Context Window Management

This handles the messy reality of keeping LLM conversations under token limits without losing important context. You get four practical patterns: tiered strategies that switch between full context, summarization, and RAG based on size; serial position optimization that exploits how models weight the beginning and end more heavily; importance-based summarization that preserves critical info over recency; and token budget allocation across system prompts, history, and queries. The patterns are implementation-agnostic but include TypeScript examples showing the decision logic. Most useful when building chat applications or agents that need multi-turn conversations, especially once you hit that point where naive truncation starts breaking things.

Install to Claude Code

npx -y skills add sickn33/antigravity-awesome-skills --skill context-window-management --agent claude-code