← ClaudeAtlas

build-queuelisted

Queue N pending phase plans and run them unattended overnight. For each item invoke /plan-expedite then /build-phase via the Skill tool, isolating each in its own worktree. On any halt, park the phase by filing a GitHub issue with halt context and proceed to the next queue item. Polls a kill-switch file between items. Emits a morning summary. Invoke as "/build-queue --queue <path>".
aberson/skill-mesh · ★ 2 · AI & Automation · score 71
Install: claude install-skill aberson/skill-mesh
# Build Queue `/build-queue --queue <path>` runs an unattended overnight orchestration of multiple phase plans. Each queue item gets the full `/plan-expedite` -> `/build-phase` chain. Halts are converted to GitHub issues, not retries — the operator triages in the morning from the summary report. --- ## Constraints Execution model: autonomous, queue-driven, halt-then-proceed (HEAVY). This skill exists so the operator can stage 3-5 pending phase plans, kick off `/build-queue` before bed, and wake up to a triage list. The whole point is that no halt — even a reviewer-gate failure on a code step — pulls the operator out of bed. Therefore: 1. **Halt-then-proceed, not halt-then-retry.** This is the load-bearing design choice. When any queue item halts (via the `/build-phase` halt contract — see [`.claude/rules/code-quality.md` § "Build-phase halt contract"](../../rules/code-quality.md)), this skill does NOT auto-retry, does NOT mutate the plan, does NOT try a different approach. It captures the halt context, files a single GitHub issue on the plan's repo titled `[build-queue parked] <plan name>: <halt class>`, marks the queue item as `PARKED`, and proceeds to the next item. Auto-retry would fight the halt contract (5 legitimate halts, anything else is a defect) and burn tokens on undiagnosable failures. The operator's morning triage IS the retry decision. 2. **Sequential, never parallel.** Queue items run one at a time. Parallel `/build-phase` on different plans has burned to