Context Window Management

This covers the practical stuff you need when your LLM app starts bumping into token limits or losing track of what matters in long conversations. It's about strategies like tiered context (different approaches based on size), smart summarization that prioritizes importance over recency, and avoiding the lost-in-the-middle problem where models miss info buried in long contexts. The anti-patterns call out naive truncation and one-size-fits-all approaches, which is helpful. If you're building anything with conversation memory or multi-turn dialogues, you'll hit these problems eventually. Pairs well with RAG and prompt caching skills.

Install to Claude Code

npx -y skills add davila7/claude-code-templates --skill context-window-management --agent claude-code