Follow and recover a run

Inspect progress and delivery, stop work, or restore a failed attempt.

Follow progress

Open Runs and select a run. Inspect its source revision, preparation logs, graph activity, and confirmed delivery link. You see your own runs; owners and admins can also see GitHub App and organization-delegation runs.

Queued work is waiting for capacity. Preparing environment runs setup and startup before graph execution. Read a failure's recorded reason before deciding whether to retry.

CLI inspection
zeroshot list --target platform
zeroshot status RUN_ID --target platform
zeroshot logs RUN_ID --target platform

Stop or disconnect

Choose Stop or run zeroshot force-stop RUN_ID --target platform. Wait until shutdown and cleanup finish. Closing the browser, CLI, or MCP client only disconnects your view; it leaves the remote run running.

Restart or resume

Saved recovery data remains available for 30 days after failure, cancellation, or expiry. Check status for recovery availability; an early failure may have no snapshot to restore. Success retires the lineage's recovery data. Recovery creates a new run ID with the original graph, input, source, environment, host policy, and delivery lineage.

CLI recovery
zeroshot checkpoints RUN_ID --target platform
zeroshot resume RUN_ID --target platform
zeroshot resume RUN_ID --from-checkpoint CHECKPOINT_ID --target platform

Resume without a checkpoint restores the latest workspace and restarts where that run began, with the conversations it began with. A submitted run starts at the graph root with new conversations; a successor keeps its original starting point. Selecting a checkpoint restores its files, execution state, and saved agent conversations, removes later edits, and reruns that boundary. Conversations scoped to one execution start new. If conversation saving failed, the affected agents start new conversations. Services and Docker state are not restored. Submit a new run to start from the graph root. Setup and startup run again; stored credentials and their destinations resolve afresh. Submit a new run to use an edited environment.

Workspace recovery cannot undo external side effects. Inspect earlier delivery before repeating work. Use discard-workspace only to remove recovery data you no longer need.

Reconnect to history

CLI and MCP clients can continue with the saved run ID and cursors. An unavailable or incomplete history is not proof of success; check status for the outcome. The Zeroshot run-control reference covers live attach, cursor replay, and recovery commands.