Link Outputs

Skill: link-outputs

Record how the project's products relate to each other (and to external identifiers), so the dataset, paper, executable article, and agent bundle form one linked compendium rather than scattered artifacts. The links are DataCite RelatedIdentifier relations, stored in each product's relations[] in the ledger (Metadata) and making the multi-product set resolvable as a coherent whole (Distributability). You own which links are meaningful; you delegate the ledger save to the datalad doer.

Load plugins/disseminate/references/datacite-relations.md for the valid relationType terms and their inverses before recording any relation.

When to use

• Two or more products exist in products[] (via analyze/manage-product), typically after some

are released (disseminate/dataset-release), and you want to record how they relate.

• Do NOT use to create products (manage-product) or to mint DOIs / cut versions

(dataset-release). This skill only records relations between things that already exist.

Steps

1. Identify the relations — for each link, determine the source product id, the

relationType (from the reference — e.g. IsSupplementedBy, IsSourceOf, IsDocumentedBy), and the target: another product id in this ledger (internal) or an external resolvable DOI/URL. Ask the user for anything ambiguous; do not guess how products relate.

1. Validate both ends — the source must be a product in products[]. For an internal target,

confirm the target id exists. Warn (do not block) if an internal target is not yet released (no DOI) — the id link still resolves once it is released; suggest dataset-release if the user wants a DOI now. Never fabricate a DOI for an unreleased target.

1. Record the relation (and its inverse) — upsert into the source product's relations[]:

{ relation: , target: } (per docs/project-ledger.md; do not duplicate an identical relation). For an internal target, also record the inverse relation on the target product (see the reference table) so the graph is consistent both ways. For an external target, record only the forward relation.

1. (Optional) mirror external links into dataset_description.json for the dataset product

(e.g. a DOI the dataset References), keeping the ledger as the canonical record.

1. Log it — append { ts, op: link-outputs, stage: disseminate, note: " (+inverse)", branch: }.
2. Save — delegate to the datalad doer: "save: datalad save -m 'link-outputs: '."
3. Report — the relation graph (source → relation → target for each link), any targets still

needing a DOI, and whether the compendium is now fully linked. When every product is released and linked, the multi-product research object is complete.

Constraints

• Use only valid DataCite relationType terms (see the reference); record the inverse for internal

product-to-product links.

• Never fabricate a DOI — link by product id (internal) or a real resolvable identifier only.
• relations[] is upserted per the ledger conventions (no duplicate {relation,target} pairs);

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

• Relations describe existing products — this skill never creates, versions, or moves a product.
• Delegate every DataLad save 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.