Skip to main content
Run an agent and stream execution progress story by story.

Options

required
The agent to run, as a handle, UUID, or FQN.
default:"the agent's deployment stage"
dry_run, pilot, or live. Omit it and the backend resolves the mode from the agent’s current stage — a plain run on a pilot-staged agent runs as pilot, not as an implicit dry run.
default:"{}"
Input data as a JSON string.
Show an output preview for each step as it completes.
A key used to deduplicate. Re-sending the same key won’t start a second run.
Emit the result as JSON.

Following a run

Progress streams live as the run executes:
The stream carries run_started, story_started, story_completed, story_failed, run_completed, and run_failed events. It times out after 300 seconds; the run itself keeps going, and you can rejoin with jstm agent run show. When the run ends, the CLI prints the run id and how to review it.

When a run needs input

If the agent hits a point where it needs something from a person, it pauses and raises an input request rather than guessing. The CLI prints the exact commands to unblock it:
See jstm input-requests.

jstm agent run show

Show a run’s status, stories, output, rejection reason, and provenance.
The second form is a sequence address: <agent>/runs/N, where N counts from 1 per agent. See Addressing.
Emit the raw payload as JSON.
Provenance tells you exactly which spec version, agent version, package version, and deployment produced this run.

Listing runs

Passing a run id instead of an agent id shows that run’s full story detail.

Listing stories

See also