datalad

Run datalad get on the paths you need, since a fresh clone contains only symlinks or pointer files, not content. Use datalad get -r . for everything including subdatasets, or datalad get -n -r . to install subdataset structure without file content.

Use DataLad when files are too large for Git, data lives in multiple locations, analyses must be re-executable, or you consume published datasets from OpenNeuro or DANDI. Use plain Git for code-and-text repositories where everything fits comfortably and no partial checkouts are needed.

The file is a git-annex pointer, not corrupted data; its content has not been retrieved yet. Run datalad get on the path before reading it, and treat file existence as insufficient evidence that content is present locally.

Register an image once with datalad containers-add using a docker:// or shub:// URL, then execute with datalad containers-run declaring inputs and outputs. The image is tracked in the dataset, so datalad rerun can re-execute the exact recorded command and environment.

The Git history was published without the annexed content, or the default --data auto-if-wanted transferred nothing because the remote has no preferred-content settings. Create a storage sibling with git annex initremote, set --publish-depends on the Git sibling, and push with --data anything.

datalad run refuses when the dataset has unsaved modifications, because a record built on an unknown starting state is unreliable. Save or discard changes first with datalad save, or pass --explicit to declare the listed inputs and outputs as the complete scope.