Executable Article

Skill: executable-article

Turn a product into a re-executable article — figures regenerate from the provenanced outputs in the project's pinned container, rather than being pasted in. That makes the paper Actionable (re-runs) and Portable/Ephemeral (rebuilt from spec). You delegate history/ledger reads and the save to the datalad doer.

Load plugins/disseminate/references/neurolibre-structure.md for the scaffold and how each piece maps to the harness before generating.

When to use

• A product (usually a paper) is analyzed and (ideally) released, and the author wants a

reproducible executable-article form.

• Do NOT use to write prose (draft-manuscript) or to release/DOI a version (dataset-release);

this generates the executable build around the results.

Steps

1. Identify inputs — the product id, its comparisons' outputs (derivatives/cmp-/), the

containers/ recipe (for the environment), and the dataset's published location (the dataset-release DOI or the DataLad sibling — from the ledger). If the dataset is unreleased, note that repo2data should point at the sibling and a DOI be wired at release.

1. Scaffold the article (per the reference) at article/ (or the author's path): myst.yml,

paper.md, content/ figure notebooks, and binder/ with an environment derived from containers/ plus a data_requirement.json (repo2data) resolving the published dataset.

1. Wire figures to provenance — each content/ notebook computes its figure from the

provenanced derivatives/… outputs (not a static image), so the build reproduces them.

1. Register + log — add the article path to the product's outputs[]; append

{ ts, op: executable-article, stage: disseminate, note: "NeuroLibre article for ", branch: }.

1. Save — delegate to the datalad doer: "save: datalad save -m 'executable-article: scaffold '."
2. Report — the article path, how to build/preview (myst build / Jupyter Book), what still

needs wiring (e.g. the dataset DOI once released), and the next step: link-outputs to relate the article to the dataset (Documents) and paper (IsSupplementTo).

Constraints

• Figures must regenerate from provenanced outputs + the pinned environment — do not embed static

images as the source of truth; the article's value is re-executability.

• The binder/ environment derives from the project's container recipe/digest — keep it consistent

with what analyses actually ran in; do not invent dependencies.

• repo2data points at the published dataset (DOI or sibling) — never a local absolute path.
• Record the article under the product's outputs[]; keep log: append-only and the ledger

schema-valid. Delegate reads/saves to the datalad doer.

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.