What problem does it solve? Shipping a groomed backlog by hand means repeatedly checking which PRs are green, which issues are free to start, and how many agents are already running. This Skill automates that loop: each invocation reconciles in-flight PRs, computes remaining capacity, and dispatches new work only from issues marked Ready. ## Core Features & Use Cases - Idempotent tick loop: Each run reconciles in-flight PRs (merge-ready, failed, conflicting, blocked) against live GitHub state, then tops the pipeline up to the configured max_in_flight limit. - Ready-only selection with filters: Pulls exclusively from Ready issues, skipping claimed, blocked, site-mismatched, dependency-blocked, and file-collision candidates, with stack-aware exemptions. - Failure and recovery accounting: Tracks a single shared recovery allowance per PR via authenticated issue comments, demotes repeat failures and conflicting PRs to blocked, and never parks failures back in Ready. - Use Case: Run it under /loop 5m to continuously drain a repo's Ready queue — it merges green PRs via pr-shepherd, dispatches one worktree sub-agent per new issue, and stops at a terminal state when the queue empties or stalls. ## Quick Start Ask the agent to work the ready queue by saying "dispatch ready items in the backlog" after configuring the repo with setup-config.
