Skills
The smallest reusable unit. Single capabilities with typed I/O, required tools, and evaluator bindings.
Signed, attested, version-pinned skills, agents, and policy templates that auditors can actually trace.
Every artifact your agents depend on - registered, versioned, and signed.
The smallest reusable unit. Single capabilities with typed I/O, required tools, and evaluator bindings.
Compositions of skills with an orchestration plan, prompt pack, and trust tier ceiling.
LLM-as-judge scorers plus custom code, rating outputs against declared failure modes.
Tool definitions registered against the gateway, with scopes and decision-boundary templates.
OPA/Rego bundles - SOX, IFRS, materiality, segregation of duties - parameterized per entity.
Versioned prompt collections from Langfuse with golden datasets attached.
645+ ingestion adapters: Gmail, SAP, Snowflake, and the long tail, formally registered.
Domain Context Graph versions. Pin to them, extend them, never collide.
No skill reaches a workspace without passing every one.
Local to publisher
Evaluators + golden set
Publisher key + attestations
Discoverable in catalog
Bound + trust tier set
Entity types, trust tiers, publishing gates, and connectors — all accounted for.
Regulated tenants restrict to official. Everyone else opts in.
Platform-signed
Vetted publishers
Off by default
Join the registry. Ship skills that regulated workspaces can trust.