🪢 Open source agent evals & observability: Trace, evaluate, and improve LLM applications with one open platform.
TypeScript analyticsautogenevaluationlangchainlarge-language-modelsllama-indexllmllm-evaluationllm-observabilityllmopsmonitoringobservabilityopen-sourceopenaiplaygroundprompt-engineeringprompt-managementself-hostedycombinator
/* eslint-disable @typescript-eslint/no-use-before-define */
import {
hasProjectKind,
type Action,
type Effect,
type OrganizationId,
type ProjectId,
type ResourceId,
type TenantId,
} from "@langfuse/shared/rbac";
import { type Policy } from "./types";
import {
forbiddenError,
type AuthorizationContext,
type Decision,
} from "@/src/features/auth/policy/types";
/** wildcardProjectId is the project-kind wildcard an org role binds to; it matches every project of its tenant but is never a valid authorization target. */
const wildcardProjectId: ProjectId = "project/*";
/** wildcardOrgId is the org-kind wildcard a cross-tenant role binds to; as a tenant it matches every tenant, and as a resource it matches every org target, but it is never a valid authorization target itself. */
const wildcardOrgId: OrganizationId = "organization/*";
/** authorize evaluates the tenant's policies for action on resource: with no matching grant, or any matching deny, it 403s; otherwise it succeeds. A policy on the organization/* tenant matches every tenant, and an organization/* resource matches every org target — the machinery the all-tenant admin role evaluates through. */
export function authorize(
ctx: AuthorizationContext,
tenant: TenantId,
action: Action,
resource: ResourceId,
): Decision {
if (hasWildcard(resource)) return forbiddenError();
const policiesForResource = matchesResource(ctx, tenant, resource);
const grant = policiesForResource.filter(byAction(action));
if (grant.length === 0 || grant.some(byEffect("DENY"))) {
return forbiddenError(
`Access denied. The "${action}" permission is required to access this resource.`,
);
}
return { success: true };
}
/** matchesResource scopes the context to the resource's tenant, preferring an exact-resource match over the kind's wildcard: a project target falls back to project/*, an org target to organization/*. */
function matchesResource(
ctx: AuthorizationContext,
tenant: TenantId,
resource: ResourceId,
): Policy[] {
const owning = ctx.policies.filter(byTenant(tenant));
const exact = owning.filter(byResource(resource));
if (exact.length > 0) return exact;
const wildcard = hasProjectKind(resource) ? wildcardProjectId : wildcardOrgId;
return owning.filter((p) => p.resources.includes(wildcard));
}
/** hasWildcard reports whether resource is a kind wildcard, never authorizable in itself. */
const hasWildcard = (resource: ResourceId): boolean =>
resource === wildcardProjectId || resource === wildcardOrgId;
/** byTenant matches a policy scoped to the resource's tenant, or to the organization/* wildcard tenant that spans every tenant. */
const byTenant = (tenant: TenantId) => (p: Policy) =>
p.tenantId === tenant || p.tenantId === wildcardOrgId;
/** byResource matches a policy listing the exact resource. */
const byResource = (resource: ResourceId) => (p: Policy) =>
p.resources.includes(resource);
/** byAction matches a policy granting the action explicitly; actions are never wildcarded. */
const byAction = (action: Action) => (p: Policy) =>
(p.actions as readonly string[]).includes(action);
/** byEffect matches a policy of the given effect. */
const byEffect = (effect: Effect) => (p: Policy) => p.effect === effect;