shaowen-ye

FigDataX — Data skill for Claude Code

Data community

高精度科学图表数据提取 Claude Code 技能 High-precision scientific figure data extraction skill for Claude Code.

How to install FigDataX

This entry records only its repository, not the path inside it, so there is no exact command to give. Open shaowen-ye/FigDataX and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What FigDataX does

高精度科学图表数据提取 Claude Code 技能 | High-precision scientific figure data extraction skill for Claude Code. ±0.5% accuracy.

Alternatives in Data

Bitbucket Data Center 71.9k ★
Financial Services — Reference agents, skills, and data connectors for the financial-services workflows we see most — investment ba 34.5k ★
Data Contract — This document defines which files belong to the system (auto-updatable) and which belong to the user (never to 32.5k ★

README

FigDataX

**Fig**ure **Data** e**X**traction — high-precision scientific figure data extraction, as an agentic **Claude Code skill**.

Give Claude a figure image (or several) cropped from a paper; it extracts the numbers at up to **±0.5% accuracy** and proves it with a validation overlay. The engine measures the geometry (plot area, tick pixel positions, series colors, sub-pixel centroids); Claude reads the semantics (tick values, legend) with vision; you are asked only when something is genuinely ambiguous.

Supported: scatter, line, bar (grouped/stacked), box, pie, heatmap, polar, error bars, log axes, multi-panel figures. Output: CSV per figure + optional multi-sheet Excel with provenance (method, calibration RMSE, validation rounds).

中文说明见 [中文说明.md](中文说明.md)。

Install

git clone https://github.com/Shaowen-Ye/FigDataX ~/.claude/skills/FigDataX
bash ~/.claude/skills/FigDataX/scripts/setup.sh      # creates .venv, installs deps (uv)
~/.claude/skills/FigDataX/.venv/bin/python -m scripts.figdatax self-test

The folder name is **`FigDataX`** (case-sensitive on Linux).

Use

In Claude Code, just ask:

> 提取 /path/to/figure.png 图片数据
> Extract data from ./results/fig3.png
> 这几张图都提取一下，汇总成一个 Excel

Claude runs the autonomous loop in [SKILL.md](SKILL.md): classify the chart → engine geometry pass (`figdatax geometry`: bbox + tick positions + series colors) → visually verify the annotated overlay → pair tick values by vision → calibrate (hard RMSE < 1% gate) → extract sub-pixel centroids → visually check the validation overlay → iterate up to 3 rounds → deliver CSV/Excel + a provenance report.

Zero-typing entry points (macOS)

bash ~/.claude/skills/FigDataX/integrations/macos/install.sh

installs two shortcuts so you barely type at all:

/figx slash command — in Claude Code, type /figx and drag the image(s) into the terminal. Full interactivity (Claude can still ask at the two gates).
Finder Quick Action — sele