Additive-only: creating absent assets without overwriting existing paths.
Apply: explicit permission for a local modifying operation.
Capability: a product-neutral outcome implemented by one or more integrations.
Capability definition: immutable semantic behavior with independent identity, permissions, compatibility, maturity, and evidence.
Capability package: signed distribution containing one or more independently governed capability definitions.
Capability release: immutable package version with hashes, provenance, compatibility, and evidence.
Capability Studio: ForgeHub's schema-driven authoring experience for capability profiles; not a separate product layer.
Canonical document: The single approved source of truth for a policy, architecture boundary, contract, or roadmap decision.
Documentation catalog: Machine-readable inventory of document authority, ownership, lifecycle, freshness, audience, versions, and relevance.
Historical record: Preserved phase, audit, RC, release, or retired evidence that does not represent current product authority.
Standards crosswalk: Evidence mapping to an external framework; not a certification or compliance claim.
AgentSpace: ForgeHub's agents-and-teams discovery and lifecycle area; not a registry, runtime, or independent product authority.
Activation: governed enablement of an installed and configured capability after dependency, permission, and policy resolution.
Source adapter: versioned detector and acquisition adapter for an external registry, repository, catalog, archive, or language package source.
Converter: bounded semantic transformer that creates canonical capability definitions and reports loss, permission differences, and unsupported behavior.
Deployment mode: local, private, offline, air-gapped, or optional cloud operating context; independent from installation scope.
Installation scope: global, organization, workspace, project, shared, or read-only ownership and precedence boundary.
Host adapter: versioned translator between canonical Forgevena capabilities and a host-native format.
Loss report: machine-readable record of native, exact, translated, degraded, manual-adaptation, or unsupported semantics during conversion or host projection.
Bounded loop: orchestration loop with explicit iteration, duration, token, cost, cancellation, termination, and failure limits.
Agent team: governed collection of typed agent roles with declared topology, authority, memory boundaries, budgets, and human control.
Consent gate: approval boundary for external commands, data egress, deployment, or managed deletion.
Integration: lifecycle metadata and adapters for an upstream tool.
Managed asset: file created by the platform and recorded with a content hash.
MCP: Model Context Protocol connection registered and activated per host.
Module: first-party project capability with a common lifecycle contract.
Provider: model API, local model runtime, or agent host.
Registry: non-secret workspace lifecycle state.
Template: stack-specific set of additive source and delivery assets.
Workspace: a project augmented with .ai-workspace state.