Agent Observer Docs
Reference

Status Model

Canonical runtime state and label semantics used across Agent Observer UI.

Status Model

This page defines the language used for runtime status and activity labels.

Agent States

  • idle: no active work
  • thinking: model planning/analysis
  • tool_calling: tool actions running
  • streaming: output actively streaming
  • waiting: blocked on external state/input
  • done: completed successfully
  • error: completed with failure

Scheduler Status

For scheduled tasks:

  • idle: no run yet or paused
  • running: active execution
  • success: last run succeeded
  • error: last run failed

Trigger types:

  • cron
  • manual

Todo Runner Status

For todo-runner jobs:

  • idle: job defined but not running
  • running: currently processing an item
  • success: latest item run succeeded
  • error: latest item run failed

Trigger types:

  • auto (queue progression)
  • manual (explicit user action)

Progress fields to watch:

  • completedTodos
  • failedTodos
  • blockedTodos
  • currentTodoIndex

Activity Label Rules

  • Labels should reflect real underlying actions.
  • CRON is only for scheduled repeated actions.
  • Todo Runner progress is represented through job/todo status, not CRON labels.

Consistency Requirements

  • UI status should update from runtime state transitions.
  • Error surfaces must include actionable text.
  • Panel-level status and automation card status should not diverge.

On this page