Skip to content
Back to search

humanize-ml-paper

MarketplaceWorks with Claude CodeGitHub

The humanize-ml-paper skill for ML conference papers

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/lehengthu--humanize-ml-paper/badge.svg)](/listings/lehengthu--humanize-ml-paper)

HTML

<a href="/listings/lehengthu--humanize-ml-paper"><img src="/api/listings/lehengthu--humanize-ml-paper/badge.svg" alt="findsafeskills safety rating" /></a>

Install

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

**Claude Code** (plugin marketplace):

/plugin marketplace add LehengTHU/humanize-ml-paper
/plugin install humanize-ml-paper@humanize-ml-paper

**Other agents** (Codex, Cursor, Gemini CLI and others that read `SKILL.md`):

npx skills add LehengTHU/humanize-ml-paper
# or, with the GitHub CLI
gh skill install LehengTHU/humanize-ml-paper humanize-ml-paper

**Manual:**

git clone https://github.com/LehengTHU/humanize-ml-paper
cp -r humanize-ml-paper/skills/humanize-ml-paper ~/.claude/skills/

The skill needs Python 3.8+ for its checker and a LaTeX engine ([tectonic](https://tectonic-typesetting.github.io) is the smallest) to build papers. After a manual install, `bash ~/.claude/skills/humanize-ml-paper/scripts/check_env.sh` reports anything missing and builds a one-page test PDF. The full setup is in [`skills/humanize-ml-paper/README.md`](skills/humanize-ml-paper/README.md).

