What is context engineering?

Context engineering is the process of curating the smallest possible set of high-signal tokens for an LLM task to maximize reasoning quality while minimizing costs.

How does this skill help reduce AI costs?

It provides strategies for context compaction and caching that target a 50-70% reduction in token usage with less than 5% loss in output quality.

When should I implement compaction triggers?

According to the skill guidelines, you should start warning at 70% token utilization and trigger active optimization or compaction at 80%.

What are the core principles of this skill?

The core principles include favoring context quality over quantity, using progressive disclosure, isolating work across sub-agents to prevent degradation, and measuring performance before optimizing.

Context Engineering

Name: Context Engineering
Author: piotroq

bypiotroq

0•

数据科学与机器学习

Optimizes LLM performance by curating high-signal context windows and managing token efficiency for complex AI agent systems.

The Context Engineering skill provides a comprehensive framework for mastering context management within AI agent architectures. It focuses on the principle of 'quality over quantity,' helping developers maximize reasoning accuracy while minimizing token overhead through techniques like compaction, progressive disclosure, and isolation. Whether you are debugging 'lost-in-the-middle' performance degradation or building sophisticated multi-agent pipelines, this skill provides the benchmarks, patterns, and evaluation metrics needed to build cost-effective and high-performing LLM applications.

主要功能

01LLM-as-Judge evaluation frameworks for measuring agent performance

02Multi-agent coordination patterns for effective context isolation

030 GitHub stars

04Memory system architectures including knowledge graphs and persistence

05Token usage optimization and compaction strategies to reduce overhead by 50-70%

06Automated context health analysis and degradation detection scripts

使用场景

01Building complex multi-agent systems requiring precise information partitioning

02Reducing API latency and costs by optimizing token-per-task efficiency

03Debugging agent failures caused by context limits or token poisoning

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add piotroq/adwokat-trzebnica-html-to-php-com context-engineering

For use in Claude.ai and ChatGPT

Download Skill