TenantAtlas/specs/108-provider-access-hardening/research.md
ahmido 0dc79520a4 feat: provider access hardening (RBAC write gate) (#132)
Implements provider access hardening for Intune write operations:

- RBAC-based write gate with configurable staleness thresholds
- Gate enforced at restore start and in jobs (execute + assignments)
- UI affordances: disabled rerun action, tenant RBAC status card, refresh RBAC action
- Audit logging for blocked writes
- Ops UX label: `rbac.health_check` now displays as “RBAC health check”
- Adds/updates Pest tests and SpecKit artifacts for feature 108

Notes:
- Filament v5 / Livewire v4 compliant.
- Destructive actions require confirmation.
- Assets: no new global assets.

Tested:
- `vendor/bin/sail artisan test --compact` (suite previously green) + focused OpsUx tests for OperationCatalog labels.
- `vendor/bin/sail bin pint --dirty`.

Co-authored-by: Ahmed Darrazi <ahmed.darrazi@live.de>
Reviewed-on: #132
2026-02-23 00:49:37 +00:00

3.4 KiB

Research — Provider Access Hardening v1 (Intune Write Gate)

Repository Reality Check (existing implementation surfaces)

Write start surfaces

  • Restore execution start (wizard submit / create path): App\Filament\Resources\RestoreRunResource::createRestoreRun()
    • Creates a queued RestoreRun and a corresponding OperationRun of type restore.execute.
    • Dispatches App\Jobs\ExecuteRestoreRunJob with the OperationRun instance.
  • Restore rerun path (action path inside RestoreRunResource): also enqueues restore.execute and dispatches ExecuteRestoreRunJob.

Job execution layer (defense-in-depth insertion points)

  • App\Jobs\ExecuteRestoreRunJob::handle()
    • Calls RestoreService::executeForRun(...) (this is the Graph-mutation path).
  • App\Jobs\RestoreAssignmentsJob::handle()
    • Calls AssignmentRestoreService::restore(...) (Graph-mutation path for assignments).

Tenant view RBAC UX surfaces

  • Tenant view page: App\Filament\Resources\TenantResource::infolist() currently renders a full "RBAC" section with many raw fields.
  • Tenant view header actions: App\Filament\Resources\TenantResource\Pages\ViewTenant::getHeaderActions() includes TenantResource::rbacAction().
  • RBAC setup surface already exists: App\Filament\Resources\TenantResource::rbacAction() creates the "Setup Intune RBAC" action (no new wizard page required).

Existing gating and observability primitives

  • Operation dedupe + provider connection blocking: App\Services\Providers\ProviderOperationStartGate (existing pattern for start surfaces).
  • Long-running work visibility: OperationRun is the canonical monitoring primitive.
  • Audit logging: App\Models\AuditLog exists and Tenant already relates to it.

Decisions (resolved clarifications)

Gate evaluation during health check

  • Decision: Gate evaluates persisted Tenant state only; an in-progress health check has no special effect.
  • Rationale: Keeps the gate DB-only and deterministic; avoids introducing locks/sentinels.
  • Alternatives considered:
    • Block all writes while health check is running (safer but adds operational friction and requires tracking “running” state).

"Setup Intune RBAC" CTA destination

  • Decision: CTA deep-links to the existing Tenant view RBAC section / surfaces (no new wizard page in this feature).
  • Rationale: Minimizes scope and leverages TenantResource::rbacAction() which already exists.
  • Alternatives considered:
    • Build a dedicated setup wizard page for RBAC (adds navigation, testing, and more UX requirements).

Audit logging for blocked attempts

  • Decision: Use existing AuditLog model for UI-level blocked write attempts (P3), in addition to OperationRun failures.
  • Rationale: Meets compliance intent without introducing new schema.
  • Alternatives considered:
    • Create a new audit log table (unnecessary; model already exists).

Config toggle behavior

  • Decision: tenantpilot.hardening.intune_write_gate.enabled=false causes writes to proceed, but logs a warning per evaluation that the gate is bypassed.
  • Rationale: Rollback safety without silent loss of protection.
  • Alternatives considered:
    • Silent bypass (bad observability).
    • Startup-only warning (can be missed; per-evaluation warning is explicit).

Key constraints reaffirmed

  • No synchronous Graph calls in the gate.
  • Start surface must block before job enqueue.
  • Job-level must block before any Graph mutation call.
  • Reason codes are stable and sanitized.