What problem does it solve? Parallel agent fan-outs can silently exceed their intended budget: launchers inherit expensive parent models, helpers forward every worker transcript to premium reviewers, and repeated turns reprocess large contexts. This Skill provides a discipline for planning, bounding, and auditing the real cost of multi-agent investigations. ## Core Features & Use Cases - Deliberate model and scale selection: Verify launcher model inheritance, state model/tier, effort, and agent count ceilings before launch, and match model strength to task uncertainty. - Consultation multiplication control: Detect helpers that forward transcripts to stronger models, aggregate shard evidence before commissioning shared reviews, and cap nested agent work. - Context and checkpoint management: Choose between resuming large histories versus spawning focused helpers, and have workers save independently usable partial results with atomic writes. - Outcome auditing: Compare planned versus actual model, width, and cost; verify saved evidence; and avoid converting bounded searches into false claims of impossibility. - Use Case: A coordinator launches extraction workers that each invoke a premium reviewer, blowing the budget despite correct output. Use this Skill to verify launcher inheritance, cap nested consultations, and review combined evidence instead. ## Quick Start Ask the agent to plan a parallel investigation with an explicit model tier, agent count ceiling, and checkpoint policy before launching any workers.
