Skip to content
Get packs

Guide · Agent skill

How to use the CELLxGENE Census Querier agent skill

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

The CELLxGENE Census Querier is an agent skill: a folder of instructions and reference files that a coding agent such as Claude loads when your request matches its purpose. This one lets the agent query the CZ CELLxGENE Census, a versioned public collection of standardized single-cell and spatial transcriptomics data. After this guide you will know how to install the skill, how to phrase requests so it triggers, and what it can and cannot do.

CELLxGENE Census QuerierK-Dense-AI/scientific-agent-skills · MIT

Quick start

  1. 1Download the ZIP. Download the skill ZIP from this page.
  2. 2Unzip into a skills folder. Unzip into ~/.claude/skills/ for personal use or .claude/skills/ for one project.
  3. 3Prepare Python. Use Python 3.10 or newer on Linux or macOS. The skill was tested on Python 3.12.
  4. 4Install the packages. Run uv pip install "cellxgene-census==1.18.0" and add the spatial or tiledbsoma-ml packages if you need them.
  5. 5Ask a natural question. Describe the public single-cell data you want, such as cell counts by tissue or an AnnData slice for a disease.
  6. 6Check the output. Confirm the pinned Census version and that is_primary_data filtering was applied.

What the skill makes the agent do

Once installed, the agent can search and pull slices of the CZ CELLxGENE Census. It can count cells by cell type, tissue, disease or assay, fetch expression data into AnnData format for analysis with Scanpy, and prepare data for machine learning. It pins a specific Census release so results stay reproducible.

  • Explore metadata such as cell types, tissues, diseases, donors and datasets
  • Count cells and summarise cross-dataset statistics
  • Fetch gene expression slices into AnnData
  • Access pre-calculated embeddings and spatial data, and look up source H5AD files
  • Prepare data for PyTorch training through TileDB-SOMA-ML

Install the skill from the ZIP

Download the skill ZIP from this page. Then choose one of two routes.

For a personal install, unzip it into ~/.claude/skills/. For a single project, unzip it into the project's .claude/skills/ folder. In the Claude apps, you can instead upload the ZIP under Settings → Capabilities → Skills.

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

Install the Python packages the skill needs

The skill's instructions use these packages. Run the installs in a dedicated environment on Linux or macOS. Pin the versions the skill was tested with.

  • Core API: cellxgene-census 1.18.0, installed with uv pip
  • Spatial workflows: the cellxgene-census[spatial] extra
  • PyTorch training: tiledbsoma-ml together with PyTorch
uv pip install "cellxgene-census==1.18.0" uv pip install "cellxgene-census[spatial]==1.18.0" uv pip install "cellxgene-census==1.18.0" tiledbsoma-ml

Trigger the skill with natural requests

You do not need to name the skill. Describe what you want from public single-cell data. The skill is meant for requests about population-scale metadata, expression slices, summary counts, source H5AD downloads, embeddings, spatial data and reference atlas comparisons.

If you want to analyse your own local single-cell files, ask for scanpy, anndata or scvi-tools instead. The skill itself points there.

  • How many B cells are in primary human lung datasets in the Census?
  • Pull T cells from COVID-19 patients as an AnnData object for Scanpy.
  • Which datasets measured CD4 and CD8A in human data?
  • Compare my cluster markers against a reference atlas of human immune cells.

Keep queries reproducible and small

The skill's examples pin the Census build 2025-11-08 and tell the agent to record it. Always filter for primary data so cells are not counted twice. Count selected cells before loading a big matrix, because cell count alone does not estimate memory use.

Exact equality filters select only cells whose whole field matches that label, so complete disease cohorts need the multi-value workflow in the reference files.

  • Pin census_version, for example 2025-11-08
  • Include is_primary_data == True unless you need duplicates
  • Select only the metadata columns you need
  • Check gene presence with feature_id, not gene symbol alone
obs_value_filter="cell_type == 'B cell' and is_primary_data == True"

Safety notes: scripts and network access

This skill runs scripts and uses the network. It needs access to public HTTPS manifests and S3 storage. The skill's metadata declares allowed tools as Read, Write, Edit and Bash, so review generated code before running it on a machine you care about. No Census credentials are required, according to the skill's compatibility note.

Large queries can exhaust memory. The skill warns that a high cell count alone does not tell you how much RAM you need, so start with a count and a narrow filter.

  • Runs scripts: yes
  • Uses network: yes
  • Allowed tools: Read, Write, Edit, Bash

What people use it for

  • Count cells of a given type across tissues or diseases before designing a study.
  • Pull an AnnData slice of one cell type from a disease cohort for Scanpy analysis.
  • Check which public datasets measured a set of marker genes.
  • Compare your findings against a reference atlas of human or other organism single-cell data.
  • Prepare training data for a PyTorch model using TileDB-SOMA-ML.

Frequently asked questions

Is the CELLxGENE Census Querier skill free?

The skill is published under the MIT licence, as listed on its page. The Census itself is public data, and the skill requires no Census credentials.

What do I need to run the cellxgene-census skill?

You need Python 3.10 or newer, Linux or macOS, and network access to public HTTPS manifests and S3. The skill was tested with Python 3.12, cellxgene-census 1.18.0 and TileDB-SOMA 2.3.0. Spatial and PyTorch workflows need optional extras.

Where do I install an agent skill like this?

Unzip the skill into ~/.claude/skills/ for personal use or into a project's .claude/skills/ folder. In the Claude apps you can upload the ZIP under Settings → Capabilities → Skills.

Can the skill analyse my own single-cell data?

No, that is not its purpose. The skill is for public Census data. For your own local files the skill points you to scanpy, anndata or scvi-tools.

Why does my query return fewer cells than expected?

Exact equality filters match only cells whose whole field equals that label. For disease cohorts, use the multi-value workflow in the reference files, and always check is_primary_data handling.

How does this compare with downloading datasets myself?

The skill queries the Census without downloading whole datasets first, and pins a release for reproducibility. Downloading individual datasets yourself gives you full files but no cross-dataset query layer.

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 bioinformatician, researcher or student who needs to query public CELLxGENE single-cell data programmatically and wants a pinned, reproducible release. Pick something else if your data is local, if you need clinical or diagnostic conclusions, or if you cannot run scripts and network calls on your machine.

Download the skill