Datalad Clone

Skill: datalad-clone

Obtain a copy of a DataLad dataset from a URL or local path. Optionally register it as a subdataset inside an existing DataLad dataset (YODA-style nested layout).

Steps

1. Identify source — read the source URL or path from $ARGUMENTS. If not provided,

ask the user for the source before continuing. If the URL begins with ria+, this is a RIA store. Accepted formats: ria+ssh://user@host/path, ria+http://host/path, ria+file:///local/path. Load ${CLAUDE_PLUGIN_ROOT}/../references/siblings-and-remotes.md for RIA store URL details.

1. Determine mode — decide whether this clone should be:

• Standalone: a new independent dataset (no -d flag)
• As a subdataset: nested inside the current dataset (-d .)

Check whether the current working directory is a DataLad dataset: ``bash ls .datalad/ 2>/dev/null ``

• If inside a dataset and the user mentions "input data", "nested", "subdataset",

or a target path like inputs/raw/sub-01 — default to subdataset mode

• If not inside a dataset or the user says "standalone" — use standalone mode
• If ambiguous — ask:

> "Should this be registered as a subdataset of the current dataset, or cloned > as a standalone dataset?"

1. Determine destination path — read from $ARGUMENTS or derive from the source

name (last path/URL component, minus .git). Show the planned destination and confirm if it differs from what the user expects.

1. Construct and show command:

• Standalone:

`` datalad clone [] ``

• As subdataset:

`` datalad clone -d . ` When cloning on the same filesystem (e.g., HPC scratch → project directory), --reckless=shared-local` skips the safety copy and is substantially faster. Only use on trusted local filesystems. Always show the full command before executing.

1. Execute — run the command. Report the installed path and the commit SHA recorded.

1. Post-clone reminder (subdataset mode only) — after cloning as a subdataset, the

superdataset's working tree is modified but not yet committed. Remind the user: > "The subdataset pointer is not yet saved. Run: > datalad save -m 'add as subdataset' > to record it in the superdataset history."

Reference

Load ${CLAUDE_PLUGIN_ROOT}/../references/subdataset-patterns.md when the user asks about nested layouts, why -d . is needed, or the YODA neuroimaging project structure.

Load ${CLAUDE_PLUGIN_ROOT}/../references/yoda-layout.md when the user asks about the YODA directory layout, .gitattributes rules, or how inputs/ and outputs/ are structured.

Constraints

• Never use git clone for DataLad datasets — use datalad clone to preserve annex

configuration and special remote registrations.

• Never use datalad install — it is a deprecated alias for datalad clone.
• Never use git submodule add directly — DataLad manages submodule entries via

datalad clone -d ..

• Always show the full command before executing.
• Always prompt for a follow-up datalad save after a subdataset clone.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: neuroanalytics
• Source: neuroanalytics/data-science-harness
• License: MIT

Install and usage instructions live in the source repository linked above.