe tool runs nothing and makes no network calls.
- The skill writes text and draws figures, with helper scripts checking the draft locally.
This is third-party code. Review the repository files before installing.
What it does
Best-Paper-Writing-Skill turns a research description or a draft abstract into a title, abstract, introduction, method and experiments in the style of CVPR, ICCV and NeurIPS award winners. The rules come from a sentence-by-sentence annotation of 55 papers (60 analysed in total). Introductions credit specific prior work, name the gap and its cause, and methods unfold as need, mechanism, property. The skill plans and draws figures: a planning table shows which claim each picture carries, picks a Figure 1 type and applies one style. It routes by eight paper types, from method improvement to dataset. A built-in checker flags typical AI wording, stray dashes and invented numbers. Next to the draft it writes a `report.md` with the core claim, an evidence table and points for the Limitations section.
Who it is for. Machine learning and computer vision researchers preparing a conference paper who want a strong draft rather than a paraphrase.
Good fit when
- A short description of the work exists and an abstract and introduction must come from it.
- A finished abstract goes through dissection: which fields exist, which are vague, which are missing.
- Figure 1 and a method overview have to be planned in one style.
- Notes for the Limitations section are collected in a separate `report.md`.
Not a fit when
- The paper lies outside machine learning and computer vision.
- Ready results and citations are expected, while the skill invents none and leaves `[TODO]`.
- Only spelling checks without restructuring are wanted.
Example request
Use Best-Paper-Writing-Skill to write the Title, Abstract and Introduction from this method description and these results.Limitations
The skill targets ML and computer vision papers: rules come from the CVPR, ICCV and NeurIPS winner corpus, with five 2024 papers held out for testing. Numbers, baselines and citations you did not supply stay as `[TODO]` and have to be filled in by you. Helper scripts need Python 3. The main documentation is in Chinese, with an English README as well. The text still needs proofreading, since the skill gives a draft rather than a finished paper.
How to disable. Delete the `bestpaper-writing` folder from `~/.claude/skills/` or `~/.agents/skills/` and restart the agent.
FAQ
How do I install the skill?
Clone the repository and run `./install.sh claude` for `~/.claude/skills/` or `./install.sh codex` for `~/.agents/skills/`. `all` installs both and `--project` installs into the current project.
How do I pass in my research description?
Fill in the offline form `skills/bestpaper-writing/assets/intake_form.html`, export the JSON and send it to the agent, or paste a finished abstract.
What does the skill do with a finished abstract?
It splits the abstract into fields, reports which exist, which are vague and which are missing, asks about the gaps and returns a table of edits.
Does it invent numbers and citations?
No. Anything you did not supply stays as `[TODO]`, and the checker looks for invented numbers in the draft.
Does it work in Claude.ai?
Yes, zip the `bestpaper-writing` folder and upload it on the Skills page. Codex has an explicit `$bestpaper-writing` call.
Discussion
No questions or reviews yet.
Related
SkillsA skill that checks a manuscript against practices of 883 ICLR, ICML and NeurIPS Oral papers and suggests edits
A skill for writing and revising ML, CV and NLP papers: section structure, paragraph flow and reviewer-style self-review
A skill turns an ML research repository into a conference paper draft in LaTeX with a claim-evidence map.
The paper-writing skill checks a LaTeX paper against MLNLP community rules: equations, tables, references and academic English.
A Claude Code skill for research papers: a five-stage pipeline, section rules, a style audit and an independent red-team review.
A skill for researchers who draft in Chinese: drafts and results become an English manuscript, with citations and claims checked
README in short
The project presents the skill as a way to get an abstract and introduction written from winner statistics. The repository holds the annotation corpus, the `PAPER_LIST.md` list of papers, an installer and the skill with checker scripts and evals.
README badge
Original SKILL.md
--- name: bestpaper-writing description: Draft or rewrite the title, Abstract, Introduction, Method and Experiments of an ML/CV paper in the style of CVPR/ICCV/NeurIPS best papers, distilled from 60 award papers, and plan and draw its figures (Figure 1, method overview, result figures) with award-paper layout and style. Use when the user either (A) fills in the intake form (assets/intake_form.html JSON) or describes problem, idea, method and results, or (B) pastes an existing abstract (or abstract plus introduction) to make it more convincing. Can write directly into a LaTeX template such as the CVPR author kit. Not for related-work surveys, rebuttals or plain proofreading. --- # Best-Paper Writing (v1.2): Abstract, Introduction, Method, Experiments, Figures Turn a description of a research project (or a draft abstract) into an Abstract and Introduction that read like a recent CVPR/ICCV/NeurIPS award paper. This skill is grounded in a sentence-by-sentence annotation of **55 award papers** (2021–2025; 60 papers are analysed in total, and 5 of them (from 2024) are held out for testing, see `PAPER_LIST.md` in the repository root): what functions their abstracts and introductions contain, in which order, and how they read (`references/corpus-profile.md`). Award papers persuade through: 1. **A sharp motivation** — credit to specific prior work, then a concrete gap (what fails or is missing, when, and why), set up as a tension the paper resolves.