End a cloud instance

Statuses an instance can transition to as it stops doing work. Terminal for the pod (which exits and releases its quota slot); whether the task can continue is tracked separately. 'skipped' signals that the agent chose not to do work (e.g., a dedupe gate matched) and is distinct from 'terminated' which signals an interruption. 'needs_attention' signals the agent paused on a user-input request (execute_ask_for_help) — the pod stops but the session remains continuable.

The reason for terminating a cloud instance. 'infra_shutdown' covers any shutdown signal from the runtime environment (K8s pod eviction, Cloud Run instance cycling, etc.) — kept generic so it applies regardless of where the agent runs. 'dispatch_failed' means the instance never started — its start message failed to publish, or expired in the queue before any runner claimed it.

The id of the cloud instance

The id of the workspace

The id of the agent session this instance belongs to

The status of the latest agent instance driving a session. The same enum is used for cloud and local instances. Cloud-only values (queued, terminating, rate_limited, skipped) are set by the cloud instance lifecycle (start/terminate/end). Common values (running, needs_attention, completed, failed, terminated) are written by either cloud or local clients. merged is a session-level state applied after a completed authoring task's branch is merged into master (via the branch merge endpoint or the session's auto_merge setting); the underlying cloud instance stays completed. accepted and closed are session-level review outcomes for a finished authoring task, set only through their own endpoints; the underlying cloud instance keeps its own status. accepted records that the user kept the authored test. closed records that the user discarded it — the authored test and its steps are deleted while the session's conversation is retained. Both are terminal in the same sense as merged — neither is resumable, and no lifecycle transition leaves them. resuming is a transient, server-set-only state on the session (no instance holds it) — the cloud TAA continuation flow flips a resumable session to resuming while it plans the answer, then to queued when the new instance spawns (or back to a resumable status on re-clarification, or failed on error). It is the concurrency guard, so a second answer to a resuming session is rejected. The session's instance_type field indicates which kind of instance owns the session. Use none in query parameters to match sessions without any status.

The platform the agent operates on

The reason for terminating a cloud instance. 'infra_shutdown' covers any shutdown signal from the runtime environment (K8s pod eviction, Cloud Run instance cycling, etc.) — kept generic so it applies regardless of where the agent runs. 'dispatch_failed' means the instance never started — its start message failed to publish, or expired in the queue before any runner claimed it.

The timestamp when the cloud instance entered the queued status in epoch milliseconds

The timestamp when the cloud instance started running in epoch milliseconds

The timestamp when the cloud instance entered the terminating status in epoch milliseconds

The timestamp when the cloud instance entered a terminal status (completed, failed, or terminated) in epoch milliseconds

The timestamp of the cloud instance creation in epoch milliseconds

The id of the user who created the cloud instance

The timestamp of the cloud instance last update in epoch milliseconds

The id of the user who last updated the cloud instance