Guardian Command Centre
A single protective control plane for the Caelomere Core, client work, identities, integrations, secrets and recovery.
Protect the Core without ever destroying the client’s work. Security action must preserve state, create a recovery checkpoint and leave an audit trail before any high-impact change is allowed.
Combined view of identity, secrets, data isolation, incident posture, backup health, audit coverage and unresolved risk.
Every client workspace must have isolated persistent state, backups and recovery metadata.
Actions Guardian recommends but will not execute until an authorised person approves them.
Measures whether the Core can recover saved state, client work and audit history after a failure.
Guardian Intelligence
No critical compromise is visible in this proving state. Two priority items require attention: a dormant privileged account and one integration scope broader than its observed need. Neither has been changed automatically.
Core protection layers
Never lose the client’s work
Automatic checkpoints
Last-action persistence
Version history
Safe retries
Core Vault
The protected boundary around the Caelomere Core and its critical services.
Core configuration, service state, provider mappings, tenant boundaries, recovery metadata and security policy should be protected separately from ordinary front-end code.
Core configuration
Tenant boundaries
Recovery metadata
Destructive-action protection
Identity & Access
Zero-trust access with least privilege, explicit roles and continuous review.
Being an administrator does not automatically mean being authorised to approve business, finance or security actions.
Privileged identities
Access warnings
Threat Intelligence
Explainable detection across identity, APIs, integrations, data access and abnormal behaviour.
Every finding must show confidence, evidence and impact. A suspicious event is not automatically treated as a confirmed compromise.
| Finding | Severity | Confidence | Evidence | Status | |
|---|---|---|---|---|---|
| Dormant privileged account | High | 97% | No sign-in 67 days; admin retained | Review | |
| Over-broad integration scope | High | 93% | Write permission exceeds observed use | Review | |
| Failed login burst | Medium | 88% | 8 failures in 4 minutes from same source | Monitored |
State & Memory Protection
Ensure no service forgets the last action and no client loses their work.
Workflow state belongs in persistent storage. AI should reconstruct context from authoritative saved state every time it resumes.
Persistence controls
Current local proving state
Data Protection
Protect client records with isolation, encryption, minimisation and traceable access.
Workspace isolation
Encryption
Retention
Secrets & Integrations
Keep API keys and provider credentials out of browsers, logs and user-facing code.
Provider credentials should live in a secure secret store. Guardian tracks age, scope, rotation readiness and downstream dependency health without exposing the secret value.
| Integration | Scope | Secret status | Rotation | Action |
|---|---|---|---|---|
| OpenAI | AI gateway only | Server-side | Policy managed | |
| Cloudflare | Deployment / security | Scoped token | Review due | |
| Banking provider | Read-first / consent | Not connected | N/A |
Backup & Recovery
Recover client work and Core state after provider, deployment, database or human error.
Protecting the Core means being able to restore it. Guardian should verify backups are recent, readable and actually restorable — not merely present.
Recovery sequence
Incident Response
Contain threats without destroying evidence or client state.
Guardian should isolate risky access paths where possible before taking destructive action. Evidence, state and recovery checkpoints are preserved throughout.
Immutable Audit
Every sensitive recommendation, approval and execution leaves evidence.
Production audit should be append-only/tamper-evident, tied to workspace, actor, action, before/after state and provider receipt.
| Time | Actor | Action | Result | Receipt |
|---|---|---|---|---|
| 19:20:18 | Guardian AI | Containment recommendation | Prepared | GDN-4018 |
| 19:14:03 | System | Workspace checkpoint | Verified | CHK-8821 |
| 19:02:44 | Guardian AI | Privilege review | Held for approval | GDN-3997 |
Security Approvals
High-impact security changes remain under authorised human control.
| Action | Risk reduced | Recovery checkpoint | Approval | |
|---|---|---|---|---|
| Disable dormant privileged account | High | Created | Required | |
| Reduce integration permission scope | High | Created | Required | |
| Rotate critical API secret | Medium | Staged rollback ready | Required |
7-Day Assault Test
Try to break Guardian before anyone relies on it.
Security is incomplete if a protective action saves the system but destroys the customer’s work. Every day includes state-preservation and restore testing.