What problem does it solve? Long-running background work dies silently with the session, and completion events get swallowed, leaving users staring at silence while finished results sit unreported. This Skill routes shell jobs and LLM subagents through a durable, observable, steerable Postgres-backed queue so work survives restarts and results provably reach the user. ## Core Features & Use Cases - Unified job orchestration: Submit deterministic shell jobs (gbrain jobs submit shell) and LLM subagent jobs (gbrain agent run) through one queue with structured progress, token accounting, and transcripts. - Steerable lifecycle: Pause, resume, cancel, replay, and send mid-flight steering messages to running agents via MCP-callable tools. - Durable execution doctrine: A capability ladder (Minion job + deadman, cron-checked progress file, foreground buffering) plus content-addressed stage checkpoints for expensive multi-stage pipelines. - Use Case: You need to run gbrain extract all, a 40-minute operation. Submit it as a Minion job with a one-hour timeout, arm a deadman check in the same action block, and get a recovery report even if the completion event is swallowed. ## Quick Start Ask the agent to submit your long-running command as a gbrain job and arm a deadman so you are notified even if the completion event is lost.
