Model or dataset
ChenLiu-1996/figures4papers avatar
ChenLiu-1996/figures4papers

figures4papers: A Personal Figure Toolkit Repackaged as an AI Agent Skill

My Python scripts to make high-quality figures for publications in top AI conferences and journals.

11,737 stars817 forksPythonNOASSERTION

At a glance

What is it?
Chen Liu's figures4papers is a collection of Python plotting scripts from published AI papers, now structured as a skill for AI coding agents. Its value lies in the concrete examples and style conventions, not in a formal library or API.
Who is it for?
Adopt figures4papers if you produce figures for machine learning papers and want a consistent, publication-ready style without building a plotting pipeline from scratch. The repository is most useful as a reference and as a promptable skill for AI coding agents like Cursor or Claude Code, where it provides concrete patterns and design rationale.
Can I use it commercially?
Check first. The repository uses a licence we do not classify automatically, so read its LICENSE file before any commercial use.
Is it still maintained?
Yes. The repository last received commits 1 day ago.
What is it written in?
Mainly Python, according to GitHub's language statistics.

Answers come from the project's GitHub data, last synced on November 15, 2026, and from our analysis. They are not legal advice.

From daily snapshots

Star history

  • ChenLiu-1996/figures4papers11,737 ★ · +6,823
Star history
Repository2026-09-102026-11-15
ChenLiu-1996/figures4papers4.9k11.7k

6,823 Stars in the last 66 days, measured from the daily GitHub snapshot. Compare with other projects

Editorial analysis

What the repository actually contains

figures4papers is not a package. It is a folder of Python scripts that Chen Liu, a PhD candidate at Yale, used to generate figures for papers published at venues like Nature Machine Intelligence, ICML, NeurIPS, and ECCV. The README shows example outputs: bar plots for quantitative comparison, composition breakdowns, 3D spheres, radar plots, line plots, concept plots, and trend plots. The scripts live in project-specific folders such as figure_ImmunoStruct, figure_Brainteaser, figure_VIGIL, or figure_Dispersion. The repository also includes a separate folder, scientific-figure-making, which packages the author's approach into a skill for AI coding agents. The intended audience is researchers who need publication-quality figures and who either want to imitate the style or want an AI assistant to generate scripts that follow that style. There is no installable module, no command-line entry point, and no formal API beyond the conventions described in the skill's reference files.

The skill layer is the real contribution

The most distinctive part is scientific-figure-making, a directory that acts as a skill definition for AI coding agents. It contains a SKILL.md file and a references folder with api.md, common-patterns.md, demos.md, design-theory.md, and tutorials.md. The skill is meant to be consumed by an agent like Cursor or Claude Code. The README offers two ways to use it. The first is path-based: open the repository in the agent and reference the skill files directly in a prompt. The second is installation via symlink: for Cursor you run mkdir -p ~/.cursor/skills then ln -s "$(pwd)/scientific-figure-making" ~/.cursor/skills/scientific-figure-making, and similar commands for Claude Code and Codex. After linking, the agent can invoke the skill by name. This design turns a personal collection of scripts into a repeatable style guide that an AI can follow. The trade-off is that the skill's effectiveness depends entirely on the clarity of the reference files, which are not reproduced in the README, so a user cannot judge their quality without opening the repository.

How the scripts are meant to be used

The README's workflow is straightforward: pick a target folder like figure_PROJECT_NAME, ask the AI to create or update a plotting script there, and tell it to follow the skill conventions. A sample prompt template is provided. It asks the agent to implement or adapt patterns such as apply_publication_style, make_* helpers, and finalize_figure, and to produce both PNG and PDF outputs. The figure_* folders serve as reference implementations. This suggests a workflow where the AI reads existing scripts, extracts the style logic, and applies it to new data. The user then runs the generated script and checks the exported figure. The top-level file list has no dependency file or setup.py. Anyone adopting this repository must infer the required Python packages from the scripts themselves, which is a barrier for non-experts. The design assumes a user who is comfortable with Python and with iterating on AI-generated code.

What the scripts show about figure style

From the sample images and the project names, the figures span common academic visualization types: grouped bar charts for method comparison, stacked or segmented bars for composition, radar charts for multi-metric evaluation, line charts for trends over time, and conceptual illustrations rendered in 3D. The README labels these as 'high-quality figures' and points to publications in top venues as proof of acceptance. The design-theory.md file is supposed to explain the style rationale. What is visible is that the author values consistency: the skill's api.md defines a palette and helper functions so that every figure follows the same visual language. That consistency is valuable for a single author's papers, but it also means the style is opinionated. A researcher who prefers a different look, or who must match a specific journal template, may find the conventions too rigid or too personal.

