update_job replaces configuration, rather than patching one field. Always call get_job first. Build a write object containing jobId, name, slug, promptTemplate, executionTargetId, runNameTemplate, concurrencyLimit, model, optional effort, and the full intended triggers array. Preserve each existing trigger’s id, slug, kind, enabled, and writable config. Strip read-only response fields (jobId inside triggers, timestamps, reflection, secretConfigured and destination). Do not pass the entire read response unchanged.
Supply expectedUpdatedAt from the fetched job’s updatedAt to protect against concurrent edits. On stale_job, fetch again and reapply the desired changes. Omitting it retains last-writer-wins behavior. Omitted optional configuration resets to defaults; omitted triggers remove the configured automatic triggers. Preserve all triggers you want to keep.
After saving, call get_job and compare prompt, target and triggers. enable_job and disable_job change enabled state without rebuilding configuration. Disabling automation does not stop an existing run. Job saves can trigger edited lifecycle subscribers. Deleting a job removes its execution history/artifacts; inspect and retain anything needed first.