What problem does it solve? Complex multi-step work often fails when parallel agents collide on shared files, duplicate effort, consume stale decisions, or get cancelled mid-flight. This Skill provides an explicit coordination model that decomposes large tasks into a dependency-aware task and decision graph, assigns one owner per decision domain and write boundary, and integrates evidence-backed results safely. ## Core Features & Use Cases - Hierarchical Scheduling: Build a task and decision graph, appoint subplanners for exclusive decision domains, and dispatch ready work up to safe capacity. - Ownership & Conflict Routing: Enforce one writer per shared surface, route semantic conflicts to decision owners, textual conflicts to merge arbiters, and acceptance conflicts to independent verifiers. - Ephemeral Shared Memory: Materialize an owned, single-writer-per-file run root for cross-agent state, with strict placement, lifecycle, and cleanup rules. - Scheduler Auditing: Diagnose hidden serialization, decision split-brain, stale contracts, and verification gaps using the bundled audit reference. - Use Case: When migrating a large repository across multiple modules, invoke this Skill to freeze shared contracts, dispatch isolated workers per module, preserve healthy long-running workers, and integrate only evidence-backed results. ## Quick Start Ask the assistant to use $multitask-coordinator to plan and execute your large multi-part task with parallel subagents and verified integration.
