What problem does it solve? Working through a backlog of open GitHub issues one at a time is slow, and naive parallelization causes agents to collide on the same files, double-claim issues, or conflict on shared secrets. This Skill orchestrates a race-free fan-out: triage, claim, spawn, land, and report, with every issue claimed before any agent starts. ## Core Features & Use Cases - Race-free claiming: Claims every issue under the orchestrator's own worktree name before spawning, with stale-claim reaping and deferral handling via findings.py. - Conflict-aware batching: Groups issues so no two agents touch the same Ansible role, at most one batch rotates SOPS secrets, and broad changes run alone. - Placement and lifecycle management: Launches headless Opus agents across hosts based on memory headroom, polls status, reconciles landed or no-report batches, and cleans up worktrees. - Use Case: You have a dozen open findings on your homelab repo. Run the fan-out to triage them into safe batches, dispatch one Opus agent per batch in its own worktree, and collect a table of issue-to-PR results. ## Quick Start Ask the agent to fan out on the open GitHub issues, triage them into batches, and dispatch one Opus agent per batch with its own worktree and PR.
