minion-orchestrator

You can run resilient background jobs by submitting them to a Postgres-backed durable queue that supports pausing, resuming, canceling, and replaying. This ensures long-running work remains observable and restart-safe while allowing mid-flight redirection for running agents.

Yes, unified job routing allows you to orchestrate both deterministic shell commands and LLM subagent jobs through a single durable queue interface. This eliminates the need to choose between scripts and agents upfront when submitting background work.

Executing parallel fan-out tasks with parent-child DAGs involves submitting a parent job that spawns child task graphs with configurable failure handling. The queue rolls up progress and tokens across the DAG, enabling consolidated results from complex workflows.

You monitor and control a running background job by using lifecycle operations including get_job, list_jobs, pause_job, resume_job, and cancel_job. These mutating operations provide full visibility and steering capability over active execution.

Yes, durable job orchestration for long-running work requires a Postgres-backed queue interface to ensure persistence. This backing store enables restart-safe execution, progress tracking, and reliable message passing for steerable background tasks.

To replay a failed job or send messages to a running LLM subagent, use the replay_job and send_job_message operations provided by the orchestration interface. These allow you to retry failed work and steer active agents without restarting the entire workflow.