yjz211/vivid-figures-skill

让 AI 用你的数据生成科研图表：108 个配方，鲜艳与稳重两种风格，多套配色，交付图片与源码。仅限个人非商业使用。

What it is

Vivid Figures is an Agent Skill – a collection of data‑driven chart templates and colour palettes that an AI assistant can use to automatically create scientific figures. The repository bundles:

143 ready‑made chart “recipes” (e.g., ridge plots, heatmaps, 3‑D grouped bar charts, SHAP importance visualisations, etc.)
7 colour‑scheme packs (each with 5‑8 colours) stored in a single JSON file
Python code that knows how to read the recipe, load a user’s CSV/Excel/JSON data, pick a suitable chart type, render the figure, and output PNG, PDF and the reproducible source script.

The skill follows the Agent Skills open specification, meaning it can be dropped into any AI‑assistant platform that can read files, run Python, and display images.

Who it’s for

Researchers writing papers, competition reports, or modelling notebooks.
Anyone who has raw data (tables, model outputs, spatial coordinates) and wants a polished figure without manually choosing chart types or tweaking colours.
AI assistants that support the Agent Skills format (e.g., Claude‑based bots, OpenAI‑function‑calling agents, or custom “assistant‑as‑a‑service” tools).

How it works (high‑level flow)

Load the skill – the assistant imports the whole vivid-figures-skill folder.

User prompt – you tell the assistant what you want, e.g.:

用 vivid-figures-skill 读取 results.csv，比较不同方法的得分分布。用珊瑚青绿配色。

Data parsing – the Python helper reads the CSV/Excel file, infers column meanings, and matches them to a chart template.
Template selection – based on the data shape and the user’s intent, the skill picks one of the 143 recipe cards (each card contains a markdown description, a preview image, and the full source code).
Rendering – the associated Python script runs, producing a PNG for quick preview, a PDF for publication, and the full source code (Matplotlib/Seaborn/Plotly etc.) for later edits.
Post‑processing – the assistant can adjust legends, font sizes, or move elements as instructed, preserving the original colour palette and design intent.

Key features

Large library of chart recipes – covers basic plots (scatter, violin, histogram) and advanced visualisations (3‑D grouped bar, multi‑Y‑axis gradient histograms, SHAP dependence & contribution maps, relief‑style correlation heatmaps).
Seven curated colour palettes – each palette is a JSON file; the default is Olive‑Apricot but any palette can be selected via a simple natural‑language command.
Open Agent‑Skill format – metadata in SKILL.md (YAML + markdown) and all assets referenced by relative paths, making the skill portable across different AI‑assistant hosts.
Output flexibility – PNG for quick checks, PDF for LaTeX/Word insertion, and the exact Python source so the user can reproduce or tweak the figure later.
Non‑commercial, personal‑use licence – the repository is free for academic or personal projects but forbids commercial redistribution without written permission.
Self‑checking – each template includes “fidelity points” that the assistant can verify (e.g., keep gradients, retain layer order) to avoid over‑simplifying the design.

Getting started (quick checklist)

Prerequisites – Python 3.10+, Git, Bash (or Git Bash on Windows). Optional: Node.js 22+ if you want to browse the HTML catalog locally.
Install – follow docs/INSTALL.md (clone the repo, run pip install -r requirements.txt).
Load into your AI assistant – place the folder where the assistant expects Agent Skills, or use the host‑specific import command.
Run a prompt – as shown in the README, ask the assistant to generate a figure from your data file.
Find results – the assistant writes a figures/ directory containing the PNG, PDF, and the Python script.

When it might not be useful

If you need a fully custom visualisation that isn’t covered by the 143 templates, you’ll still have to write code yourself.
The skill is designed for AI assistants that can execute Python and view images; a plain CLI tool without an LLM won’t be able to drive the selection logic.
Commercial use is prohibited without a separate licence.

Bottom line

Vivid Figures is a ready‑to‑use, AI‑driven chart‑generation skill that lets a language model turn raw data into publication‑quality figures with minimal prompting. It bundles a rich set of pre‑tested templates, colour palettes, and a reproducible Python pipeline, all packaged in the open Agent Skills specification.