Skip to main content
The Health domain’s base read is an integration smoke test: a 200 with { "status": "ok" } means the API answered, your credential is valid and it holds platform:read. A valid credential without platform:read answers 403. It does not probe downstream capability providers.

Operations

Reads:
  • GET /api/v3/health — getHealth — authenticated liveness check
  • GET /api/v3/health-stops — getStopState — global machine-visible stop state
  • GET /api/v3/health-notifications — getRequiredNotificationHealth — required workflow notification reporting-plane health
  • GET /api/v3/usage-summary — getUsageSummary — Platform API request counts by MCP tool or route, and by outcome

Common workflow

The response data follows the Health schema. Over MCP, forge_call_operation with operationId getHealth reaches the same read.

Emergency controls

Machines can observe whether Forge is stopped or global spend is frozen through GET /api/v3/health-stops or forge_get_stop_state. Agent state reports paused or deprecated, factory state reports PAUSED or ARCHIVED, and refused runs report platform_killed, budget_frozen, agent_unavailable, or factory_paused as applicable. Every stop, freeze, resume, and posture change is a person action in the Console under switches:use, with a named actor, a reason, and an audited rollback. Operators use /trust-fabric/emergency-stops and the agent and factory kill-switch panels. While Forge is stopped, machine writes and knowledge answers are refused with 403 operator_stopped; reads continue so a machine can inspect its state.