latent-briefing

Name: latent-briefing
Availability: InStock
Author: guanyang

Compact orchestrator trajectories into worker KV caches via attention matching.

947|163|Updated Jan 13, 2026

Search Tags:#multi-agent #token-efficiency #state-sharing #kv-cache #attention-matching #latent-briefing #orchestrator-worker

Ready to run

npx skills add https://github.com/guanyang/open-agent-hub --skill latent-briefing-guanyang

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: latent-briefing
Source: https://github.com/guanyang/open-agent-hub/tree/main/skills/latent-briefing
Command: npx skills add https://github.com/guanyang/open-agent-hub --skill latent-briefing-guanyang

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

Latent Briefing provides representation-level sharing of orchestrator state in multi-agent systems, reducing token costs by compacting the central trajectory into the worker's KV cache rather than replaying full text.

Core Features & Use Cases

Representation-level sharing: share task-specific latent state between orchestrator and workers without text replay.
Attention Matching-based compaction: uses a task-guided query and a shared global mask to reduce KV cache size while preserving attention relevance.
Use Case: in hierarchical agent architectures, enable efficient handoffs for long-running reasoning without incurring large token overhead.

Quick Start

Enable latent briefing by providing the orchestrator with the current worker task prompt to generate a compact KV representation for the next worker call.

Frequently Asked Questions about latent-briefing

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I reduce token usage in hierarchical multi-agent systems?▼

Multi-agent token usage is reduced by compacting the orchestrator trajectory into a smaller KV cache. This representation-level state sharing replaces replaying full text when delegating tasks to workers.

What is attention matching based KV cache compaction for multi-agent workflows?▼

Attention matching KV cache compaction is a technique that uses a task-guided query and a shared global mask to reduce cache size. It preserves attention relevance while sharing latent state between orchestrator and workers.

How do I share orchestrator state with worker agents without replaying full text?▼

You can share orchestrator state by providing the orchestrator with the worker task prompt to generate a compact KV representation. This enables efficient handoffs in long-running reasoning tasks without incurring large token overhead.

Do I need access to worker prompts to enable latent state sharing in orchestrator-worker architectures?▼

Yes, enabling latent state sharing requires access to worker prompts, the orchestrator trajectory, a task-guided query, and a shared mask. You must also have the ability to perform attention-based KV compaction.

When should I use KV cache compaction instead of full text replay for agent handoffs?▼

KV cache compaction is ideal for hierarchical agent architectures where token costs scale with context length. It provides efficient handoffs for long-running reasoning tasks without the overhead of replaying full text.

What are the limitations of using attention-based KV compaction for cross-agent state sharing?▼

Attention-based KV compaction requires specific components like a shared mask and task-guided query to preserve attention relevance. It is strictly applied in hierarchical multi-agent tasks and cannot function without worker prompt access.