This skill provides a comprehensive framework for understanding and engineering the state available to language models at inference time. It covers the essential components of context—system prompts, tool definitions, retrieved documents, and message history—while introducing critical concepts like the attention budget constraint and the progressive disclosure principle. By implementing these patterns, developers can move beyond simple prompting to create efficient, high-signal context strategies that prevent model degradation and ensure long-horizon task success.
主要功能
01Guidance on optimal prompt 'altitude' and structural clarity using XML/Markdown
02Strategies for progressive disclosure to manage token constraints
0331,431 GitHub stars
04Comprehensive breakdown of context anatomy and attention mechanics
05Patterns for just-in-time document retrieval and tool output management
06Practical triggers for context compaction and budget monitoring
使用场景
01Designing and architecting new autonomous agent systems
02Optimizing RAG systems to reduce token costs and improve retrieval relevance
03Debugging unexpected agent behavior related to context overflow or degradation