What problem does it solve? Coordinating multi-step work across several specialist agents is error-prone: tasks get executed by the wrong profile, dependencies are lost, and progress vanishes on restart. This Skill turns an orchestrator profile into a disciplined router that decomposes goals into a dependency-linked Kanban task graph persisted in SQLite, instead of doing the work itself. ## Core Features & Use Cases - Task Graph Decomposition: Drafts a task graph (e.g., researchers in parallel, then analyst, then writer), shows it to the user for confirmation, then creates tasks via kanban_create with parents=[...] dependency links that auto-promote children when parents complete. - Route-Don't-Execute Enforcement: Codifies anti-temptation rules so the orchestrator always assigns work to specialist profiles (researcher, analyst, writer, reviewer, backend-eng, ops, pm) rather than implementing it directly. - Large-Scale Refactoring Pattern: Provides a batching playbook for architectural migrations — define the target state first, launch 3-4 independent subtasks in parallel, validate everything in one sweep, update meta-files last. - Stuck Worker Recovery: Documents reclaim, reassign, and model-change workflows for crashed or hallucinating workers, with audit events preserved. - Use Case: Ask whether to migrate to Postgres; the orchestrator queues two parallel researcher tasks, an analyst synthesis gated on both, and a writer memo gated on the analysis — all tracked on the board. ## Quick Start Decompose my goal of evaluating a Postgres migration into Kanban tasks assigned to the right specialist profiles, showing me the task graph before creating anything.
