Skill collection¶

A skill collection is a curated, licensed and citable release of skills and tools. It is serialized as a collection.yaml manifest described by the class SkillCollection in src/asb_schema/asb_skill_collection.yaml, which imports asb_skill_bundle.yaml. Each released version of a domain collection has its own directory, collections/<domain>/v<N>/, and its own manifest. The released collections are listed by a collection registry.

asb_skill_collection.yaml 0.3.0 https://w3id.org/holobiomicslab/asb-schema/skill-collection/v0.3 normative 6 of 25 slots required

Example¶

The start of collections/metabolomics/v2/collection.yaml in asb-skill-collections, shortened: the description, attribution blocks and the keys after the skill list are omitted.

Example

'@context':
  '@vocab': https://schema.org/
  asb: https://w3id.org/holobiomicslab/asb-skill/
  edam: http://edamontology.org/
'@type': asb:SkillCollection
'@id': https://w3id.org/holobiomicslab/asb-skill/collection/metabolomics/v2
iri: https://w3id.org/holobiomicslab/asb-skill/collection/metabolomics/v2
title: ASB Metabolomics Skill Collection v2 (expansion)
slug: metabolomics
version: 2
license: CC-BY-4.0
doi: 10.5281/zenodo.20794027
zenodo_doi: 10.5281/zenodo.20794027
skills_count: 5859
tools_count: 909
workflows_count: 21
skills:
- 2d-nmr-pulse-sequence-implementation
- 2d-tic-preprocessing
- 4d-lcimmsms-feature-extraction
# 5,856 more

SCHEMA=$(python -c "from asb_schema import get_schema_path; print(get_schema_path('asb_skill_collection'))")
linkml-validate --schema "$SCHEMA" --target-class SkillCollection collection.yaml
# No issues found

The four manifests of the collections repository pass this closed validation, and the repository's release workflow runs the same check on every collection.yaml.

Fields¶

Required slots are in bold. The slots jsonld_context, jsonld_type and jsonld_id serialize as @context, @type and @id.

Group	Slots	Content
Identity	`iri`, `slug`, `version` (integer), `title`, `description`, `@context`, `@type`, `@id`	IRI of the release, `<N>` matching the `vN` of the IRI
License and citation	`license` (SPDX), `doi`, `zenodo_doi` (pattern `^10\.5281/zenodo\.[0-9]+$`), `released_at`	Terms of reuse and the archived deposit
Content	`skills`, `tools`, `skills_count`, `tools_count`, `workflows_count`	The released skills and tools, by slug or IRI, and their counts
Topics	`edam_topics`, `domain_topics`	EDAM topic IRIs of the whole release
Lineage	`source_collections`, `catalogue_membership`, `ro_crate_path`	Prior collections, the registry that lists this one, the release crate
Attribution	`contributors`, `lead_curators`, `roles`	People and roles; see below

Each slot is in the field reference. The release gate report gate_report.json of each collection binds the payload tree by SHA-256.

Attribution¶

Slot	Class	Attributes
`contributors`	`Person`	`name` (required), `orcid` as a URI (`https://orcid.org/...`), `affiliation`
`lead_curators`	`Curator`	`name` (required), `orcid` as a bare iD (pattern `^[0-9]{4}-[0-9]{4}-[0-9]{4}-[0-9]{3}[0-9X]$`), `github`
`roles`	`CollectionRoles`	`generator` (the program that produced the content), `promoter`, `sponsor`, `curators` (people who modified or validated the content), `zenodo_authors` (creators of the deposit, in order)

The manifest writer derives contributors from lead_curators, converting each bare iD to its URI form. CollectionRoles mirrors the per-skill attribution block of the released leaves.

Release layout¶

collections/<domain>/v<N>/
├── collection.yaml       the manifest (SkillCollection)
├── leaves/<slug>/SKILL.md   one leaf per skill (Agent Skills layout)
├── tools/<slug>.yaml     one record per tool
├── workflows/            multi-skill workflows, when declared
├── gate_report.json      release gate report with payload hashes
└── .zenodo.json          deposit metadata

How a collection is assembled and released is described in How the artifacts are produced; how to publish one is in Converting and publishing.

Collection registry¶

A collection registry lists released collections by IRI; it does not hold their content. The class SkillCollectionRegistry is the root of asb_skill_collection_registry.yaml (0.2.0, https://w3id.org/holobiomicslab/asb-schema/skill-collection-registry/v0.2), which imports asb_skill_collection.yaml for Person.

Field	Type	Required	Definition
`registry_iri`	uriorcurie	yes	Stable IRI of the registry (`schema:url`)
`registry_title`	string	yes	Title (`schema:name`)
`collections`	list of uriorcurie	yes	IRIs of the listed `SkillCollection` releases, one per version
`registry_description`	string		Description (`schema:description`)
`registry_maintainers`	list of `Person`		People who curate the registry
`generated_at`	datetime		Time of the snapshot (`prov:generatedAtTime`)
`external_repos` reserved	list of uriorcurie		Future federation: other repositories whose collections the registry aggregates

A minimal registry:

{
  "registry_iri": "https://w3id.org/holobiomicslab/asb-skill/registry",
  "registry_title": "ASB skill collections",
  "collections": [
    "https://w3id.org/holobiomicslab/asb-skill/collection/metabolomics/v2"
  ]
}

SCHEMA=$(python -c "from asb_schema import get_schema_path; print(get_schema_path('asb_skill_collection_registry'))")
linkml-validate -s "$SCHEMA" --target-class SkillCollectionRegistry registry.json
# No issues found

The registry IRI https://w3id.org/holobiomicslab/asb-skill/registry resolves to the catalog catalogue.jsonld of the collections repository, and each collection IRI https://w3id.org/holobiomicslab/asb-skill/collection/<slug>/v<N> to the release directory. The catalog is a schema.org view of the same list, with name and description and the collections as inline objects. linkml-validate selects its loader by file extension, so a .jsonld file is copied to .json before validation.