# Implementation Plan: System Console Control Tower (Spec 114)

**Branch**: `114-system-console-control-tower` | **Date**: 2026-02-27

## Summary

Implement a platform-only `/system` Control Tower that provides:

- Global health KPIs + top offenders (windowed)
- Cross-workspace Directory (workspaces + tenants) with health signals
- Global Operations triage (runs + failures + stuck) with canonical run detail
- Minimal Access Logs (platform auth + break-glass)

Approach: extend the existing Filament System panel and reuse existing read models (`OperationRun`, `AuditLog`, `Tenant`, `Workspace`) with DB-only queries and strict data minimization/sanitization.

## Technical Context

**Language/Version**: PHP 8.4 (Laravel 12)
**Primary Dependencies**: Filament v5 (Livewire v4), Pest v4, Laravel Sail
**Storage**: PostgreSQL
**Testing**: Pest v4
**Target Platform**: Web (Filament/Livewire)
**Project Type**: web
**Performance Goals**: p95 < 1.0s for `/system` list/index pages at typical volumes
**Constraints**: DB-only at render time; strict data minimization; no cross-plane session bridging
**Scale/Scope**: cross-workspace platform operator views; growing `operation_runs` volumes

**Non-negotiables**

- `/system` is a separate plane from `/admin`.
- Wrong plane / unauthenticated: behave as “not found” (404).
- Platform user missing capability: forbidden (403).
- DB-only at render time for `/system` pages (no Microsoft Graph calls while rendering).
- Data minimization: no secrets/tokens; failures and audit context are sanitized.
- Mutating actions are confirmed + audited.

**Spec source**: `specs/114-system-console-control-tower/spec.md`

## Constitution Check (Pre-design)

PASS.

- Inventory-first + read/write separation: this feature is read-first; v1 manages ops with strict guardrails.
- Graph contract isolation: no render-time Graph calls; any future sync work goes through existing Graph client contracts.
- Deterministic capabilities: capability checks use a registry (no raw strings).
- RBAC-UX semantics: 404 vs 403 behavior preserved.
- Ops observability: reuse `OperationRun` lifecycle via `OperationRunService`.
- Data minimization: `RunFailureSanitizer` + `AuditContextSanitizer` are the contract.
- Filament action safety: destructive/mutating actions require confirmation.

## Project Structure

### Documentation (this feature)

```text
specs/114-system-console-control-tower/
├── spec.md
├── plan.md
├── research.md
├── data-model.md
├── quickstart.md
└── contracts/
    └── system-console-control-tower.openapi.yaml
```

### Source Code (repository root)

```text
app/
├── Filament/
│   └── System/
│       └── Pages/
├── Models/
├── Services/
└── Support/

config/
database/
routes/
tests/
```

**Structure Decision**: Single Laravel web application. System Console features live as Filament Pages under `app/Filament/System/Pages` using existing Eloquent models.

## Phase 0 — Research (Complete)

Output artifact:

- `specs/114-system-console-control-tower/research.md`

Resolved items:

- System panel already exists and is isolated by guard + session cookie middleware.
- Existing audit stream already captures platform auth and break-glass events.
- Existing ops primitives (`OperationRun`, sanitizers, links) are sufficient and should be reused.

## Phase 1 — Design & Contracts (Complete)

Output artifacts:

- `specs/114-system-console-control-tower/data-model.md`
- `specs/114-system-console-control-tower/contracts/system-console-control-tower.openapi.yaml`
- `specs/114-system-console-control-tower/quickstart.md`

Post-design Constitution Check:

- PASS (design remains DB-only, keeps plane separation, uses sanitization contracts, and Spec 114 documents UX-001 empty-state CTA expectations + v1 drilldown scope).

## Phase 2 — Implementation Planning (for `tasks.md` later)

This section outlines the implementation chunks and acceptance criteria that will become `tasks.md`.

### 2.1 RBAC + capabilities

- Extend `App\Support\Auth\PlatformCapabilities` to include Spec 114 capabilities.
- Ensure all new `/system` pages check capabilities via the registry (no raw strings).
- Keep 404/403 semantics aligned with the spec decisions.

### 2.2 Information architecture (/system routes)

- Dashboard (KPIs): global aggregated view, windowed.
- Directory:
  - Workspaces index + workspace detail.
  - Tenants index + tenant detail.
- Ops:
  - Runs list.
  - Failures list (prefiltered/saved view).
  - Stuck list (queued + running thresholds).
  - Canonical run detail: remove current runbook-only scoping so it can show any `OperationRun` (still authorization-checked).
- Security:
  - Access logs list (platform login + break-glass only for v1).

### 2.3 Ops triage actions (v1 manage)

- Implement manage actions with capability gating (`platform.operations.manage`).
- Actions:
  - Retry run: only when retryable.
  - Cancel run: only when cancelable.
  - Mark investigated: requires reason.
- All actions:
  - Execute via Filament `Action::make(...)->action(...)`.
  - Include `->requiresConfirmation()`.
  - Produce an `AuditLog` entry with stable action IDs and sanitized context.

### 2.4 Configuration

- Add config keys for “stuck” thresholds (queued minutes, running minutes).
- Ensure defaults are safe and can be overridden per environment.

### 2.5 Testing (Pest)

- New page access tests:
  - non-platform users get 404.
  - platform users without capability get 403.
- System auth/security regression verification:
  - `/system` login is rate-limited and failed attempts are audited via `platform.auth.login` (existing coverage in `tests/Feature/System/Spec113/SystemLoginThrottleTest.php`).
  - break-glass mode renders a persistent banner and audits transitions (`platform.break_glass.*`) (existing coverage in `tests/Feature/Auth/BreakGlassModeTest.php`).
- Access logs surface tests:
  - `platform.auth.login` and `platform.break_glass.*` appear.
- Manage action tests:
  - capability required.
  - audit entries written.
  - non-retryable/non-cancelable runs block with clear feedback.

### 2.6 Formatting

- Run `vendor/bin/sail bin pint --dirty --format agent` before finalizing implementation.