@@ -0,0 +1,21 @@
+---
+'@objectstack/formula': minor
+'@objectstack/spec': patch
+---
+
+feat(formula): `current_user.hasCapability(name)` — a page predicate can ask whether the viewer holds a system capability, from the same list the action gate reads
+
+Clause-②: yes (widening)
+
+- **The predicate.** The CEL standard library registers `hasCapability` beside `can`, receiver-only, so the one spelling is `current_user.hasCapability('<name>')` (the `user` / `ctx.user` / `os.user` aliases work too). It answers `true` when the name is in the acting subject's held system capabilities and `false` when it is not. A name no action declares is just a string: there is no registry lookup.
+- **The data.** `EvalContext` gains `systemPermissions` (type `EvalSystemPermissions`, a list of capability names): the `systemPermissions` array of the published `/auth/me/permissions` response, unchanged — the list an action's `requiredPermissions` is checked against (ADR-0066 D4). `PermissionBinding` gains the same field, optional. Pass it beside `permissions`:
+
+  ```ts
+  celEngine.evaluate(expr, { user, permissions: toEvalPermissions(me.objects), systemPermissions: me.systemPermissions });
+  ```
+
+- **Loud, like `can`.** With no `systemPermissions` on the context the call fails (`ok: false`, `kind: 'runtime'`) rather than answering; an empty list is a real answer and evaluates to `false`. A non-list, a non-string or empty name, or a receiver that is not the acting subject fails the same way. A bare `hasCapability(name)` does not resolve.
+- **The authoring gate.** The visibility-predicate rule's function-existence check reads the evaluation environment, so a page `visibleWhen` calling `current_user.hasCapability(…)` now passes `os validate`, `os lint` and `os build`, where before it was refused as an unknown function (`visibility-predicate-unknown-function`). An invented sibling (`hasCapabilities`) is still refused.
+- **`can` points at it.** `current_user.can(object, verb)` with a verb outside the object-permission vocabulary still fails, and its message now names `current_user.hasCapability('<name>')` for a capability question.
+- **Docs.** The page component `visibleWhen` describe names `current_user.hasCapability('<name>')` as the way to gate a component on a capability: `visibleWhen` hides, `requiredPermissions` refuses.
+- **Hosts.** A host that evaluates page predicates supplies the list for the predicate to answer; one that does not makes it fail, and a failing page `visibleWhen` renders the component.
@@ -311,7 +311,7 @@ View filter rule
 | **style** | `Record<string, string>` | optional | Inline styles or utility classes |
 | **className** | `string` | optional | CSS class names |
 | **responsiveStyles** | `{ large?: Record<string, string \| number>; medium?: Record<string, string \| number>; small?: Record<string, string \| number>; xsmall?: Record<string, string \| number> }` | optional | Per-breakpoint scoped style maps (ADR-0065) |
-| **visibleWhen** | `string \| { dialect: Enum<'cel' \| 'cron' \| 'template'>; source: string; ast?: any; meta?: object }` | optional | Visibility predicate (CEL) — component rendered only when TRUE. Contract-bound roots: `record`, `current_user` (ADR-0068 aliases `user` / `ctx.user` — one object, three spellings), and page state as `page.<var>`. The shipping renderer additionally mounts `features` and `os.user` — renderer behaviour, NOT contract-guaranteed (ADR-0068 rules the user object only). It mounts NO `data` root on an element that no `api` data source binds: there a `data.*` read faults and a faulting predicate renders the component, so such a gate never gates; the page's row is `record` — the record a record page opens, or the row an app page's `record` binding reads (none without one; a block's own `dataSource` binds no `record` for its siblings). A count or a plan is a field on that row (a `summary` roll-up: `record.environment_count > 1`; `record.plan != 'free'`), never `features.*`, which are the deployment's public auth flags, the same for every organisation. `data` is mounted only on an element whose `dataSource` is an `api` data source (`provider: 'api'`): there, and on every element nested inside it, `data` is the value the binding selects, and the element's `visibleWhen` is decided once the read answers. ⚠️ A `page:tabs` item `visibleWhen` is a different evaluator: it binds `data` to the record ROW. e.g. "page.selectedProjectId != ''" |
