What problem does it solve? Draining a large backlog with parallel AI agents silently degrades without discipline: agents self-merge into race conditions, ship confident no-ops, and drop work on timeouts. This Skill provides a repeatable nine-phase protocol that keeps a fleet of delegated agents honest from triage through integration. ## Core Features & Use Cases - Nine-phase orchestration: Triage, ground, advisor review, clustering, wave spawning, serialized integration, run-to-exhaustion, retro-of-retros, and documented closeout. - Worktree isolation with claims ledger: Each agent works in its own git worktree branched from a pinned base commit, with ownership recorded in a claims file to prevent domain collisions. - Adversarial verification and review quorum: The orchestrator re-runs each branch's gates, scales reviewer count to blast radius, and never lets agents self-merge or self-approve. - Use Case: You have twenty open plans and audit findings in a repo. Invoke the protocol to cluster independent items, spawn a wave of agents that each plan and implement in a worktree, then merge their branches one at a time with verification between each merge. ## Quick Start Ask the AI to drain the backlog with agents using the agent protocol, triaging open plans and running one wave of worktree-based agents with serialized integration.
