Task Event Types

Every type a task event record (nfltr.task_event.v1: nfltr orch task events --json) can carry. Generated from the code that declares them.

Type	Meaning
`artifact.egress.allowed`	is recorded when the egress policy allowed an artifact to leave to the worker.
`artifact.egress.denied`	is recorded when the egress policy refused to send an artifact to the worker.
`artifact.offered`	is recorded when the worker offers artifacts it produced.
`artifact.requested`	is recorded when the worker asks for artifacts it needs.
`artifact.transfer.completed`	is recorded when an artifact transfer finishes with all its bytes moved.
`artifact.transfer.failed`	is recorded when an artifact transfer fails.
`artifact.transfer.governed`	is recorded with the decision on a governed artifact transfer.
`artifact.transfer.outcome`	is recorded with the outcome of a governed artifact transfer.
`artifact.transfer.progress`	marks each quarter of an artifact's bytes moved.
`artifact.transfer.retried`	marks a new attempt for an artifact after an earlier one broke off; its resume_at_byte/bytes show where it resumed.
`artifact.transfer.started`	marks the first attempt to move an artifact.
`governance.decision`	is recorded when a governance policy or dispatch hook decides on a task start (allow, require approval or deny).
`result.manifest.signed`	is recorded when a signed result manifest is created for the task.
`result.manifest.verified`	is recorded with the outcome of verifying a task's result manifest.
`task.accepted`	is recorded when the worker accepts the task, or when a recovered worker snapshot shows it accepted or running.
`task.cancelled`	is recorded when the task is cancelled. It is terminal.
`task.completed`	is recorded when the task finishes with a result. It is terminal.
`task.dispatched`	is recorded when the task assignment is sent to the worker (or queued for dispatch).
`task.duplicate_terminal`	is evidence that the task store rejected a second terminal transition for an already-terminal task; the stored status is unchanged.
`task.execution_stop_pending`	is a relay-owned extension event. It is deliberately distinct from worker progress: replaying it must retain the transport-disconnect marker while execution ownership is reconciled.
`task.failed`	is recorded when the task fails; the payload is the TaskError. It is terminal.
`task.handoff_requested`	is recorded when the worker asks to hand work to another worker.
`task.input.sent`	is recorded when follow-up input (an answer or message) is sent to the worker.
`task.integration.recorded`	is recorded when the task's result carries an integration record, such as a landed change.
`task.loop.decision`	is recorded when a loop decision (accept, rework, reassign, escalate, stop) is reported for the task.
`task.progress`	is recorded for each progress report from the worker, setup step or watchdog (including the hard-timeout notice).
`task.question`	is recorded when the worker asks a follow-up question and waits for an answer.
`task.rate_limited`	is recorded instead of task.failed when the failure is a rate-limit notice. It is terminal.
`task.reclaimed`	is recorded when a task that had been marked failed or disconnected is reopened by later worker evidence, such as a late result, progress or replay.
`task.redispatched`	is recorded when auto-resume re-dispatches a task the worker no longer knew.
`task.remote_call`	is recorded when a remote call made for the task is audited.
`task.remote_call.policy`	is recorded with the policy decision on a remote call made for the task.
`task.rework.requested`	is recorded when the worker or a verifier asks for the work to be redone, with instructions.
`task.setup_approval.decision`	is recorded when an approval decision on a worker setup step is applied.
`task.steer`	is recorded when an operator steers a running task with a message.
`task.stream.failed`	is recorded when the task stream to the worker failed or closed before the task finished.
`task.superseded`	records that the task's owner (a hub) moved past this attempt (SupersedeTask): it is never revived, resumed or redispatched again, and its late output is recorded as task.duplicate_terminal only.
`task.superseded_stopped`	records that the worker which ran a superseded attempt acknowledged the owner's stop for it (StopSupersededTask): it no longer runs there. The task's state is unchanged.
`task.transport_disconnected`	is recorded when the worker transport drops while the task is open. A retryable disconnect is self-healing.
`task.transport_recovered`	is recorded when the worker transport comes back and the task resumes.
`task.verification.recorded`	is recorded when a verification of the task's work is reported, with its outcome.
`task.worker_poll_unreachable`	is recorded when a proactive status poll cannot reach the worker holding the task.

A few older entries, kept so earlier task histories still read, are not listed.