CellPainting-Claw

CellPainting-Claw was built to make Cell Painting work less fragmented and more usable through agents. Instead of asking users to stitch together data-access utilities, CellProfiler steps, pycytominer processing, DeepProfiler preparation, and agent glue by hand, the project exposes one documented skill catalog for both direct use and agent-mediated use.

The same documented skills can be used in two ways:

• run the skills directly

• run the same skills through an agent

Project Layout

CellPainting-Claw is organized as a simple three-part structure:

People can use the same skills in two ways:

• directly, through cellpainting_skills

• through an agent, through OpenClaw

Both paths rely on the same foundation packages underneath. The lower-level cellpainting_claw package remains available for advanced direct toolkit use, but it is not the main starting point for most users.

Core Packages

The foundation packages are easiest to understand in workflow order.

Capability	Packages	Main capability
Data access	boto3, quilt3, cpgdata	dataset discovery and download
Measurement extraction	CellProfiler	profiling tables, segmentation masks, labels, outlines, and crop-ready object data
Classical profile generation	pycytominer	aggregation, annotation, normalization, and feature selection
Deep feature extraction	DeepProfiler	learned single-cell feature extraction

Main Entry Paths

For most users, CellPainting-Claw should be understood through two main usage paths.

Purpose	Start with	Outcome
run documented tasks yourself from Python or from the command line	cellpainting_skills	you call the documented skills directly, such as segmentation, pycytominer processing, or DeepProfiler tasks
tell an agent in plain language what you want done	OpenClaw	the agent maps your request onto the same documented skills and runs them through the same toolkit

Public Skill Catalog

Skills are the core public task interface of the project.

Data Access

Skill	Main result
data-inspect-availability	inspect configured sources and write an availability summary
data-plan-download	resolve a download request into a saved download plan
data-download	download one dataset slice into a local cache

Profiling

Skill	Main result
cp-extract-measurements	write CellProfiler measurement tables
cp-build-single-cell-table	merge CellProfiler tables into one single-cell measurements table
cyto-aggregate-profiles	write the aggregated classical profile table
cyto-annotate-profiles	attach metadata to aggregated profiles
cyto-normalize-profiles	write the normalized classical profile table
cyto-select-profile-features	write the feature-selected classical profile table
cyto-summarize-classical-profiles	turn classical profile outputs into readable summaries and PCA views

Segmentation

Skill	Main result
cp-prepare-segmentation-inputs	prepare the load-data table used by segmentation
cp-extract-segmentation-artifacts	write masks, labels, outlines, and segmentation tables
cp-generate-segmentation-previews	write preview PNGs for quick review
crop-export-single-cell-crops	export masked or unmasked single-cell crop stacks

Deep Features

Skill	Main result
dp-export-deep-feature-inputs	build DeepProfiler-ready metadata and location files
dp-build-deep-feature-project	prepare a runnable DeepProfiler project directory
dp-run-deep-feature-model	run the DeepProfiler model and write raw feature files
dp-collect-deep-features	collect raw feature files into tabular outputs
dp-summarize-deep-features	turn DeepProfiler tables into readable summaries and PCA views

Start Here

Start with these pages:

• Introduction

• Quick Start

• Skills

• CLI

• OpenClaw