What problem does it solve? Parallel AI subagents often suffer from context drift, peer chatter deadlocks, interface friction between branches, and unverified self-certified work. This Skill enforces a governance protocol so multi-agent execution converges deterministically. ## Core Features & Use Cases - Role Tier Calibration: Assigns strict roles (Orchestrator, Architect/Seam Auditor, Worker, Detached Auditor) with model-tier guidance and tool restrictions, enforcing that planners never write code. - Worker Isolation & Bounded Tasks: Requires private git worktrees, explicit blast-radius file lists, and machine-verifiable acceptance criteria for every worker contract. - 4-Point Handoff & Convergence Loop: Standardizes worker returns (Changes, Deviations, Concerns, Verification Proof) plus seam audits, green-sweep reconciliation, circuit breakers, and blind detached verification. - Use Case: When splitting a large feature across three parallel coding subagents, use this protocol to dispatch bounded tasks, audit interface seams before merging, and have a fresh auditor verify the merged branch exits 0. ## Quick Start Ask the agent to apply the swarm-governance protocol to decompose your feature into bounded worker tasks with 4-point handoffs and detached verification.
