skills-comm

A community collection of neuroimaging skills for LLM coding agents.

skills-comm gathers Agent Skills that teach LLM agents (Claude Code and other agents that read SKILL.md/AGENTS.md) how to drive real neuroimaging tools correctly and reproducibly. Instead of letting an agent improvise commands for FSL, ANTs, FreeSurfer, DataLad, nipoppy, or a SLURM cluster, a skill encodes the operational knowledge — the right flags, the safe order of operations, the quality-control checks, and the reproducibility guardrails — so the work is repeatable across people and machines.

The goal is operationalization: turn “how an expert actually runs this tool” into a package an agent can load on demand.

Who this is for

• Method developers & tool authors who want their software used the way it was meant to be used, even when an LLM is at the keyboard.
• Neuroimaging researchers who want a coding agent that follows community best-practice (BIDS, YODA/DataLad provenance, explicit tool versioning) by default.
• Contributors who want to share a reproducible recipe as a reusable skill.

What’s in here

Each contribution lives under plugins/ as its own directory. A contribution is either a full Claude Code plugin (with a .claude-plugin/plugin.json manifest) or a bare skill collection (one or more SKILL.md directories). Current contributions:

skill-iterations/ is a teaching example: the same brain-extraction skill captured at six successive stages of refinement, showing how a raw draft is hardened into a reliable skill (fixing YAML, removing “Claude-isms”, generalizing across tools, improving QC). Read it to see what “good” looks like before you write your own.

Skill anatomy