flare-skills

flare-skills

FLARE lab's Claude Code plugin marketplace. Lab-wide skill set, packaged as one plugin so members can install everything at once and get auto-updates.

The current release ships the medical-imaging dataset workflow subset — three composable skills covering dataset acquisition, DICOM→NIfTI conversion, and nnUNet v2 formatting — absorbed from AIHubSkillSet. Additional lab skills (modeling, writing, reproducibility tooling) will land in this repo over time without changing the install path.

Install (recommended — auto-update on /plugin update)

/plugin marketplace add Bardli/flare-skills
/plugin install flare-skills@flare-skills

Then /plugin list to confirm and /plugin update flare-skills@flare-skills whenever you want to pull new revisions.

What's in the current release (v0.1.0)

Skill	Purpose	Triggers on
dataset-acquisition	Download from TCGA/GDC, Kaggle, HuggingFace, Google Drive; generate SLURM sbatch scripts	"grab the CESC slides", "pull this Kaggle competition", "download from HF and pin the revision", "sbatch script for this download"
dicom-converter	DICOM series → NIfTI; RTSTRUCT/SEG handling; SOP-UID-anchored routing for multi-acquisition data; 10-check audit script; multi-RTSTRUCT OR-union; debug recipes for label misalignment	"convert these DICOMs to NIfTI", "handle this RTSTRUCT", "debug this label/image mismatch", "audit this DICOM dataset"
nnunet-converter	Format imaging datasets into nnUNet v2 layout (imagesTr/labelsTr/dataset.json/splits_final.json); handles 2D PNG/BMP/TIFF, 3D NIfTI/MHA/NRRD, 3D TIFF, multi-modal, classification labels, ignore label, region-based	"make this nnUNet-ready", "prepare for nnUNet training", "generate dataset.json"

These three skills are independent — Claude loads each on demand based on the task — but they compose into the standard medical-imaging pipeline: acquire → convert DICOM → format for nnUNet.

        ┌──────────────────────┐
        │ dataset-acquisition  │   download from TCGA/GDC, Kaggle, HF, gdrive
        └─────────┬────────────┘
                  ▼
        ┌──────────────────────┐
        │ dicom-converter      │   DICOM → NIfTI (only if input is DICOM)
        └─────────┬────────────┘
                  ▼
        ┌──────────────────────┐
        │ nnunet-converter     │   nnUNet v2 layout + dataset.json + splits
        └──────────────────────┘

Skip any step that doesn't apply: if the source is already NIfTI/MHA, skip dicom-converter; if you don't need nnUNet, skip nnunet-converter.

Relationship to AIHubSkillSet

AIHubSkillSet is the upstream public release of these three medical-imaging skills. flare-skills is the FLARE-lab umbrella that absorbs that set as its founding subset and will grow with additional lab-internal skills.

• Use AIHubSkillSet if you only want the three medical-imaging skills as a standalone bundle.
• Use flare-skills if you're inside FLARE lab and want everything we ship under one install.

The skill content here tracks AIHubSkillSet at v0.1.2; future divergence will be noted in this README.

Manual install (without the marketplace)

If you prefer to vendor the skills directly into a project (no auto-update):

git clone https://github.com/Bardli/flare-skills.git
mkdir -p ~/.claude/skills            # or .claude/skills/ for project-scoped
cp -r flare-skills/skills/* ~/.claude/skills/

You can also copy individual skills if you only want one or two:

cp -r flare-skills/skills/nnunet-converter ~/.claude/skills/

Repository layout

flare-skills/
├── .claude-plugin/
│   ├── marketplace.json          # marketplace declaration
│   └── plugin.json               # this repo IS the plugin (one plugin, N skills)
├── skills/
│   ├── dataset-acquisition/      # progressive-disclosure skill
│   │   ├── SKILL.md
│   │   ├── references/           # tcga_gdc, kaggle, huggingface, google_drive, sbatch_template
│   │   └── scripts/              # gdc_manifest.py, hf_download.py
│   ├── dicom-converter/          # progressive-disclosure skill
│   │   ├── SKILL.md
│   │   ├── references/           # 9 topical .md files (audit, SOP-UID routing, multi-RTSTRUCT, etc.)
│   │   └── scripts/              # audit_dicom_dataset, build_sop_to_acq, parse_rtstruct_union, qc-overlay
│   └── nnunet-converter/         # progressive-disclosure skill
│       ├── SKILL.md
│       ├── references/           # 10 topical .md files
│       └── scripts/              # convert_template, simple-CLI, manifest writer
└── README.md

Design notes