agent-run-schema

A unified agent run event schema provides a canonical set of 10 observability events, such as task_started and tool_call_failed, that share an agent_run_id to reconstruct execution timelines across debugging sessions and dashboards.

You track subtask activity by emitting canonical events like task_started and task_completed with a shared agent_run_id, enabling cross-tool correlation and reliable run reconstruction for your agent pipelines.

You debug agent pipelines by emitting the 10 defined observability events with a shared agent_run_id, allowing you to join all run and subtask activity into a single timeline for cross-tool correlation.

Yes, the agent observability schema supports unified dashboards and sinks including Postgres, SQLite, and log streams to capture run-level activity and reconstruct reliable run timelines.

Yes, agent run logging requires that each event includes a shared agent_run_id and conforms to the defined event shapes, ensuring cross-tool correlation and reliable run reconstruction across your observability dashboards.

The best way to capture memory misses and tool call failures is to use a canonical 10-event observability schema that defines specific events for these outcomes, tied together by a shared agent_run_id for analytics.