@adonis-agora/agent/dist/src/a2a/gate.d.ts

Governed, durable-backed AI agent (chat + tool-calling + governance) for AdonisJS — part of the Agora ecosystem.

import { type ScopeResolver } from '../skills.js';
import type { RolesPolicy } from '../spi/roles-policy.js';
/** The role a delegated scope becomes. Namespaced, so a scope can never pose as an app role. */
export declare const SCOPE_ROLE_PREFIX = "scope:";
export declare function scopeRole(scope: string): string;
/**
 * Wrap a {@link RolesPolicy} so personal agents reach only what was written for them.
 *
 * An A2A actor carries {@link PERSONAL_AGENT_ROLE}, plus one {@link scopeRole} per delegated scope.
 * For it, a tool must DECLARE one of those roles, and the wrapped policy must agree too. A tool with
 * no `roles` is unreachable: under the default policy "no roles" means "anyone", and an app's
 * existing tools were never written with an outside caller in mind. Every other actor is decided by
 * the wrapped policy alone, unchanged.
 */
export declare function personalAgentGate(inner: RolesPolicy): RolesPolicy;
/**
 * Wrap the {@link ScopeResolver} of memory or skills so a personal agent resolves NO scope. Under
 * delegation its actor id is the account's — without this it would read the account's memories
 * (and its tenant's) and could write new ones, a standing instruction to the user's own assistant.
 * Built-ins like `remember` come from config, not the registry, so the tool gate never sees them.
 */
export declare function personalAgentScopes(inner: ScopeResolver | undefined): ScopeResolver;
//# sourceMappingURL=gate.d.ts.map