Agent Watch
Governance drifts. A tool you meant to keep behind a human gate slips into auto-execute; a refactor quietly pushes your agent from advisory to autonomous. Agent Watch computes your agent's AgentAz Trust Level from its agentaz.json and, every time you re-check it, tells you exactly what changed.
Runs entirely in your browser — your spec is never uploaded. Track as many agents as you like; re-paste a spec after you edit it to catch regressions before they ship.
Why this matters
The most common way an agent becomes unsafe isn't a bad design — it's a small, unnoticed change: a tool moved out from behind its approval gate, a bound removed, a capability added. Those changes rarely get a second look. Agent Watch makes the Trust Level a thing you can regression-test, the way you'd test anything else that matters.
Want a one-off grade instead of tracking? Scan a system prompt or generate a Trust Level badge.