# Agent Live Monitor Model

## Rule
Certain agents may always monitor internal system health. External up-to-second information requires explicit provider permission gates and source quality checks.

## Always-Live Internal Monitors
- /deployment-health-gate: Catch build/deployment failures before release. · access: internal only
- /telemetry-alert-gate: Watch internal warning signals and escalate abnormalities. · access: internal only
- /telemetry-provider-health-gate: Watch provider readiness without uncontrolled external access. · access: internal/provider readiness only
- /system-heartbeat-gate: Monitor heartbeat, runtime health, and stale state. · access: internal only
- /registry-review-workbench: Watch registry drift, duplicate growth, and unresolved route decisions. · access: internal only
- /route-divergence-map: Watch route divergence and old/new system mismatch. · access: internal only
- /agent-arena/shell: Watch shell integrity, window routing, and command surface health. · access: internal only

## Permissioned External Live Monitors
- /unified-search-browser-bridge-map: Plan permissioned browser/search bridge access. · provider gate: required
- /live-connection-readiness: Prepare live connection intake and source handling. · provider gate: required
- /weather-source-health: Monitor weather source quality when approved. · provider gate: required
- /weather-activation-gate: Gate weather activation and prevent stale source misuse. · provider gate: required
- /event-anchoring-os: Anchor live events to sources and timestamps. · provider gate: required
- /provider-health-polling-os: Poll provider status only inside approved limits. · provider gate: required

## External Access Rules
- No uncontrolled browsing.
- No secret exposure.
- No provider calls without explicit permission scope.
- Every live result needs source, timestamp, confidence, and freshness.
- Agents may request access but cannot expand their own access.
- Up-to-second info must be labeled as live, source-bound, and permissioned.