Skip to content
Get packs

Guide · Agent skill

How to use the Paper Figure Generator agent skill

Updated · written with AI from the project's own documentation and checked against our catalogue data

The paper-figure skill is an agent skill for Claude that generates publication-quality figures and tables from your experiment results. This guide is for researchers, students and teams who want plots for a LaTeX paper or thesis without writing every matplotlib script by hand. By the end you will know what the skill does, how to install it, how to ask for figures in plain language, and what to check before you use the output.

Paper Figure Generatorwanshuiyin/Auto-claude-code-research-in-sleep · MIT

Quick start

  1. 1Download and unzip the skill. Download the skill ZIP from this page and unzip it into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder.
  2. 2Or upload it in the Claude apps. If you use the Claude apps, go to Settings → Capabilities → Skills and upload the ZIP.
  3. 3Prepare the figure plan and data. Make sure PAPER_PLAN.md has a Figure Plan table and your results are JSON or CSV files in figures/ or the project root.
  4. 4Ask for the figures in plain language. Say something like 'generate figures for my paper from figures/exp_results.json', or use 画图 or 作图.
  5. 5Review the scripts and the output. Read the one-script-per-figure Python files, then check each saved plot in figures/ against your real data before using it.
  6. 6Add manual figures and continue. Draw architecture diagrams and other manual figures yourself, place them in figures/, then run /paper-write.

What the paper-figure skill makes the agent do

An agent skill is a folder of instructions that Claude follows when a request matches the skill's description. This one is for turning experiment data into figures for a paper. When it runs, the agent can build line plots such as training curves, bar charts for method comparisons, scatter plots, heatmaps, box and violin plots, and multi-panel figures that combine several plots in one grid.

It can also write comparison tables as LaTeX, for example a table comparing prior error bounds or method features. The skill sets up a shared matplotlib style, writes one Python script per figure, and saves the results to a figures folder (FIG_DIR, which defaults to figures/).

  • Data-driven plots: line, bar, scatter, heatmap, box/violin, multi-panel
  • Comparison tables written as LaTeX
  • Output as pdf (default, vector) or png
  • Existing figures you added by hand are preserved

Install the skill from a ZIP or the Claude apps

Download the skill ZIP from this directory page, then unzip it so that the skill folder ends up in one of two places. For a personal setup, use ~/.claude/skills/. For a single project, put it in that project's .claude/skills/ folder. The skill's main file is SKILL.md.

If you use the Claude apps instead of Claude Code, you can upload the ZIP under Settings → Capabilities → Skills. Choose whichever route matches how you work; the skill itself is the same.

  • Personal install: ~/.claude/skills/
  • Project install: .claude/skills/ inside your project
  • Claude apps: Settings → Capabilities → Skills, then upload the ZIP
  • Files included: SKILL.md

Trigger the skill with a natural request

The skill activates when your request matches its description. The source lists these trigger phrases: 画图, 作图, generate figures, paper figures, and requests for plots for a paper. You do not need a special command; a plain sentence with one of these ideas is enough.

The skill also accepts an optional argument hint, figure-plan-or-data-path, so you can point it at a plan file or a data path directly.

  • generate figures for my paper from figures/exp_results.json
  • 画图：根据 figures/ 里的实验数据生成论文图表
  • make the paper figures from PAPER_PLAN.md
  • I need a training curve and a comparison bar chart for the paper

Prepare your plan and data before you ask

The skill works best when a figure plan exists. It reads the Figure Plan table in PAPER_PLAN.md, which the source says comes from /paper-plan. Each row has an ID, a type, a description, a data source and a priority. Rows marked manual are left for you to create.

Experiment data should be JSON or CSV files in figures/ or the project root, or screen logs. If there is no PAPER_PLAN.md, the skill scans for data files and asks you which figures to generate.