+| **visibleWhen** | `string \| { dialect: Enum<'cel' \| 'cron' \| 'template'>; source: string; ast?: any; meta?: object }` | optional | Visibility predicate (CEL) — component rendered only when TRUE. Contract-bound roots: `record`, `current_user` (ADR-0068 aliases `user` / `ctx.user` — one object, three spellings), and page state as `page.<var>`. The shipping renderer additionally mounts `features` and `os.user` — renderer behaviour, NOT contract-guaranteed (ADR-0068 rules the user object only). It mounts NO `data` root on an element that no `api` data source binds: there a `data.*` read faults and a faulting predicate renders the component, so such a gate never gates; the page's row is `record` — the record a record page opens, or the row an app page's `record` binding reads (none without one; a block's own `dataSource` binds no `record` for its siblings). A count or a plan is a field on that row (a `summary` roll-up: `record.environment_count > 1`; `record.plan != 'free'`), never `features.*`, which are the deployment's public auth flags, the same for every organisation. `data` is mounted only on an element whose `dataSource` is an `api` data source (`provider: 'api'`): there, and on every element nested inside it, `data` is the value the binding selects, and the element's `visibleWhen` is decided once the read answers. To gate a component on a system capability, write `current_user.hasCapability('<name>')` with the name the gated action's `requiredPermissions` declares — it answers from the viewer's `systemPermissions`, the list the action gate reads, never a proxy such as `current_user.can('<object>', 'edit')`; a host that binds no capability list makes it fault, and a faulting predicate renders the component. ADR-0066 D4: `visibleWhen` hides, `requiredPermissions` refuses. ⚠️ A `page:tabs` item `visibleWhen` is a different evaluator: it binds `data` to the record ROW. e.g. "page.selectedProjectId != ''" |
 | **visibility** | `never` | optional | [REMOVED] `page.components[].visibility` was removed in @objectstack/spec 18 (ADR-0089 D2) — it was the page component's deprecated spelling of the conditional-visibility predicate, which the parse folded into `visibleWhen` until this release. Use `visibleWhen`. Rename the key; the value (a CEL predicate) is unchanged. Where a component writes both, `visibleWhen` is the one that was in effect: delete `visibility`. |
 | **dataSource** | `{ object: string; view?: string; filter?: object[]; sort?: object[]; … } \| { provider: 'api'; read: object; path?: string }` | optional | Per-element data binding: an object binding `{ object, view?, filter?, sort?, limit? }` that overrides the page object, or an API binding `{ provider: 'api', read: { url }, path? }` that reads one same-origin endpoint |
 | **responsive** | `never` | optional | [REMOVED] `page.components[].responsive` was removed in @objectstack/spec 17 (ADR-0049 D2) — no renderer ever read it, so per-breakpoint layout overrides (columns/order/visibility) parsed, validated, and then did nothing. Delete the key. For breakpoint behaviour that IS applied, use the sibling `responsiveStyles` (ADR-0065) — per-breakpoint CSS maps compiled to id-scoped CSS at render, e.g. `responsiveStyles: { xsmall: { display: 'none' } }` to hide a component on the narrowest screens. Run `os migrate meta --from 17` to list the mechanical edits for existing sources; `--write` applies the ones it can prove, and you apply the rest by hand. |
@@ -483,7 +483,7 @@ Sort field and direction pair
 | **style** | `Record<string, string>` | optional | Inline styles or utility classes |
 | **className** | `string` | optional | CSS class names |
 | **responsiveStyles** | `{ large?: Record<string, string \| number>; medium?: Record<string, string \| number>; small?: Record<string, string \| number>; xsmall?: Record<string, string \| number> }` | optional | Per-breakpoint scoped style maps (ADR-0065) |
-| **visibleWhen** | `string \| { dialect: Enum<'cel' \| 'cron' \| 'template'>; source: string; ast?: any; meta?: object }` | optional | Visibility predicate (CEL) — component rendered only when TRUE. Contract-bound roots: `record`, `current_user` (ADR-0068 aliases `user` / `ctx.user` — one object, three spellings), and page state as `page.<var>`. The shipping renderer additionally mounts `features` and `os.user` — renderer behaviour, NOT contract-guaranteed (ADR-0068 rules the user object only). It mounts NO `data` root on an element that no `api` data source binds: there a `data.*` read faults and a faulting predicate renders the component, so such a gate never gates; the page's row is `record` — the record a record page opens, or the row an app page's `record` binding reads (none without one; a block's own `dataSource` binds no `record` for its siblings). A count or a plan is a field on that row (a `summary` roll-up: `record.environment_count > 1`; `record.plan != 'free'`), never `features.*`, which are the deployment's public auth flags, the same for every organisation. `data` is mounted only on an element whose `dataSource` is an `api` data source (`provider: 'api'`): there, and on every element nested inside it, `data` is the value the binding selects, and the element's `visibleWhen` is decided once the read answers. ⚠️ A `page:tabs` item `visibleWhen` is a different evaluator: it binds `data` to the record ROW. e.g. "page.selectedProjectId != ''" |
