skill

htan-claude

ncihtan/htan-claudecommunitylistedunreviewed

Set up

Skills are installed by placing their folder where your agent loads skills.

Download ZIP from GitHub Open the repository

The download comes from github.com/ncihtan/htan-claude itself, not from our server, so what you get is exactly what its author published.

Install instructions come from the README

Skills do not use an MCP config. Follow the steps in the README below.

README

synced Oct 26, 2026

HTAN Skill for Claude Code

A Claude Code plugin for working with the Human Tumor Atlas Network (HTAN) — an NCI Cancer Moonshot initiative constructing 3D atlases of the dynamic cellular, morphological, and molecular features of human cancers as they evolve from precancerous lesions to advanced disease.

frame_0008

What It Does

Capability	Auth Required	Description
Portal queries (ClickHouse)	Synapse team membership	Query file metadata, clinical data, download coordinates
Data model	None	Query HTAN data model components, attributes, controlled vocabularies
PubMed search	None	Search HTAN-affiliated publications by keyword, author, year
File mapping	None	Resolve HTAN file IDs to Synapse/Gen3 download coordinates
Synapse download	Synapse token	Download open-access data (processed matrices, clinical)
Gen3/CRDC download	Gen3 credentials + dbGaP	Download controlled-access data (raw sequencing)
BigQuery metadata	Google Cloud ADC	Query HTAN metadata tables in ISB-CGC

Install (Users)

1. Add the plugin to Claude Code

# From the marketplace
/plugin marketplace add ncihtan/htan-claude
/plugin install htan@htan-claude

2. Set up and go

Invoke the skill with /htan. On first use, Claude will:

Check for the htan CLI and, if missing, install it from PyPI into a .venv in your project
Check your credentials, and run uv run htan init only if something is missing
Suggest adding Bash(uv run htan *) to your project permissions, unless your settings already allow it

The CLI itself lives at ncihtan/htan-cli — file CLI bugs and feature requests there.

Then just ask:

"List all scRNA-seq files from breast cancer in HTAN"
"Search HTAN publications about spatial transcriptomics"
"What attributes are required for scRNA-seq Level 1 manifests?"

Demo Outputs

Example outputs from running the skill headlessly via claude -p:

Prompt	Output
Give me an overview of what data is available in HTAN	01_overview.md
Find the 10 smallest open-access scRNA-seq breast files	02_scrna-breast.md
What clinical demographics are available for HTAN OHSU?	03_demographics-ohsu.md
Participants per cancer type with both scRNA-seq and imaging (BigQuery)	04_bigquery-multimodal.md
HTAN publications about spatial transcriptomics from 2024	05_pubs-spatial.md
Required attributes for scRNA-seq Level 1 manifest	06_model-scrna.md
Download the smallest open-access scRNA-seq breast file	07_download-open.md
Download a controlled-access file via Gen3 (dry run)	08_download-controlled.md

Re-run with: bash demo/run_demo.sh demo/output

Develop (Contributors)

This repo contains the Claude Code plugin (skill definition + reference docs + demo). The Python CLI it relies on lives in a separate repo.

# Plugin (this repo)
git clone https://github.com/ncihtan/htan-claude.git
cd htan-claude
claude --plugin-dir .              # use as a local plugin

# CLI (separate repo)
git clone https://github.com/ncihtan/htan-cli.git
cd htan-cli
uv venv && uv pip install -e ".[dev]"
uv run pytest tests/               # 323 tests

Authentication

Service	How to Set Up
Portal	Join HTAN Claude Skill Users team, then run `uv run htan init`
Synapse	Get a Personal Access Token from synapse.org, set `SYNAPSE_AUTH_TOKEN` or configure `~/.synapseConfig`
Gen3/CRDC	Request dbGaP access for study `phs002371`, download credentials from the CRDC portal
BigQuery	Run `gcloud auth application-default login` and set `GOOGLE_CLOUD_PROJECT`

See skills/htan/references/authentication_guide.md for detailed instructions.

CLI Reference

The htan command is the single interface — used by Claude (via uv run) and by you directly. See the htan-cli README for the full reference. A few examples:

uv run htan query portal files --organ Breast --assay "scRNA-seq" --limit 20
uv run htan query portal sql "SELECT atlas_name, COUNT(*) as n FROM files GROUP BY atlas_name"
uv run htan pubs search --keyword "spatial transcriptomics"
uv run htan model components
uv run htan files lookup HTA9_1_19512
uv run htan query bq tables
uv run htan config check

All commands accept --help for full usage.

Architecture

htan-claude/                         # this repo (plugin)
├── skills/htan/
│   ├── SKILL.md                     # Skill definition (teaches Claude the CLI)
│   └── references/                  # Reference docs (schema, auth, atlases)
├── commands/setup.md                # /htan:setup slash command
├── .claude-plugin/plugin.json       # Plugin metadata
├── demo/                            # Headless example outputs
└── CLAUDE.md                        # Plugin development notes

ncihtan/htan-cli                     # separate repo (the CLI)
└── src/htan/                        # `pip install htan` — depended on by the plugin

No MCP server. The skill teaches Claude the CLI commands. Claude runs them via uv run with blanket Bash(uv run htan *) permission.

License

MIT