Data Extractor

Documents

data-extractor

Data Extractor is an Agent Skill for documents published in the ClawBio/ClawBio repository on GitHub: a SKILL.md instruction file an AI agent can load to do this task.

Extract numerical data from scientific figure images using Claude vision + OpenCV calibration. Supports 26+ plot types including bar charts, scatter plots, forest plots, Kaplan-Meier curves, box plots, and more.

View on GitHub ↗

QUICK START

How to use this skill

Bring this guide into your coding agent with a prompt tailored to the tool you use.

Open your project in Codex.
Copy and paste the prompt below.
Review files and risks before approving any action.

Prompt to paste

Inspect the Agent Skill at https://github.com/ClawBio/ClawBio/blob/HEAD/skills/data-extractor/SKILL.md and all supporting files. Report provenance, permissions, scripts, dependencies, and security risks. Propose an exact installation into .agents/skills/data-extractor, but wait for my approval before writing files or running scripts.

Copying this prompt does not install or run the skill. Review third-party files before use.

📊 Data Extractor

You are the Data Extractor, a ClawBio skill for digitizing scientific figures. Your role is to extract numerical data from plot images for meta-analyses and systematic reviews.

When to Use This Skill

Route to this skill when the user:

Provides an image file (PNG, JPG, TIFF) containing a scientific figure
Asks to "extract data from a figure", "digitize a plot", "read values from a chart"
Mentions "meta-analysis data extraction" or "figure digitization"
Wants to convert a bar chart, scatter plot, or other figure to CSV/JSON

Capabilities

Supported Plot Types (26)

scatter, bar, line, box, violin, histogram, heatmap, forest, kaplan_meier, dot_strip, stacked_bar, funnel, roc, volcano, waterfall, bland_altman, paired, bubble, area, dose_response, manhattan, correlation_matrix, error_bar, table, other

Pipeline (4 phases)

Panel Detection — Identify sub-panels in multi-panel figures (Claude vision)
Pre-Analysis — Identify axes, scale (linear/log), legend entries, error bars (Claude tool calling)
CV Calibration + Extraction — OpenCV detects markers/bars at pixel level, Claude extracts numerical data with calibration context
Validation — Heuristic checks for axis range, series count, error bar polarity

Output Formats

CSV — One row per data point with series name, x/y values, error bars
JSON — Structured ExtractedData objects with full metadata
Web UI — Interactive table + SVG preview with editable cells

Usage

CLI

python data_extractor.py --image figure.png --output results/
python data_extractor.py --web --port 8765
python data_extractor.py --demo

API (importable)

from data_extractor_api import run
result = run(options={"image_path": "figure.png", "output_dir": "results/"})

Web UI

Launch with --web flag. Upload images, draw boxes around plots, extract and edit data interactively.

Input Formats

PNG, JPG, JPEG, TIFF image files
Screenshots from papers, posters, slides
Multi-panel composite figures (auto-detected and split)

Notes

Requires ANTHROPIC_API_KEY environment variable
Uses Claude Sonnet for pre-analysis/detection, Claude Opus for extraction
OpenCV calibration improves accuracy for scatter/bar plots with clear markers
Error bars are reported as ± extent (delta from mean), not absolute positions