Fleet 2.6.0 is out.See what's new →
FleetFleet

Runs and governance

A run is the durable record of one workflow execution: what it received, what each step produced, where it paused, who decided, and how it ended.

Run states

pendingQueued or requeued for a worker.
awaiting_codeA coding agent is working and the run is waiting for its PR.
awaiting_approvalA human must approve or reject the current gate.
doneThe workflow completed successfully.
failedExecution failed or exhausted its configured recovery path.
rejectedA human rejection ended the run.

Terminal runs are immutable. A late worker report or stale approval cannot resurrect a completed run.

Approval gates

When a run reaches an approval step, Fleet stores the documents and findings the approver must see, records the pending gate, and removes the run from the worker queue. The approval inbox links to that exact evidence. Approval requeues the run; rejection records the person, time, and reason before following the workflow's rejection route or ending the run.

Artifacts and provenance

Document outputs are stored by content hash. Action steps also leave durable markers: PR URL and commit for code, decision for approval, shared link for publish, and merged PR for merge.

Lineage

Workflow, step, and upstream inputs.

Agent type

The role bound to the agent-running step.

Rules snapshot

The exact house-rule versions pinned when the run began.

Model

The model default that governed the step when known.

Run history

Each workflow keeps its own history even when renamed because runs reference the workflow's stable ID. Open a run to inspect step results, flags, approvals, route rounds, artifacts, and terminal outcome. Parked or failed runs can appear in the intervention queue when they exceed the workflow's stuck SLA.

Next: build a coding workflow →