Context Engineering

A practical guide for keeping LLM agents from choking on their own context windows. You get concrete strategies for the four main escape hatches: write externally, select what matters, compress aggressively, or isolate across sub-agents. The metrics are helpful (start compacting at 70% token utilization, aim for 50-70% reduction). What's refreshing is the focus on token efficiency as a first-class concern rather than an afterthought. The anti-patterns section alone will save you from common mistakes like burying critical information in the middle where attention drops off. Best used when you're building agent systems that need to stay coherent across long sessions or coordinate multiple agents without blowing your token budget.