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

.claude-plugin/marketplace.json · Manifest · 15 lines
  1. {
  2. "name": "humanize-ml-paper",
  3. "description": "The humanize-ml-paper skill for ML conference papers",
  4. "owner": {
  5. "name": "Leheng Sheng",
  6. "url": "https://github.com/LehengTHU"
  7. },
  8. "plugins": [
  9. {
  10. "name": "humanize-ml-paper",
  11. "source": "./",
  12. "description": "Write or revise ML conference papers (ICLR, NeurIPS, ICML) in the style of human-written papers that follow academic conventions."
  13. }
  14. ]
  15. }
README.md · README · 113 lines
  1. # humanize-ml-paper
  2. **An agent skill for writing ML conference papers (ICLR, NeurIPS, ICML) in the style of human-written papers.**
  3. English | [中文](README.zh-CN.md)
  4. [![License: MIT](https://img.shields.io/badge/License-MIT-blue.svg)](LICENSE)
  5. [![Agent Skills](https://img.shields.io/badge/Agent%20Skills-SKILL.md-2A4E8F)](https://agentskills.io)
  6. [![Blog post](https://img.shields.io/badge/Blog-The%20Art%20of%20Writing-2A4E8F)](https://lehengthu.github.io/blog/2026/paper-skill-ai-detection/)
  7. <a href="https://lehengthu.github.io/blog/2026/paper-skill-ai-detection/"><img src="media/demo.gif" width="800" alt="Opening film of the blog post: an agent writes a draft page and revises it with the skill's style sheet"></a>
  8. We distilled the writing practices of human-written papers that follow academic conventions into a workflow, rules, a phrase bank and checking tools. An agent equipped with the skill can write fluent, logical prose with a lower AI detection rate and can even write a complete conference paper in one run.
  9. The full study is the blog post [The Art of Writing: Agents with Well-Curated Skills Can Cheat AI Detectors](https://lehengthu.github.io/blog/2026/paper-skill-ai-detection/). You can also watch the [37-second film](media/demo.mp4).
  10. ## Results
  11. <img src="media/result-en.png" width="760" alt="Share of text that Pangram 4.1 labels AI on four papers: 94% without the skill, 29% with the skill, 29% for the human-written text">
  12. - **Text generated with the skill is less likely to be detected by Pangram.** We simulate the writing process from a draft to a finished text. On four papers, Pangram labels 94% of the text that Claude Opus 5.5 writes without the skill as AI, compared with 29% with the skill, the same as the human-written text.
  13. - **The text follows academic conventions.** The skill's checker finds 6 habits of machine-written drafts in the text generated with the skill, compared with 75 without it.
  14. - **One prompt, one complete paper.** Given a single prompt, an agent with the skill chose an idea, ran the experiments and typeset an ICML 2027 submission alone in about 82 minutes.
  15. - **The skill does not replace your thinking.** If the content is AI-generated to begin with, revising it with AI cannot correct it: papers written from a topic alone stay at 100% AI. A complete one-shot paper reads as human but is still labeled AI because of its length.
  16. > [!CAUTION]
  17. > **Scope:** the texts that pass the detector are mostly short ones written from a human draft. Text generated directly by AI, or a long paper, is still likely to be labeled AI.
  18. ## Advanced usage: one prompt, one complete paper
  19. The skill can also take an agent from an idea to a typeset submission. Given one prompt, Claude Opus 5.5 with the skill wrote [SourceStein](examples/sourcestein-icml2027/), a 14-page ICML 2027 submission, in a single run of 209 turns and about 82 minutes, including 37 minutes of its own experiments on a CPU-only Mac. Nobody edited the paper.
  20. <a href="examples/sourcestein-icml2027/"><img src="examples/sourcestein-icml2027/preview/pages.jpg" alt="All 14 pages of SourceStein, the paper written in one run"></a>
  21. [Read the paper](examples/sourcestein-icml2027/SourceStein.pdf) · [The prompt](examples/sourcestein-icml2027/prompt.md) · [Source, code and results](examples/sourcestein-icml2027/source/)
  22. > [!CAUTION]
  23. > **Risk:** for a complete generation, although the writing reads as human, it is still labeled AI because of its length.
  24. ## Install
  25. **Claude Code** (plugin marketplace):
  26. ```
  27. /plugin marketplace add LehengTHU/humanize-ml-paper
  28. /plugin install humanize-ml-paper@humanize-ml-paper
  29. ```
  30. **Other agents** (Codex, Cursor, Gemini CLI and others that read `SKILL.md`):
  31. ```bash
  32. npx skills add LehengTHU/humanize-ml-paper
  33. # or, with the GitHub CLI
  34. gh skill install LehengTHU/humanize-ml-paper humanize-ml-paper
  35. ```
  36. **Manual:**
  37. ```bash
  38. git clone https://github.com/LehengTHU/humanize-ml-paper
  39. cp -r humanize-ml-paper/skills/humanize-ml-paper ~/.claude/skills/
  40. ```
  41. The skill needs Python 3.8+ for its checker and a LaTeX engine ([tectonic](https://tectonic-typesetting.github.io) is the smallest) to build papers. After a manual install, `bash ~/.claude/skills/humanize-ml-paper/scripts/check_env.sh` reports anything missing and builds a one-page test PDF. The full setup is in [`skills/humanize-ml-paper/README.md`](skills/humanize-ml-paper/README.md).
  42. ## Usage
  43. Ask in plain language; the skill triggers on paper-writing tasks. For example:
  44. - "Turn my method notes and results in `notes.md` into a full ICLR paper."
  45. - "Rewrite the introduction in `chapters/1_introduction.tex` in this style."
  46. - "Make the main result table from `results.csv`, with empty cells for the missing runs."
  47. - "Plan the page budget for an ICML submission and tell me what goes to the appendix."
  48. - "Review my draft against the style checklist."
  49. The checker also runs on its own: `python3 skills/humanize-ml-paper/scripts/check_paper.py <paper_dir> --stage submission`.
  50. ## What is inside
  51. | Path | Contents |
  52. |---|---|
  53. | `SKILL.md` | The workflow from a proposal to a finished paper, in eight steps, and the style checklist |
  54. | `references/` | Venue limits and page budgets, paper types, structure, prose rules, the argument chain of the Introduction, concepts, math, experiments, figures and tables, LaTeX conventions, annotated examples |
  55. | `assets/lexicon/` | The phrase bank: 1,272 expressions taken from the papers, each with its counts and a tier (core, occasional, retired, reference only, banned) |
  56. | `tools/` | `ps.py` looks up how often a phrase occurs, from a shipped index of word-sequence counts that holds no sentences; `build_corpus.py` and `ngrams.py` build the same from your own papers |
  57. | `scripts/check_paper.py` | Word budgets, the page limit, required statements, sentence shape and the 27 habits of machine-written drafts, with line numbers |
  58. | `templates/` | ICLR and NeurIPS (one column) and ICML (two column) LaTeX projects, table templates, a plot style and a TikZ diagram style |
  59. ## Responsible use
  60. - A draft that passes a detector is still an LLM draft. Venues such as ICLR require you to disclose any use of an LLM, and this applies to papers written with this skill.
  61. - Detector scores should not decide authorship.
  62. - What the paper says and how it argues are still for you to decide. Read and revise everything the agent writes.
  63. ## Citation
  64. ```bibtex
  65. @misc{sheng2026paperskill,
  66. title = {The Art of Writing: Agents with Well-Curated Skills Can Cheat AI Detectors},
  67. author = {Sheng, Leheng and Zhang, An and Wang, Xiang},
  68. year = {2026},
  69. howpublished = {Blog post},
  70. url = {https://lehengthu.github.io/blog/2026/paper-skill-ai-detection/}
  71. }
  72. ```
  73. ## Acknowledgement
  74. I thank my advisors Xiang Wang and An Zhang for permission to distill the papers, the early users for their feedback, and my girlfriend Xieyang Zhao for her support and help with the design of the blog.
  75. ## License
  76. [MIT](LICENSE)

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.