multi-agent-long-task-schedulinglisted

# Skill: Multi-Agent Long-Task Scheduling ## Purpose Use this skill to make long, complex, or interruption-prone projects recoverable, verifiable, mergeable, and deliverable. MALTS is single-agent first. Multi-agent work is a controlled division-of-work mechanism enabled only when needed; every real sub-agent dispatch requires an explicit launch review first. This skill does not exist to spawn more agents. It exists to reduce loss of control. ## Trigger Use this skill only when one of these is true: - The user explicitly enables MALTS, long-task mode, or multi-agent long-task scheduling. - The task is multi-stage, multi-file, or likely to exceed a comfortable single-session context. - Independent exploration, implementation, or verification can run in parallel without file conflicts. - The project needs recoverable state, batch execution, growth/memory records, and acceptance-ready delivery. Do not use this skill when: - The task is small enough for the main controller to finish directly. - Requirements are unclear. - The verification method is unclear. - Subtask boundaries cannot be defined. - The scheduling cost is higher than the likely benefit. ## Multi-Agent Fit Assessment Before suggesting or enabling multi-agent scheduling, classify the task by type and difficulty. Multi-agent exists to stabilize long work, reduce loss of control, and improve independent verification; it does not exist to maximize agent count. | Level | Task Shape | Default Action | |---|---