@agenticprimitives/capability-claims

Part of Agentic Primitives — the open-source trust substrate for agentic applications: identity that can sign, authority checked at act time, evidence the owner carries. Developer kit · All packages

Capability claim credentials (SkillClaimCredential) and on-chain SkillDefinitionRegistry helpers (spec 251).

Not Agent Skill packages (SKILL.md). See ADR-0051 and .cursor/rules/skill-terminology.mdc.

Vocabulary

| Term | Meaning | |------|---------| | Capability definition | Stable AP semantic capability (CapabilityDefinition) | | Capability claim | Private VC asserting possession/proficiency | | OASF skill taxon | Classification projection | | A2A AgentSkill | Advertisement projection | | Agent Skill package | SKILL.md procedural playbook | | MCP tool | Executable operation |

Card curation (spec 347 §5)

Pure helpers that take vault claims to a public A2A card without ever widening disclosure:

import { selectPublicSkillClaims, buildA2ASkillCandidates, mapSkillClaimToTaxonomy } from '@agenticprimitives/capability-claims';

const { selected, omitted } = selectPublicSkillClaims(claims, { policyId: 'card-1', maxSkills: 12 });
const { candidates, diagnostics } = buildA2ASkillCandidates(selected, definitions); // id = the stable CURIE
const { mapping, status } = mapSkillClaimToTaxonomy('adv:asset-allocation', mappings, { taxonomy: 'oasf', taxonomyVersion: '1.1.0' });

• Only public / public-coarse claims are ever selected; a policy naming a private visibility discloses nothing. Every omission is reported with a reason.
• A candidate's id is the capability's stable id (the CURIE computeSkillId hashes); proficiency, evidence commitments and nonces never leave the claim.
• SkillTaxonomyMappingV1.relation (exact | close | broad | narrow) is a SKOS mapping relation (skosRelationFor). A mapping asserted for another taxonomy version is review-required, never silently reused; broad/narrow without evidence yields an evidenceNeeded diagnostic.

Import

import { buildSkillClaimCredential, type CapabilityDefinition } from '@agenticprimitives/capability-claims';

Legacy alias: @agenticprimitives/agent-skills re-exports this package.

Seeded catalog (ADR-0051) — src/catalog/

The definitions an agent SELECTS from. listCapabilityDefinitions · getCapabilityDefinition · mappingsFor · isKnownCapabilityId.

Read-only this wave. SkillDefinitionRegistry can enumerate on-chain definitions and is empty; nothing here is minted to it yet, so this module is the source of truth for what may be claimed. When definitions are published on chain it becomes a projection of that registry and the ids do not change — an id is a CURIE, and its keccak is the on-chain skillId.

Why a catalog: the editor used to slug free text into an id, so two agents describing one ability produced two ids and a matcher could not tell they matched.

Mappings assert something, so they are sparse on purpose. No OASF or HCS-26 class id appears unless it can be cited against a published schema; the 45 advisory rows are unmapped, and the explorer says "no mapping at this taxonomy version" rather than showing an invented class that would read as authority. hcs26 is the skill taxon — hcs14 is UAID, an agent identifier, and a capability does not map to one.