What problem does it solve? Hierarchical multi-agent systems waste tokens by replaying the full orchestrator trajectory into every worker call, while summarization adds latency and loses information. This Skill explains how to transfer orchestrator state to workers at the representation level by compacting the worker model's KV cache instead of re-serializing context as text. ## Core Features & Use Cases - Task-guided KV compaction: Score trajectory positions by attention from the current worker task prompt, keeping only relevant tokens. - Shared global mask with MAD thresholding: Aggregate scores across layers and heads into one mask using a median + tau * MAD threshold for batched, low-latency compaction. - Decision framework: Compare Latent Briefing against prefix caching, summarization, and RAG to pick the right cross-agent memory mechanism. - Use Case: In a recursive orchestrator-worker system answering long-document questions, compact the growing orchestrator trajectory before each worker call to cut worker tokens while preserving task-relevant reasoning state. ## Quick Start Ask the assistant to explain how to share orchestrator memory with workers via KV cache compaction using Latent Briefing and Attention Matching.
