dwi-skill

What it does

The skill acts as a modality-layer interface for diffusion MRI work, outlining WHAT needs to be done and WHICH tool skill to delegate to, without containing implementation code itself. It plans tasks like preprocessing, metric extraction, and tractography/connectome workflows and specifies that outputs should be saved under dwi_output/. It emphasizes preferring QSIPrep + MRtrix3-style downstream processing for tractography/connectome when data and inputs support it, and to report Missing required input if key options are missing. It delegates concrete execution to tool skills via claw-shell and relies on QSIPrep, MRtrix3, DIPY, and FSL pathways as appropriate. It also includes a structured benchmark guidance for full pipelines, ROI stats, and connectome tasks, and lists required inputs like atlas/parcellation, labels/LUT, space alignment, and shell/mode choices. It provides a detailed MRtrix3-based tractography workflow as a reference, including 5TT segmentation, label relabeling, DWI conversion, multi-tissue response estimation, FOD computation, ACT-based tractography, SIFT2 weighting, and tck2connectome. It concludes with installation notes and a concrete batch example for HCP-style tractography, and a Quick Reference of common tasks and delegation paths."

How it works

• Identify input type (DICOM / NIfTI / BIDS), single-shell vs multi-shell, and availability of reverse phase-encoded maps.
• Generate a numbered execution plan detailing steps, tools, outputs, runtime, and risks.
• Present the plan and await explicit user confirmation (e.g., “YES” / “execute” / “proceed”).
• On confirmation, delegate each step to the relevant tool skill via claw-shell.
• Save outputs into dwi_output/.
• When tractography or connectome is involved, prefer a diffusion-first route: QSIPrep -> MRtrix3 downstream processing with multi-tissue FOD, ACT, SIFT2, and tck2connectome, rather than tensor-only defaults, provided inputs allow it.
• If essential inputs (atlas/parcellation, LUT, registration targets) are missing, state Missing required input explicitly and prompt for key options before execution.
• For existing metrics or tractography cases, follow the specified downstream mainline (e.g., ROI stats or connectome construction) without restarting from raw preprocessing unless requested.
• The skill delegates execution to tools like qsiprep-tool, dipy-tool, fsl-tool, and claw-shell, and saves outputs in dwi_output/.

When to use it

• When a user wants to preprocess diffusion MRI data, compute diffusion metrics (FA/MD/AD/RD), extract ROI diffusion features, or run tractography/connectome workflows.
• When a full DWI pipeline ending in tractography/connectome is requested, with a preference for QSIPrep + MRtrix3-style downstream processing given inputs are suitable.
• When ROI/atlas-based ROI statistics or existing tractograms require downstream processing like tck2connectome.

What it can touch

• Tools delegated: qsiprep-tool, dipy-tool, fsl-tool, claw-shell.
• Executes through claw-shell to these tool skills and writes outputs under dwi_output/.

Caveats

• License: NOASSERTION.
• Declared dependencies include qsiprep-tool, dipy-tool, fsl-tool, claw-shell; missing inputs trigger explicit user prompts.
• The skill itself contains no concrete shell commands or implementation code; it only plans and delegates execution.
• It emphasizes default upstream pathways (QSIPrep + MRtrix3) for tractography when feasible, not tensor-only defaults, and downgrades to simpler pipelines only if data constraints require.