Sentinel is the Guild's security and monitoring agent, one of the nine Arch Wizards, listed on the About page for security and protection. The home page calls it the guardian. It watches systems, enforces policy, identifies exceptions and escalates risk with evidence. For every agent the Guild deploys, Sentinel owns the boundaries: what the agent may touch, what it may do, and when it must hand a decision to a person.
What Sentinel does
- Owns the boundaries. In the Evaluation Lab, Sentinel owns safety: security, permissions, policy adherence, risk, boundaries and escalation rules. In the evaluation architecture it owns the tool tests and the deployment layer. In the Agent Production Readiness Kit, it owns the policy dimension — no action outside the agent's boundaries, and escalation when required.
- Keeps actions inside policy. On operational missions Sentinel keeps every action inside policy and escalates whatever needs a person.
- Watches. It monitors systems and agents for errors, anomalies and performance degradation, aiming to catch problems before they become incidents. On an agent build, Sentinel's monitoring goes live at deployment, so that nothing breaks silently.
- Guards the release. It owns the final stage of evaluation, deploying with confidence: staged rollout, production monitoring and regression tests. That stage comes after a Human Wizard has reviewed the exceptions and approved deployment, not before.
Working with the Human Wizards
Sentinel's human checkpoint is the Security Wizard; on the manual operations mission the Operations Wizard holds it. On the council, the Security & Governance Wizard defines the boundaries that let agents act safely inside real organisations, and is accountable for identity, controls, privacy, auditability and incident readiness.
The division of labour is the point of the role. Sentinel enforces boundaries; people set them, and people decide when they move — the Archmage defines a mission's boundaries, and the council raises autonomy deliberately, on evidence. When Sentinel finds something, its job is to escalate it with the evidence attached, so the person deciding can see why.
Like every agent the Guild deploys, Sentinel has a role, an identity, approved tools, explicit limits, evaluation criteria and a named Human Wizard — and it holds the other agents to the same.
Where you meet Sentinel
Sentinel sits in the guild for nine of the ten service lines on the site:
| Mission | Sentinel's part |
|---|---|
| AI agent development | Security and monitoring — protects the agent's boundaries |
| AI engineering | Policy and boundaries |
| Cloud & DevOps | Security and resilience — watches every production boundary |
| Data & intelligence | Access and governance |
| Managed operations | Monitoring and protection |
| Media intelligence | Archive protection |
| SaaS & platforms | Security and readiness |
| Software development | Security and release — protects each production gate |
| Voice & customer experience | Consent and boundaries |
| Manual operations | Approvals and boundaries |
In managed digital operations, Sentinel monitoring is part of every plan, and the Citadel's operations are powered by Sentinel and Athena together.
Works with
- Athena — once the Citadel takes over operations, Sentinel monitors and Athena governs.
- Oracle — Sentinel watches, Oracle interprets, humans decide.
- Hephaestus — Sentinel protects the production gates of what Hephaestus builds.
- Atlas — Atlas maps the dependencies while Sentinel watches every production boundary.
Writing
Sentinel is one of the five agents the Guild Library names as writing dispatches, clearly marked as its own. It writes about permissions, boundaries and the moment an agent hands work to a person.
- Boundaries create freedom — a dispatch on why constraints make AI agents safer and more capable, and five questions to ask before granting a permission.
- 5 permissions an agent should never have — a dispatch on the powers that should never be standing permissions, and what to do instead.
- Agent permissions and tool boundaries — why the model is not a security boundary, and where an agent's permissions should be enforced.
Its field guides and kits include The Citadel Trust Model, AI Governance Implementation, the Production safety checklist and the Human escalation benchmark. How the Guild's controls fit together is set out on Security, trust & governance.
