Skip to content

Glossary

Terms used on this site. Schema terms take their meaning from the LinkML sources in src/asb_schema/; each term links to the page that treats it.

Agent Skill
A directory with a SKILL.md file whose YAML front matter has name and description, followed by Markdown instructions1. Released skill leaves follow this layout.
artifact family
The six artifacts ASB Schema defines: task card, capsule, skill bundle, skill collection, collection registry and exploration trace.
ASTRA analysis
The unit of an ASTRA document (astra.yaml): inputs, outputs, decisions with named options and insights backed by evidence2. A paper becomes a root analysis and each card a sub-analysis.
bridge
A connection from ASB Schema to another format, with a status: existing, feasible or needs new code.
capsule
A directory that packages one task card as a research object, with evidence, per-file provenance and an RO-Crate metadata file (SciTaskCapsule).
closed validation
Validation that rejects keys the target class does not declare, as linkml-validate does.
collection cut
A candidate collection directory with a record MANIFEST.gen.json (asb-cut/v0) of its inputs, output tree and digests, marked released: false.
collection registry
The list of released skill collections by IRI (SkillCollectionRegistry).
direct check
An automated, deterministic check, such as file existence or a value range, in EvaluationStrategy.direct_checks.
evaluation strategy
The required card slot that plans how an attempt is checked: direct_checks and expert_review.
evidence snippet
A verbatim passage from the source with the reason it supports the card (EvidenceSnippet).
expert review
A criterion that needs human judgment, such as scientific validity, in EvaluationStrategy.expert_review.
exploration trace
The research path behind a paper's outcome as a tree of question, experiment, decision, dead-end and pivot nodes (ExplorationTree).
finding
The source's answer or result for a task, in the required slot finding. A solver receives the task text; the evaluator keeps the finding.
Harbor task
An executable task directory: instruction, environment, verifier that writes a reward and, optionally, an oracle3.
mapping (exact, close)
A LinkML link from a slot to a term of another vocabulary: exact_mappings for the same meaning, close_mappings for a similar one56.
MergeAuditRecord
One skill deduplication decision: alias, target slug, similarity evidence and a MergeDecisionEnum value.
minimal conformance
A task_card.json that validates against ASBTaskCard and a ro-crate-metadata.json that conforms to RO-Crate 1.1.
open validation
Validation against the root of a generated JSON Schema file, which accepts undeclared top-level keys.
oracle
In Harbor, a reference solution, solution/solve.sh, that shows a task can be solved4. The task author adds it to a task derived from a card.
Process Run Crate
An RO-Crate profile for tool executions recorded as CreateAction entities with an instrument7.
reference pipeline
The program maintained by the same laboratory that writes cards, capsules, skill bundles and traces from papers and releases skill collections. Writing cards does not require it.
release gate
A check run on a collection directory before publication, in the continuous integration of asb-skill-collections.
reproducibility tier
How far a card's step re-executes: certified, type-checked, bypassed, manual-only or experimental.
research question
The scientific question a task addresses, in the required slot research_question.
reuse readiness
Whether a skill applies outside its source paper, reviewed by the rubric reuse-readiness/v0.
RO-Crate
A specification for research-object packages described by one JSON-LD metadata file based on schema.org8.
SciTask
The name of the standard until 2026-05-24, kept in SciTaskCapsule, scitask_version and the aliases of asb_schema.compat.
skill
A reusable, paper-grounded computational competency, deduplicated across papers (ASBSkill).
skill bundle
A directory describing one skill: skill.md and structured sidecar files (SkillBundle). A bundle describes one skill; a collection releases many under one license.
skill collection
A curated, licensed and citable release of skills and tools, serialized as collection.yaml (SkillCollection).
slim capsule
A release copy of a capsule whose artifact_provenance.json records "profile": "slim"; its card sits in the collection's benchmark/cards/.
slot
A property a class can carry9. ASBTaskCard has 64 slots, 15 required.
support level
Whether a trace step is stated by the paper (explicit) or reconstructed (inferred).
task card
A structured record of one bounded task extracted from a paper (ASBTaskCard), serialized as task_card.json.
task kind
How a card relates to its source: reproduction, analysis, component_reconstruction or extension.
tree root
The class a LinkML file marks as the root of its documents10.
unverified
A point whose check has not been run; such points are listed under Status.

  1. Agent Skills, "Specification". https://agentskills.io/specification ↩

  2. ASTRA, "Specification". https://astra-spec.org/latest/specification/ ↩

  3. Harbor documentation, "Tasks overview". https://docs.harborframework.com/tasks/overview.md ↩

  4. Harbor documentation, "Solution". https://docs.harborframework.com/tasks/solution.md ↩

  5. LinkML Model, "Slot: exact_mappings". https://linkml.io/linkml-model/latest/docs/exact_mappings/ ↩

  6. LinkML Model, "Slot: close_mappings". https://linkml.io/linkml-model/latest/docs/close_mappings/ ↩

  7. Workflow Run RO-Crate, "Process Run Crate". https://www.researchobject.org/workflow-run-crate/profiles/process_run_crate/ ↩

  8. Research Object Crate (RO-Crate), "About RO-Crate". https://www.researchobject.org/ro-crate/about_ro_crate ↩

  9. LinkML, "Models". https://linkml.io/linkml/schemas/models.html ↩

  10. LinkML Model, "Slot: tree_root". https://linkml.io/linkml-model/latest/docs/tree_root/ ↩