overnight-queue

overnight-queue

Queue three real jobs before you leave, come back to a triage table: two verified with independent test/critic evidence, one parked at its human gate waiting for your go/no-go - nothing hung, nothing needed you at 3am.

What it demonstrates: the workflow a bare agent CLI structurally cannot offer. looprail queue runs items sequentially and unattended; goal-only items reuse one loopfile's graph shape (looprail.yaml: worker + real tester + anti-gaming critic) with per-item goals, and release-check.yaml ends at a role: gate whose gate_timeout PARKS the run - resumable, zero repeated work - instead of blocking the queue overnight. Every verified item carries a journal: what ran, what the tester executed, what the critic checked.

Run it:

cp examples/overnight-queue/*.yaml .
looprail queue          # reads ./queue.yaml; add --gate-timeout <sec> to tune parking

In the morning:

# the triage table printed at the end shows parked items' exact resume command
looprail resume <parked-run-id>   # the gate asks again; prior work is cached
looprail ui --all                 # or answer it from mission control

Adapt it: replace the two goal: items in queue.yaml with your own backlog (each needs testable success criteria - the tester and critic hold the line while you sleep), and point looprail.yaml's run: npm test at your stack's real test command. Add as many items as you like; budget rails (max_cost_usd, max_wall_minutes) cap each item independently.