+| **visibleWhen** | `string \| { dialect: Enum<'cel' \| 'cron' \| 'template'>; source: string; ast?: any; meta?: object }` | optional | Visibility predicate (CEL) — component rendered only when TRUE. Contract-bound roots: `record`, `current_user` (ADR-0068 aliases `user` / `ctx.user` — one object, three spellings), and page state as `page.<var>`. The shipping renderer additionally mounts `features` and `os.user` — renderer behaviour, NOT contract-guaranteed (ADR-0068 rules the user object only). It mounts NO `data` root on an element that no `api` data source binds: there a `data.*` read faults and a faulting predicate renders the component, so such a gate never gates; the page's row is `record` — the record a record page opens, or the row an app page's `record` binding reads (none without one; a block's own `dataSource` binds no `record` for its siblings). A count or a plan is a field on that row (a `summary` roll-up: `record.environment_count > 1`; `record.plan != 'free'`), never `features.*`, which are the deployment's public auth flags, the same for every organisation. `data` is mounted only on an element whose `dataSource` is an `api` data source (`provider: 'api'`): there, and on every element nested inside it, `data` is the value the binding selects, and the element's `visibleWhen` is decided once the read answers. To gate a component on a system capability, write `current_user.hasCapability('<name>')` with the name the gated action's `requiredPermissions` declares — it answers from the viewer's `systemPermissions`, the list the action gate reads, never a proxy such as `current_user.can('<object>', 'edit')`; a host that binds no capability list makes it fault, and a faulting predicate renders the component. ADR-0066 D4: `visibleWhen` hides, `requiredPermissions` refuses. ⚠️ A `page:tabs` item `visibleWhen` is a different evaluator: it binds `data` to the record ROW. e.g. "page.selectedProjectId != ''" |
 | **visibility** | `never` | optional | [REMOVED] `page.components[].visibility` was removed in @objectstack/spec 18 (ADR-0089 D2) — it was the page component's deprecated spelling of the conditional-visibility predicate, which the parse folded into `visibleWhen` until this release. Use `visibleWhen`. Rename the key; the value (a CEL predicate) is unchanged. Where a component writes both, `visibleWhen` is the one that was in effect: delete `visibility`. |
 | **dataSource** | `{ object: string; view?: string; filter?: object[]; sort?: object[]; … } \| { provider: 'api'; read: object; path?: string }` | optional | Per-element data binding: an object binding `{ object, view?, filter?, sort?, limit? }` that overrides the page object, or an API binding `{ provider: 'api', read: { url }, path? }` that reads one same-origin endpoint |
 | **responsive** | `never` | optional | [REMOVED] `page.components[].responsive` was removed in @objectstack/spec 17 (ADR-0049 D2) — no renderer ever read it, so per-breakpoint layout overrides (columns/order/visibility) parsed, validated, and then did nothing. Delete the key. For breakpoint behaviour that IS applied, use the sibling `responsiveStyles` (ADR-0065) — per-breakpoint CSS maps compiled to id-scoped CSS at render, e.g. `responsiveStyles: { xsmall: { display: 'none' } }` to hide a component on the narrowest screens. Run `os migrate meta --from 17` to list the mechanical edits for existing sources; `--write` applies the ones it can prove, and you apply the rest by hand. |
 
@@ -0,0 +1,143 @@
+// Copyright (c) 2026 ObjectStack. Licensed under the Apache-2.0 license.
+
+import { describe, expect, it } from 'vitest';
+
+import { celEngine } from './cel-engine';
+import { firstUnknownFunctionCall } from './unknown-function';
+import { validateExpression } from './validate';
+import type { EvalContext, EvalResult } from './types';
+
+/**
+ * `current_user.hasCapability(name)` — the capability predicate (cloud#3036
+ * ruling A: the action's declaration is the one source of truth, so a page
+ * predicate must be able to ask the question the action gate asks).
+ *
+ * The property under test is the same PAIR `permission-predicate.test.ts` pins
+ * for `can`: given the subject's capability list the predicate evaluates, and
+ * without it the predicate fails LOUDLY — never `true`, never a silent `false`.
+ */
+
+/** The `/auth/me/permissions` `systemPermissions` payload used across these cases. */
+const SYSTEM_PERMISSIONS = ['manage_users', 'view_setup'];
+
+const USER = { id: 'usr_1', positions: ['sales_rep'], organizationId: 'org_1' };
+
+function evaluate(source: string, ctx: EvalContext): EvalResult<boolean> {
+  return celEngine.evaluate<boolean>({ dialect: 'cel', source }, ctx);
+}
+
+function withCapabilities(source: string): EvalResult<boolean> {
+  return evaluate(source, { user: USER, systemPermissions: SYSTEM_PERMISSIONS });
+}
+
+/** A refusal is a reported `runtime` fault naming `needle` — ⛔ never a quiet `false`. */
+function expectLoud(result: EvalResult<boolean>, needle: string): void {
+  expect(result.ok, 'must REFUSE, not answer').toBe(false);
+  if (result.ok) return;
+  expect(result.error.kind).toBe('runtime');
+  expect(result.error.message).toContain(needle);
+}
+
+describe('`hasCapability` answers from the subject\'s held capabilities', () => {
+  it('answers TRUE for a held capability and FALSE for one not held', () => {
+    // Both directions: a pin asserting only `true` stays green under an
+    // implementation that answers `true` for everything.
