What problem does it solve? Multi-agent setups fail when an orchestrator tries to do the work itself, invents profile names that don't exist, or creates dependent tasks as independent cards. This Skill provides the decomposition playbook and anti-temptation rules for an orchestrator profile that routes work through a Kanban board instead of executing it. ## Core Features & Use Cases - Profile discovery before fan-out: Run hermes profile list or ask the user to learn which worker profiles actually exist, since the dispatcher silently drops unknown assignees. - Task graph decomposition: Extract independent lanes from a request, create parallel cards with no parents, and gate synthesis/review cards behind parents=[...] so children auto-promote from todo to ready only when parents finish. - Stuck worker recovery: Reclaim, reassign, or change the model on tasks flagged with a warning badge when a worker crashes or hallucinates card ids. - Use Case: A user asks to "research Postgres migration costs and performance, then draft a CTO memo." The orchestrator creates two parallel research cards, a synthesis card gated on both, and a memo-drafting card gated on the synthesis, each assigned to real profiles on the machine. ## Quick Start Ask the orchestrator to break down your goal into Kanban cards assigned to your available worker profiles and report the resulting task graph.
