Glossary

This glossary defines terms as they are used in the current Agent App standard.

Term	Meaning
Agent App	A complete installable intelligent application package for a host runtime.
APP.md	Required discovery entry with YAML frontmatter and human-readable guide.
Runtime package	UI, worker, storage, workflow, artifact, eval, and support files referenced by APP.md.
Host	Client or platform that installs, projects, authorizes, and runs the app.
Desktop host	Host class for desktop apps, including Lime Desktop Platform, Electron adapters, Tauri adapters, and runtime-backed shells that share the same Agent App contract.
Lime Desktop Platform	Lime's standard desktop host implementation for app center, shared settings, OAuth, OEM, billing, updates, Host Bridge, and Agent App lifecycle.
Host Bridge	Standard event bridge between sandboxed UI and the host, using lime.agentApp.bridge.
Capability SDK	Stable API surface injected by the host, such as lime.ui and lime.storage.
Capability	A host-provided service an app can call through the SDK.
Shared platform capability	Host capability such as lime.cloudSession, lime.modelSettings, lime.branding, lime.billing, or lime.appUpdates that should be shared across apps instead of reimplemented in each app.
Entry	Host-visible launch point such as page, command, workflow, artifact, background task, or settings.
Expert	Chat-first entry. In current manifests it is represented as expert-chat, not as the whole app.
Projection	Deterministic compilation from manifest to host catalog objects.
Readiness	Static setup and compatibility check before runtime execution.
Overlay	Tenant, workspace, user, or customer-specific configuration outside the official package.
Knowledge template	Slot describing required or optional Agent Knowledge binding.
Skill reference	Reference to an Agent Skill that provides procedural capability.
Tool reference	Requirement for an external or host callable tool.
Artifact type	Durable output contract created or viewed by the app.
Eval	Quality gate, readiness check, or human review rule for output acceptance.
Evidence	Provenance and support record for tasks, tools, Knowledge sources, Artifacts, and evals.
Secret slot	Credential placeholder resolved by host secret manager.
Package hash	Integrity hash for package contents.
Manifest hash	Integrity hash for the manifest or APP.md frontmatter.
Local runtime	Host-side execution model where app code runs under host authorization.
Server-assisted	Mode where remote services help with catalog, gateway, or tools without becoming hidden default runtime.

Deprecated or compatibility terms

Term	Status
scene entry	Legacy compatibility term. Current apps should use page, command, workflow, artifact, background-task, or settings.
home entry	Legacy compatibility term. Use page for app home or dashboard.

Naming guidance

Use Agent App for the standard and app for a package instance. Use domain names such as Content Factory App for product examples. Avoid names that imply a single conversation box when the package owns UI, workflow, storage, and artifacts.

Term groups

Group	Terms to review together
Packaging	Agent App, APP.md, runtime package, package hash, manifest hash.
Runtime	Host, Capability SDK, capability, local runtime, server-assisted.
UI and workflow	Entry, expert, projection, readiness, overlay.
Composition	Knowledge template, Skill reference, Tool reference, Artifact type.
Governance	Eval, Evidence, secret slot, policy, compatibility.

How to use this glossary

Use the glossary as a naming guardrail when writing manifests, SDK docs, release notes, and host UI copy. If a proposed term overlaps an existing one, prefer the existing term unless the new concept owns a different runtime boundary.

Ambiguity rules

• If it changes how the agent behaves, start with Skill.
• If it supplies trusted facts, start with Knowledge.
• If it calls an external system, start with Tool.
• If it is a durable output, start with Artifact.
• If it combines UI, workflow, storage, permissions, and lifecycle, use Agent App.