+    expect(withCapabilities("current_user.hasCapability('manage_users')")).toEqual({ ok: true, value: true });
+    expect(withCapabilities("current_user.hasCapability('manage_billing')")).toEqual({ ok: true, value: false });
+  });
+
+  it('treats a name no action declares as just a string — no registry lookup', () => {
+    // Held ⇒ true, not held ⇒ false, whatever the platform's action catalog
+    // says: the list is the subject's, and nothing else is consulted.
+    const undeclared = 'zz_no_action_declares_this';
+    expect(evaluate(`current_user.hasCapability('${undeclared}')`, {
+      user: USER, systemPermissions: [undeclared],
+    })).toEqual({ ok: true, value: true });
+    expect(withCapabilities(`current_user.hasCapability('${undeclared}')`)).toEqual({ ok: true, value: false });
+  });
+
+  it('answers FALSE from an EMPTY list — a real answer, not a fault', () => {
+    expect(evaluate("current_user.hasCapability('manage_users')", { user: USER, systemPermissions: [] }))
+      .toEqual({ ok: true, value: false });
+  });
+
+  it('answers identically under every ADR-0068 alias of the acting subject', () => {
+    for (const root of ['current_user', 'user', 'ctx.user', 'os.user']) {
+      expect(withCapabilities(`${root}.hasCapability('manage_users')`), root).toEqual({ ok: true, value: true });
+    }
+  });
+
+  it('composes with `can` and the rest of a predicate', () => {
+    expect(evaluate(
+      "current_user.hasCapability('view_setup') && !current_user.hasCapability('manage_billing')",
+      { user: USER, systemPermissions: SYSTEM_PERMISSIONS },
+    )).toEqual({ ok: true, value: true });
+  });
+});
+
+describe('`hasCapability` refuses LOUDLY', () => {
+  it('THROWS when the context carries no capability list — never a silent false', () => {
+    expectLoud(
+      evaluate("current_user.hasCapability('manage_users')", { user: USER }),
+      'carries no capability list',
+    );
+  });
+
+  it('THROWS on a capability list that is not a list of names', () => {
+    expectLoud(
+      evaluate("current_user.hasCapability('manage_users')", {
+        user: USER,
+        systemPermissions: { manage_users: true } as unknown as string[],
+      }),
+      'is not a list of capability names',
+    );
+  });
+
+  it('THROWS when called on anything but the acting subject', () => {
+    expectLoud(
+      evaluate("record.hasCapability('manage_users')", {
+        user: USER, systemPermissions: SYSTEM_PERMISSIONS, record: { id: 'r1' },
+      }),
+      'must be called on it',
+    );
+  });
+
+  it('faults on an unbound `current_user` when the evaluation carries no user, even with a list', () => {
+    // cel-js refuses the unresolved receiver before the binding is reached —
+    // loud either way, as for `can`; this pins WHICH loud.
+    const r = evaluate("current_user.hasCapability('manage_users')", { systemPermissions: SYSTEM_PERMISSIONS });
+    expect(r.ok).toBe(false);
+    expect(r.ok === false && r.error.message).toContain('current_user');
+  });
+
+  it('THROWS on a non-string or empty name', () => {
+    expectLoud(withCapabilities('current_user.hasCapability(1)'), 'must be a system capability NAME');
+    expectLoud(withCapabilities("current_user.hasCapability('')"), 'must be a system capability NAME');
+  });
+
+  it('a bare `hasCapability(name)` names no subject and does not resolve', () => {
+    const result = withCapabilities("hasCapability('manage_users')");
+    expect(result.ok).toBe(false);
+  });
+});
+
+describe('`hasCapability` exists for the publish gates', () => {
+  it('is a registered name — the function-existence oracle reports nothing', () => {
+    expect(firstUnknownFunctionCall("current_user.hasCapability('manage_users')")).toBeNull();
+    // Control: an invented sibling on the same receiver IS reported, so the
+    // null above is a reading, not a blank run.
+    expect(firstUnknownFunctionCall("current_user.hasCapabilities('manage_users')")?.name)
+      .toBe('hasCapabilities');
+  });
+
+  it('validates clean as a boolean predicate', () => {
+    expect(validateExpression('predicate', "current_user.hasCapability('manage_users')").ok).toBe(true);
+  });
+
+  it('points an out-of-vocabulary `can` verb at the capability predicate', () => {
+    expectLoud(
+      evaluate("current_user.can('crm_lead', 'approve')", { user: USER, permissions: {} }),
+      "current_user.hasCapability('<name>')",
+    );
+  });
+});