Install

pnpm add @aooth/arbac-moost @aooth/arbac @aooth/user

Documentation

Full docs, API reference, and recipes: https://aooth.moost.org/moost/

• Setup
• @ArbacAuthorize
• DB controllers (AsArbacDbController)
• Atscript wiring (@arbac.* annotations)
• Decorators
• API reference

Custom action handlers — your responsibility

The @ArbacAuthorize interceptor gates the entry to a @DbAction handler: if the user lacks the privilege, the handler doesn't run. But what the handler does inside its body is on you — the interceptor doesn't follow secondary database calls or external side effects.

If your handler reads or writes rows from the same table the entry-gate scopes, route those calls through the active scope. useArbac().getScopes<ArbacDbScope>() exposes the per-event scope array; from there:

• scope.filter — AND-merge into your WHERE clause (use the union of all scope filters via mergeScopeFilters from @aooth/arbac, not just the first).
• scope.set — overlay onto any insert/update payload after your own body keys, so scope-set wins over caller-supplied values.
• scope.projection — restrict any custom SELECT to the union of allowed fields.

The demo at packages/e2e-demo/src/controllers/_helpers.ts shows the convention (scopedFilter, scopedSet); the test ACT-08 in packages/e2e-demo/test/arbac-actions.spec.ts pins that handlers using this pattern correctly enforce row-level scope.

If your handler talks to a different system (audit log, cache, external API, sibling table) or has special logic (cross-tenant aggregations, admin escape hatches), you wire the scope to whatever surface that system exposes — there is no automatic propagation, and the trust decisions stay explicit in the handler.

Why not enforce structurally? The framework can't know, at the action-entry point, what the handler intends to read or write. A handler that calls a remote service, spans transactions across tables, or has admin-only branches won't compose cleanly with an automatic row-level gate. Keeping scope enforcement explicit makes the trust decision a code-review-visible artifact. The tradeoff is that a forgetful contributor can bypass it — mitigate via code review and ACT-08-style regression tests.

License

MIT