What problem does it solve? Coordinating multi-agent work is error-prone: tasks get bundled incorrectly, dependencies are missed, and cards get assigned to profiles that don't exist, leaving work stuck in the queue forever. This Skill provides a decomposition playbook and anti-temptation rules so an orchestrator agent routes work through a Kanban board instead of doing it itself. ## Core Features & Use Cases - Profile Discovery First: Requires checking available worker profiles (via hermes profile list or asking the user) before fanning out, since the dispatcher silently drops unknown assignees. - Task Graph Decomposition: Splits multi-lane requests into parallel independent cards and gated dependent cards using parents=[...] links in kanban_create. - Goal-Mode Cards: Supports goal_mode=True for open-ended tasks, wrapping workers in a judge-evaluated loop until acceptance criteria are met. - Stuck Worker Recovery: Documents reclaim, reassign, and model-change workflows for crashed or hallucinating workers. - Use Case: A user asks to "research Postgres migration costs and performance, then draft a decision memo." The orchestrator creates two parallel research cards, a synthesis card gated on both, and a memo-drafting card gated on the synthesis. ## Quick Start Ask the agent to decompose your project into Kanban tasks and assign each one to the appropriate worker profile on your setup.
