What problem does it solve? When acting as an orchestrator in a multi-agent Kanban setup, it is easy to do the work yourself, invent profile names that silently fail, or mislink task dependencies. This Skill provides the decomposition playbook and anti-temptation rules for routing work through a Kanban board instead of executing it directly. ## Core Features & Use Cases - Profile Discovery First: Requires listing actual configured profiles (via hermes profile list or asking the user) before fan-out, since the dispatcher silently drops unknown assignees. - Task Graph Decomposition: Splits multi-lane requests into parallel independent cards and gated child cards using parents=[...] in kanban_create, with fan-out/fan-in, pipeline, and human-in-the-loop patterns. - Goal-Mode and Recovery: Supports goal_mode=True cards with a judge loop for open-ended work, plus reclaim, reassign, and model-change workflows for stuck or hallucinating workers. - 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 agent to decompose your goal into Kanban cards assigned to your configured profiles, running independent lanes in parallel and gating dependent work with parent links.
