What problem does it solve? When a goal exceeds what a single agent run can accomplish, work must be decomposed, delegated, supervised, and aggregated across multiple agents without losing context or flooding the user with raw output. This Skill defines the operating rules for that multi-agent fleet: who spawns whom, how results roll up, and when the user is interrupted. ## Core Features & Use Cases - Orchestrator and worker role manuals: Reference guides cover spawning with missions and skill names, the explore-fork-join loop, goal watchers for supervision, and adversarial evaluation at reduce points. - Task-fleet interaction rules: Durable task registry conventions for creation, completion notification, blocker escalation, reassignment, cancellation, and automatic reminder escalation. - Effort and autonomy dials: Per-delegation control over model strength, decomposition depth, evaluation rigor, and how much the human decides, with budget ceilings metered by a usage script. - Use Case: You receive a goal too large for one run. You spawn a dedicated orchestrator that decomposes it, forks workers with named skills, supervises them with idle watchers, gathers results, and delivers one aggregated outcome while only decisions needing authorization reach the user. ## Quick Start Ask the agent to break a large goal into subtasks, spawn workers with named skills for each, and report back only at milestones or blockers.
