Datalad Addurls

Skill: datalad-addurls

Bulk-populate a DataLad dataset by reading a URL manifest and adding each file as an annexed entry with its download URL registered as the annex remote. DataLad will fetch content on demand with datalad get.

Steps

1. Verify DataLad context — check for .datalad/ in the current directory or a

parent: ``bash ls .datalad/ 2>/dev/null ` If no dataset is found, ask the user whether to initialize one first with /datalad-init`.

1. Identify the manifest file — read from $ARGUMENTS or conversation context.

The manifest is a file with one row per file to ingest. Supported formats:

• CSV / TSV: columns delimited by comma or tab
• JSON / JSON-Lines: array of objects or one object per line

If no manifest is provided, ask the user to specify the file path.

1. Inspect the manifest — read the first few lines:

``bash head -5 `` Identify the column (or JSON key) that contains:

• The URL to download from
• The local filename (relative path inside the dataset)

Ask the user to confirm or correct the column names if ambiguous.

1. Gather parameters — confirm:

• --url-format (or positional URL column): the column/key holding the URL
• --filename-format (or positional filename column): the column/key for the local path
• --jobs N: parallel downloads (default 1; suggest 4–8 for large manifests)
• -m: commit message for the resulting dataset commit (required; the auto-generated

message is unhelpful — always ask the user for a meaningful message)

• Whether to --ifexists skip (skip already-present files) or overwrite
• --missing-values: how to handle optional URL fields that are absent in some rows

(e.g., --missing-values skip to skip rows with missing URLs)

1. Construct and show the command:

For CSV/TSV: ``bash datalad addurls \ -m "" \ [--jobs ] \ [--ifexists skip] \ \ '' \ '' ``

For JSON (keys as format strings): ``bash datalad addurls \ -m "" \ --url-format '{url}' \ --filename-format '{name}' \ --jobs 4 \ .json \ '{url}' \ '{name}' ``

Display the full command and ask: > "Ready to execute? (yes / edit)"

1. Execute and report:

• On success: report the number of files added, the commit hash created, and that

content is pointer-only (not yet downloaded) — run datalad get . to fetch everything.

• On failure: show the error. Common causes: malformed URLs in manifest, wrong

column names, network errors. Suggest checking the manifest format.

1. Post-ingestion suggestions:

• To download all content immediately: datalad get .
• To download in parallel: datalad get -J 8 .
• To verify: datalad status and git annex whereis

Constraints

• Always inspect the manifest before building the command — never guess column names.
• Always show the full datalad addurls command before executing.
• Always require a meaningful -m message.
• Warn the user that addurls registers URLs as annex remotes but does not download

content by default — files will be pointer-only until datalad get is run.

• For large manifests (>100 files), recommend --jobs 4 or higher and note that the

initial run creates the commit immediately; downloading happens separately.

• Never edit the manifest file — only read it to determine column names.

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.