backlog-autopilotlisted

# backlog-autopilot The in-session skill that drains a Linear pool into clean PRs. Today it does the **attended, parallel** burst: you are present; you want a few **independent** tickets done **now**, **in parallel**, **without breaking anything**, and handed back as **one PR** to review. Each ticket is worked end-to-end by a **worktree-isolated subagent**; the green branches are reconciled into a single **integration PR** gated by the full suite. > **Consolidation in progress.** `backlog-autopilot` replaces the former `backlog-batch` and > the deleted `autonomous-backlog-loop`. Risk-gated auto-merge of an attended cluster is wired > (`--auto-merge`, see below); cluster auto-detection and the multi-cluster long-run L0 loop are > still being added per `docs/specs/2026-06-21-backlog-autopilot.md` + > `docs/specs/2026-07-01-backlog-autopilot-auto-merge-mvp.md`. The per-ticket quality cycle is now the > dedicated `harness:ticket-runner` skill (single source of truth), which each worker runs. > The attended batch below is the stable core they build on. **This is never a default.** It runs only when a human launches `/harness:backlog-autopilot`. It requires the **Workflow** tool (deterministic multi-agent orchestration) to be available and opt-in in the session — the launcher is the explicit trigger. **Attribution**: see `.source`. --- ## Why this exists Draining a handful of independent tickets in parallel, attended, is its own problem. Doing it naively (N agents editing on