What problem does it solve?
This Skill solves the problem of reliably running and controlling background work—both deterministic shell commands and LLM subagent tasks—with durable progress tracking and restart safety.
Core Features & Use Cases
- Unified job queue routing: Routes requests to deterministic shell jobs or LLM subagent jobs depending on whether the task needs a reasoning loop or reproducible command execution.
- Durable, observable execution: Uses a Postgres-native Minions queue so jobs survive gateway restarts and expose structured progress, token accounting, and session transcripts.
- Steerable job lifecycle: Supports pausing, resuming, cancelling, replaying, and steering running agents via inbox messages, including parent-child DAGs with failure policies.
Example use case: When you need to fan out parallel research and then consolidate results, submit an LLM subagent fan-out manifest so children run concurrently and an aggregator runs after completion, while you monitor and steer progress as it happens.
Quick Start
Submit an LLM subagent job for research and synthesis by asking your agent to run: "Research Acme Corp revenue and produce a cited summary, then tell me what key details are missing."
