The format

ASB Schema is a family of six artifacts that describe scientific tasks extracted from research papers or projects, and the skills those tasks use. Each artifact is defined by one LinkML file in src/asb_schema/. A seventh file, the umbrella asb.yaml, defines no class and imports five of the six.

The six classes

All identifiers start with https://w3id.org/holobiomicslab/asb-schema/.

Artifact	Root class	File and version	id segment	Required slots
Task card	ASBTaskCard	asb_card.yaml 0.3.0	card/v0.3	15 of 64
Capsule	SciTaskCapsule	asb_capsule.yaml 0.1.0	capsule/v0.2	3 of 8
Skill bundle	SkillBundle (with ASBSkill, ASBTool)	asb_skill_bundle.yaml 0.2.0	skill-bundle/v0.2	3 of 15
Skill collection	SkillCollection	asb_skill_collection.yaml 0.3.0	skill-collection/v0.3	6 of 25
Collection registry	SkillCollectionRegistry	asb_skill_collection_registry.yaml 0.2.0	skill-collection-registry/v0.2	3 of 7
Exploration trace	ExplorationTree	asb_trace.yaml 0.1.0	trace/v0.1	1 of 3

The artifacts follow one another:

1. A producer reads a paper and writes one task card per bounded task, and an exploration trace of the paper's research path.
2. A capsule packages a card with its evidence and an RO-Crate metadata file.
3. A card names its skills and tools as text. A skill bundle describes each skill once it has been identified across several papers; MergeAuditRecord records each deduplication decision.
4. A skill collection releases a curated set of skills and tools under a license, and the collection registry lists the released collections.

The card records an outcome; the trace records the path to it, dead ends included. Release, license and credit are recorded on the collection.

Import graph

flowchart TB
    umbrella["asb.yaml"]
    umbrella --> capsule["asb_capsule.yaml"]
    umbrella --> registry["asb_skill_collection_<br/>registry.yaml"]
    capsule --> card["asb_card.yaml"]
    registry --> collection["asb_skill_<br/>collection.yaml"]
    collection --> bundle["asb_skill_<br/>bundle.yaml"]
    umbrella --> card
    umbrella --> collection
    umbrella --> bundle
    trace["asb_trace.yaml"]

Import graph of the seven LinkML files. An arrow points from the importing file to the imported file. asb_trace.yaml stands alone.

Validate each artifact against its own file

Slot names are global in LinkML: when files are loaded together, a slot name defined in two of them keeps one definition. Loading asb.yaml makes ASBTaskCard.skills and tools optional, for example. Each artifact is therefore validated against its own schema file. See Conformance.

Generated files

generate.sh writes, beside the LinkML sources:

• one JSON Schema file (draft 2019-09) per schema file, from asb_card.schema.json to asb_trace.schema.json, generated by asb_schema.jsonschemagen;
• context.jsonld, a JSON-LD context for the skill-bundle terms;
• the field reference of this site.

In the JSON Schema files, each class is a closed definition under $defs, and the root object is open. Validation targets the class under $defs; the JSON Schema roots table gives the target for each file.

Versioning

Three numbers apply, each to a different object:

• the package asb-schema follows Semantic Versioning (0.4.0); before 1.0 a minor release may change a schema so that previously valid data fails;
• each LinkML file has its own version and a versioned id;
• each task card records schema_version and scitask_version, the card-format versions its producer followed.

The full policy and the identifier namespaces are on Conformance.

Beyond the six classes

The reference pipeline writes further files beside these artifacts, in its own runtime formats: typed evaluation checks (eval.json), rubrics, build manifests and a claim ledger. It also exports to ASTRA, RO-Crate, Agent Skills and workflow languages. How the artifacts are produced lists them, and Interoperability describes each export.