A genuine limitation: it is a personal archive, not a framework

The repository has no releases, and GitHub reports the licence as NOASSERTION even though a LICENSE file sits at the root. The last push was October 2026, so it is actively maintained by the author, but that maintenance is for his own needs. There is no guarantee that the scripts work with current versions of matplotlib, seaborn, or other dependencies. This is the wrong tool for a team that needs a shared, documented plotting library with versioned releases and a clear license. It is also the wrong tool for someone who wants a plug-and-play package. The value is in the examples and the skill definition, not in a stable codebase. The 'miscellaneous' section includes figures not made end-to-end in Python, which the author includes to acknowledge partial work. That honesty is refreshing, but it also means not every image in the repository is reproducible from the scripts.

Alternatives and how they differ

The obvious alternative is to use a dedicated scientific plotting library like matplotlib with a style sheet, or seaborn for statistical plots, or plotly for interactive figures. Those libraries provide comprehensive APIs, documentation, and community support. The difference is that figures4papers is not a library; it is a collection of scripts plus a promptable skill. A library gives you functions to call; this repository gives you examples to imitate and a set of conventions that an AI agent can follow. Another alternative is to use a tool like draw.io or Illustrator for manual figure design, which offers full control but no reproducibility. The Python-based approach in figures4papers has the advantage that figures can be regenerated when data changes, which matters for papers that require reproducibility. But the skill-based workflow introduces a dependency on AI agents that a direct plotting script does not.

Maintenance and upgrade considerations

Because the repository is a personal collection, there is no formal upgrade path. The author may add new figure_* folders as he publishes new papers, which is useful as a source of fresh examples. The skill files may evolve, but there is no changelog or release history to track. GitHub cannot identify the licence (NOASSERTION), so read the LICENSE file at the root before reusing the code or the skill definitions in a commercial or academic project. The README invites citation of the related papers, which suggests the author is open to academic use, but that is not a legal grant. Before using any script or the skill content, you should read the LICENSE file. The maintenance cost is low if you only use the repository as a reference, but it becomes higher if you copy scripts into your own project, because you inherit dependencies and style choices that you may need to update yourself.

Editorial conclusion

Adopt figures4papers if you produce figures for machine learning papers and want a consistent, publication-ready style without building a plotting pipeline from scratch. The repository is most useful as a reference and as a promptable skill for AI coding agents like Cursor or Claude Code, where it provides concrete patterns and design rationale. Do not adopt it if you need a maintained library with a stable API, package installation, or formal versioning; this is a personal collection with no releases and a licence GitHub cannot identify. Before relying on it, verify that the style conventions and helper functions match your venue's requirements, and check each figure_* script for dependencies you are willing to manage.

Frequently asked questions

What is figures4papers?

A repository of Python plotting scripts for publication-quality figures, written by Chen Liu, a Computer Science PhD candidate at Yale University. The scripts behind figures published at venues including Nature Machine Intelligence, ICML, NeurIPS and ECCV are collected in per-paper directories such as figure_ImmunoStruct/ and figure_Dispersion/, grouped by figure type from bar plots and 3D spheres to radar, concept and trend plots.

How do I use the scientific-figure-making skill in figures4papers?

Two ways, and the first installs nothing: open the repository in an AI coding agent such as Cursor or Claude Code and reference scientific-figure-making/SKILL.md and its references/ files by path. Alternatively symlink the folder into the agent's skills directory, for example `mkdir -p ~/.cursor/skills` then `ln -s "$(pwd)/scientific-figure-making" ~/.cursor/skills/scientific-figure-making`, and restart the agent so the skill can be invoked by name.

Can I reuse the figures and scripts from figures4papers?

The repository carries a LICENSE file at the root, and the README asks readers to cite the related papers if the figures are relevant or helpful, adding a note that this applies within the bounds of academic norms. The repository metadata does not state a licence identifier, so the LICENSE file is the only place the terms are written down, and there is a category of figures the author notes were not made end to end in Python.

Official sources

  1. ChenLiu-1996/figures4papers on GitHub
  2. Issues
  3. Project website
  4. README
Add this badge to your README

If you maintain this project, the badge below links readers to this analysis and shows its maintenance status from the daily GitHub snapshot. Paste the markdown into your README; add ?metric=license or ?metric=stars to the image URL for a different field.

Add this badge to your README

markdown
[![Hysen Labs](https://hysenlabs.com/badge/chenliu-1996-figures4papers.svg)](https://hysenlabs.com/projects/chenliu-1996-figures4papers)

Star history chart for your README

markdown
[![Star history](https://hysenlabs.com/chart/chenliu-1996-figures4papers.svg)](https://hysenlabs.com/projects/chenliu-1996-figures4papers)

Was this useful?