Dataset Release

Skill: dataset-release

Freeze a product at a named version so it can be cited and cross-linked: a version bump, a BIDS CHANGES entry, an immutable datalad version tag, and — when credentials exist — a minted DOI. A release is a durable, resolvable pointer to one exact state (Distributability) with a citable record (Metadata). You own the release judgment; you delegate tagging/saving to the datalad doer and DOI minting to the archive doer.

> Scope note: the version + CHANGES + tag path is always available and fully provenanced. DOI > minting is the gated add-on — the archive doer returns result: unminted (not a fake DOI) when no > Zenodo/OSF/DataCite credential is configured, and the release still completes. Push the tagged > state to a sibling/archive via disseminate/publish.

When to use

• A product exists in the ledger products[] (via analyze/manage-product) and is ready to be

fixed at a version — a dataset release, a paper's data/outputs snapshot, a report.

• Do NOT use to group comparisons (analyze/manage-product), to push working state to a sibling

(disseminate/publish), or to cross-link products (disseminate/link-outputs).

Steps

1. Identify the product + version — get the product id from products[] and the new version

(semver, e.g. 1.0.0; suggest the next bump and confirm). A version is immutable once tagged — never re-tag an existing version; cut a new one.

1. Ensure a clean tree (datalad doer) — a release must tag committed state:

> "status: report modified/untracked files and the current branch." If dirty, route to analyze/checkpoint first.

1. Write release metadata — at the dataset root:

• prepend a BIDS CHANGES entry: followed by bulleted notes.
• record the version where the dataset expects it; once a DOI is minted, set

dataset_description.json "DatasetDOI".

1. Tag the release (datalad doer) — delegate a tagged save:

> "save with a version tag: datalad save -m 'release v' --version-tag v." This creates the immutable, provenanced release point.

1. Mint a DOI (archive doer, gated) — delegate:

> "mint a DOI for tag v of product `` (backend auto: OSF/Zenodo/DataCite)."

• result: ok → capture the doi; set it in dataset_description.json and (step 6) the ledger.
• result: unminted → proceed without a DOI; report how to enable minting. Never invent one.

1. Update the ledger — set the product's status: released; append any minted DOI to the

product's dois[] (per docs/project-ledger.md); append a log entry { ts, op: dataset-release, stage: disseminate, note: "released v; doi ", branch: }. Delegate the save to the datalad doer (a plain datalad save, tree already tagged).

1. Report — the product, version, tag, CHANGES entry, DOI (or "unminted — set ZENODO_TOKEN /

datalad-osf to mint"), and the next step: disseminate/publish to push the tag to a sibling/archive, then disseminate/link-outputs to link this release to other products.

Constraints

• Tag committed state only (clean tree); a version tag is immutable — never move or reuse it.
• Never fabricate a DOI — a DOI enters the ledger/metadata only when the archive doer returns one.
• Every release is a datalad version tag so the exact state is reproducible via datalad from the

tag; do not create releases as plain files without a tag.

• products[] is upserted (status → released) and dois[] appended per the ledger conventions;

keep log: append-only and the ledger schema-valid (schemas/project.schema.json).

• Delegate every tag/save to the datalad doer and every deposit/mint to the archive doer — never

call git/datalad or an archive API directly.

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.