pueue job queue

Pueue is a daemon-backed shell job queue. The daemon (pueued) accepts tasks via the pueue CLI, runs them across parallelism-capped groups, persists state across reboots, and exposes status / logs as JSON. This skill teaches the agent to drive pueue end-to-end: submit single tasks or whole DAGs, cap parallelism per group, block until completion, retrieve logs, and retry.

The skill is a CLI bridge, not a scheduler. Pueue's --after is AND-only and success-only — that maps cleanly to declarative DAGs and not much beyond. For OR-deps, conditional branching, retry-with-backoff, or distributed scheduling, escalate to a real orchestrator (see "When NOT to use").

When to use

• "Run these 30 commands, max 4 at a time" → pueue group add ml && pueue parallel 4 --group ml then loop submit.sh --group ml.
• "Kick off a long training job and let me close my laptop" → submit.sh -- ./train.sh (pueue persists across reboots).
• "Run task B only after task A finishes successfully" → submit.sh --after $A_ID -- ./b.sh.
• "Fan out 4 trainings, then evaluate" → submit-dag.py dag.yaml.