kanban-orchestrator

Kanban task decomposition converts user prompts into a task graph, spawning independent lanes for parallel execution while respecting parent-child links and dependencies. It discovers available profiles, sketches the execution order, and assigns work to specialized agents without manual bottlenecks.

Kanban-driven orchestration coordinates complex, multi-lane work by routing decomposed tasks to the right profiles instead of executing them inline. It enforces a route-don't-execute discipline, creating explicit Kanban tasks with parent links, goal_mode support, and lifecycle guarantees across research, implementation, and synthesis phases.

Anti-temptation rules enforce route-don't-execute discipline by creating tasks and preventing premature work without explicit ownership. The orchestrator assigns work to discovered profiles via routing, ensuring no agent begins execution until the Kanban engine grants ownership and dependencies are satisfied.

Recovery and resilience tooling provides reclaim and reassign flows to recover stuck workers in Kanban orchestration. Goal-mode task loops handle long-running prompts, while the engine identifies misassigned or over-linked tasks and reassigns them to available profiles to keep work flowing.

Kanban orchestration supports human-in-the-loop review by gating synthesis tasks that depend on parallel fan-out lanes. Parent-child links ensure upstream work completes and is reviewed before downstream synthesis begins, providing structured lifecycle management across research, implementation, and synthesis phases.

Common pitfalls in Kanban task orchestration include over-linking tasks with unnecessary dependencies, misassigning work to wrong profiles, and misconfiguring frontmatter metadata needed for routing. The framework highlights these misconfigurations to ensure tasks are properly linked and assigned to the correct profiles.