Clair Variant Caller Skill

An agent skill for the Clair suite of variant callers. Provides intelligent model selection, command generation, and troubleshooting for germline, somatic, mosaic, and RNA-seq variant calling from long-read and short-read sequencing data.

Supported Tools

Tool	Repository	Purpose
Clair3	HKU-BAL/Clair3	Germline variant calling (DNA)
Clair3-RNA	HKU-BAL/Clair3-RNA	Variant calling from long-read RNA-seq
ClairS	HKU-BAL/ClairS	Somatic variant calling (paired tumor/normal)
ClairS-TO	HKU-BAL/ClairS-TO	Somatic variant calling (tumor-only)
Clair-Mosaic	HKU-BAL/Clair-Mosaic	Mosaic variant calling

Installation

This skill follows the Agent Skills open standard and works with Claude Code, Codex CLI, and Cursor.

Claude Code

# Global (available in all projects)
git clone https://github.com/HKU-BAL/Clair-skills.git ~/.claude/skills/clair-variant-caller

# Project-level
git clone https://github.com/HKU-BAL/Clair-skills.git .claude/skills/clair-variant-caller

Codex CLI

# Global
git clone https://github.com/HKU-BAL/Clair-skills.git ~/.codex/skills/clair-variant-caller

# Project-level
git clone https://github.com/HKU-BAL/Clair-skills.git .codex/skills/clair-variant-caller

Cursor

# Global
git clone https://github.com/HKU-BAL/Clair-skills.git ~/.cursor/skills/clair-variant-caller

# Project-level
git clone https://github.com/HKU-BAL/Clair-skills.git .cursor/skills/clair-variant-caller

Cursor also auto-discovers skills from ~/.claude/skills/ and ~/.codex/skills/.

Usage

The skill activates automatically when you mention variant calling, Clair tools, or provide BAM/FASTQ files. You can also invoke it explicitly:

• Claude Code: /clair-variant-caller
• Codex CLI: $clair-variant-caller
• Cursor: /clair-variant-caller

Example prompts

I have a BAM file at /data/sample.bam, help me call germline variants.

Call somatic variants from ONT R10.4.1 5kHz tumor-only data at /data/tumor.bam.

My data was basecalled with Dorado v5.2.0 SUP, which model should I use?

Benchmark my Clair3-RNA results against GIAB truth set.

Skill Structure

clair-variant-caller/
  SKILL.md                       # Main skill (workflow, model logic, decision rules)
  templates/
    clair3-command.sh            # Germline calling
    clair3-rna-command.sh        # RNA-seq calling
    clairs-command.sh            # Paired tumor/normal
    clairs-to-command.sh         # Tumor-only
    clair-mosaic-command.sh      # Mosaic calling
  references/
    tool-selection.md            # Decision tree, per-tool summaries
    model-selection.md           # Full PyTorch model catalog (Clair3 v2.0+)
    preprocessing.md             # FASTQ → BAM alignment recipes
    analysis.md                  # Clair-specific outputs, hap.py benchmarking
    setup.md                     # Docker / Singularity / Bioconda / step-by-step install

License

This skill is provided as-is for use with the Clair suite of tools. The Clair tools themselves are licensed under BSD 3-Clause by HKU-BAL.