Run and host
Control a run
Use a run ID to check progress and read saved output. You can also recover a failed run, follow one active graph step live, or stop the run.
Find a run and follow it
Start with list if you don't know the run ID. Once you have it, use status for a snapshot, watch for state changes, or logs for saved output.
zeroshot list --target cloud
zeroshot status <run-id> --target cloud
zeroshot watch <run-id> --target cloud
zeroshot logs <run-id> --target cloudlist | Shows runs that the selected target still stores. Copy the run ID you need. |
|---|---|
status | Shows one run's fixed details, current state, and active graph steps. |
watch | Prints saved state changes, then waits for new ones. |
logs | Prints saved logs that exclude credentials and provider internals, then waits for new lines. |
View one active step or stop the run
zeroshot attach <run-id> <execution-ref> --target cloud
zeroshot force-stop <run-id> --target cloudstatus gives each active graph step an execution reference. This is a generated identifier, so copy it exactly into attach. Because attach shows only output produced after you connect and accepts no keyboard input, use logs for earlier output.
force-stop records a hard-stop request. Keep checking status or leave watch open until the run reaches finished. Active steps can remain visible while the state is stopping.
Read the run state
queued | Cloud hasn't started the run. It is waiting for a concurrency slot, which Cloud calls a lane, or for Boost capacity. |
|---|---|
admitted | The controller has created the run, but no graph step is active yet. |
running | The run is underway. Status lists any graph steps that are active now. |
stopping | Cloud recorded a hard-stop request. Steps that are still shutting down remain visible. |
finished | No step is active. The result is succeeded(output) or failed(reason). |
queued is a Cloud scheduling state. OECP, the control protocol used by the CLI after Cloud starts the run, covers admitted through finished.
Restart or resume a failed run
Cloud keeps checkpoints for failed, cancelled, or expired attempts for 30 days. A successful attempt removes the lineage's checkpoints.
zeroshot checkpoints <run-id> --target cloud
zeroshot resume <run-id> --target cloud
zeroshot resume <run-id> --from-checkpoint <checkpoint-id> --target cloud
zeroshot discard-workspace <run-id> --target cloudWithout --from-checkpoint, resume restores the latest workspace and starts the graph at its first node. Selecting a checkpoint restores files and prerequisite outputs from before its node or outer parallel/map group, removes later workspace edits, and reruns that boundary. Either form creates a new run ID and keeps the original graph, input, source, and delivery lineage.
Cloud resolves stored provider connections again; provider sessions and prior token usage do not carry over. If private source checkout or Git delivery used a raw GH_TOKEN, export it in the current shell before resuming. Use discard-workspace to remove recovery data you no longer need.
If your terminal disconnects
watch and logs can replay saved records. The client remembers a cursor, which is its position in the run history, and resumes after the last record it received. attach has no cursor because it is live-only.
After a disconnect, run status again. If the step is still active, reconnect with its execution reference; if attach returns GONE, the step has ended and its saved output is available through logs. Losing the connection never stops the run.