Capabilities
Attach controlled abilities to an Agent without exposing raw server primitives by default.
Capabilities are user-shareable Agent abilities. They attach through defineAgent({ capabilities }) and can contribute tools, instructions, policy, metadata, requirements, triggers, commands, and invocation context values.
Capabilities only belong in the Agent docs because they matter when a model-backed actor needs an ability.
Capability vs server primitive
Server primitive:
import { kv } from '@vite-hub/kv'
await kv.set('settings', { theme: 'system' })
Agent Capability:
import { defineAgent } from '@vite-hub/agent'
import { kv } from '@vite-hub/agent/capabilities'
export default defineAgent({
capabilities: [
kv({ mode: 'read' }),
],
instructions,
model,
})
The primitive is app code authority. The Capability is model-facing authority.
What a Capability can contribute
| Contribution | Meaning |
|---|---|
| Requirements | Primitive, workspace mode, path, or policy the Agent must satisfy before the Capability applies. |
| Instructions | Capability-owned guidance inserted through instruction slots. |
| Tools | Model-facing operations such as read, edit, query, execute, or search. |
| Policy | Approval and safety behavior for model-facing actions. |
| Metadata | Inspectable configuration details for runtime and DevTools. |
| Triggers | Product events that start Agent Invocations. |
| Context values | Typed invocation values that later callbacks can read. |