  • Figure plan table in PAPER_PLAN.md with columns ID, Type, Description, Data Source, Priority
  • Data as JSON or CSV in figures/ or the project root
  • Any hand-made figures already placed in figures/
| Fig 2 | Line plot | ... | figures/exp.json | HIGH |

Know which figures are manual

The skill covers data plots and comparison tables. It does not produce architecture or pipeline diagrams, generated sample grids, photographs or screenshots. The source says architecture diagrams should be drawn yourself, for example in draw.io, Figma, Keynote or TikZ, and that the skill can produce only a rough TikZ skeleton at best.

The source estimates that for a typical ML paper the skill handles about 60% of figures. The rest, such as the hero figure and qualitative results, should be placed in figures/ before you run /paper-write, and the skill will keep them.

  • Automated: line plots, bar charts, scatter, heatmaps, box/violin, multi-panel, LaTeX tables
  • Manual: architecture and pipeline diagrams, generated image grids, photos and screenshots

Safety notes: scripts and network access

The skill runs scripts on your machine. It executes Python plotting code, and its tool list includes shell access (Bash). Read the generated scripts before running them, and run the skill in a project folder where you are comfortable letting code write files.

The listing states that the skill does not use the network. Its frontmatter also lists optional reviewer tools (codex, antigravity, grok, claude-review) used only for figure quality review. Those need separate setup and are not required for generating plots.

  • Runs scripts: yes, Python plotting code plus shell access
  • Uses network: no, according to the listing
  • Reviewer tools are optional and need separate setup
  • Do not use it if you do not want scripts run on your machine

Tips for better figures

Check every generated plot against the real data. The skill is a generator, not a reviewer, and the listing is explicit that manual review is still needed. Settings such as STYLE (publication, poster or slide), DPI (300 by default), FORMAT (pdf or png), COLOR_PALETTE (tab10, Set2 or colorblind) and FONT_SIZE (10 by default) can be changed to suit your venue or slides.

For serif text, the style script lists Times New Roman first, then Times, then DejaVu Serif as fallbacks. Times New Roman is optional, so the figures still render without it.

  • Use pdf output for LaTeX papers because it stays vector
  • Use the colorblind palette if figures must be deuteranopia-safe
  • Keep experiment data in JSON or CSV so scripts can read it directly
  • Re-run a single figure script after you change its data rather than the whole set
matplotlib.rcParams.update({'font.size': FONT_SIZE, 'font.family': 'serif', 'font.serif': ['Times New Roman', 'Times', 'DejaVu Serif']})

What people use it for

  • Turning JSON training logs into a line plot of loss against steps for a methods section
  • Building a grouped bar chart that compares your method with baselines across datasets
  • Creating a 3×3 multi-panel figure of dataset by method results for a results section
  • Writing a LaTeX comparison table of prior bounds and method features for a theory paper

Frequently asked questions

How do I install the paper-figure agent skill?

Download the skill ZIP from this page and unzip it into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder. If you use the Claude apps, upload the ZIP under Settings → Capabilities → Skills instead.

What do I need to run the paper-figure skill?

The source lists Python with matplotlib, LaTeX for comparison tables, and optionally Times New Roman for serif text. The reviewer tools in the frontmatter are only for figure quality review and need separate setup.

Is the paper-figure skill free?

The listing states the licence is MIT. Check the repository's licence file for the exact terms before you redistribute it.

Does the paper-figure skill use the internet?

The listing says it does not use the network. It does run Python scripts and has shell access, so review the generated code before running it.

Can paper-figure draw architecture diagrams?

No. The skill says architecture and pipeline diagrams must be drawn manually, for example in draw.io, Figma, Keynote or TikZ. At best it can produce a rough TikZ skeleton.

Why do my figures use a fallback font instead of Times New Roman?

Times New Roman is optional. The style script lists Times and then DejaVu Serif as fallbacks, so the figures still render if Times New Roman is not installed.

New agent skills, repos and AI tools — every week

The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.

Our verdict

Use this skill if you are a researcher, student or team who wants consistent, print-ready data plots and LaTeX comparison tables generated from JSON or CSV results. Pick something else if you need architecture diagrams, photos or qualitative image grids, or if you are not comfortable running plotting scripts on your machine.

Download the skill