latent-briefing

An agent skill by muratcankoylan, from muratcankoylan/Agent-Skills-for-Context-Engineering. Tags: automation, developer-tools, multi-agent, performance, strategy.

What it does

This skill should be used when the user asks to "share memory between agents", "KV cache compaction for multi-agent", "orchestrator worker context", "latent briefing", "reduce worker tokens", "cross-agent memory without summarization", or discusses Attention Matching compaction, recursive language models with workers, or token explosion in hierarchical agents.

Install

With the skills CLI, which installs into Claude Code, Codex, Cursor and other agents:

npx skills add muratcankoylan/Agent-Skills-for-Context-Engineering --skill latent-briefing

Or copy the skill folder into Claude Code's skills directory by hand (~/.claude/skills for every project, or .claude/skills inside one):

git clone --depth 1 https://github.com/muratcankoylan/Agent-Skills-for-Context-Engineering
cp -r Agent-Skills-for-Context-Engineering/skills/latent-briefing ~/.claude/skills/latent-briefing

Safety box score

Grade A, minimal risk: 0 of 100 risk points, rated 2026-10-20. The grade rates what the skill and its scripts can reach on the machine of whoever installs it, across eight categories from shell execution to secrets access. How the score works.

This skill is documentation-only: it explains Latent Briefing, a KV-cache compaction technique for multi-agent inference, and ships no executable scripts or installers. The main practical risk is compatibility, because it assumes control over the worker model’s runtime and KV state, so it is not meant for API-only or heterogeneous stacks.

Category levels, from 0 (none) to 3 (high)
Category	Level	Why
Code & shell execution	0 (none)	The shipped files are Markdown prose and equations only; there are no shell commands, scripts, or evaluable code artifacts in `SKILL.md` or `references/attention-matching-formulation.md`.
Network access	0 (none)	The skill only contains static reference links (for example, to arXiv and X) and does not include code that fetches from or sends data to a remote host.
Filesystem reach	0 (none)	There are no instructions to read, write, or delete files on disk; the content stays at the documentation level and does not mention paths outside the repo.
Secrets & credentials	0 (none)	The skill does not ask for API keys, passwords, `.env` files, browser sessions, or any other credential material.
Privilege & persistence	0 (none)	No sudo, global installs, services, shell startup edits, or OS-level settings are requested anywhere in the skill files.
Instruction hijack surface	0 (none)	The text contains normal scope guidance and no prompt-injection patterns such as hiding instructions, skipping confirmations, or following remotely fetched directives.
Irreversible actions	0 (none)	It describes an inference optimization technique, not an action that sends messages, deploys, purchases, publishes, or deletes anything.
Transparency	0 (none)	All shipped files are readable Markdown; there are no binaries, minified blobs, base64 payloads, or decoded-at-runtime components.

Before installing

Check that your worker runtime actually exposes KV tensors for inspection and rewriting.
Confirm your stack is not API-only or otherwise locked to hosted text-generation endpoints.
Read `SKILL.md` and `references/attention-matching-formulation.md` to verify the technique matches your model architecture and task mix.
If you plan to adopt it, benchmark against prefix caching and text-layer handoff on your own workload first.

Rated 2026-10-20 by an automated review of 2 files from the skill's repository.

Source

Repository: muratcankoylan/Agent-Skills-for-Context-Engineering (all skills from this repository)
Path: skills/latent-briefing/SKILL.md
Branch: main
Updated: 2026-09-19

context-compression — This skill should be used when long-running agent sessions need context compression, structured summarization, compaction, token-per-task optimization.
context-degradation — This skill should be used for diagnosing and mitigating context degradation: lost-in-middle failures, context poisoning, context clash, context confusion.
context-optimization — This skill should be used for improving context efficiency: context budgeting, observation masking, prefix or KV-cache strategy, partitioning.
reasoning-trace-optimizer — Debug and optimize AI agents by analyzing reasoning traces, context degradation, tool confusion, instruction drift, repeated task failures.
self-improvement-loops — This skill should be used when the harness, scaffold, workflow, or optimizer itself is the optimization target: recursive self-improvement (RSI) loops.
stitch-loop — Teaches agents to iteratively build websites using Stitch with an autonomous baton-passing loop pattern