What problem does it solve? Designing and debugging LLM agent systems requires understanding how context windows, attention budgets, and token costs affect model behavior, yet these fundamentals are often scattered or misunderstood. This Skill provides a structured foundation covering context anatomy, attention mechanics, and context budgeting so engineers can make informed architecture decisions. ## Core Features & Use Cases - Context Anatomy Reference: Breaks down system prompts, tool definitions, retrieved documents, message history, and tool outputs with guidance on each component's role and cost. - Attention & Budgeting Guidance: Explains the n-squared attention constraint, position encoding degradation, and why smaller high-signal context outperforms large low-signal context. - Progressive Disclosure Patterns: Shows how to defer loading of skills, documents, and tool results until needed, including filesystem-based and hybrid pre-loading strategies. - Use Case: When an agent's responses degrade on long tasks, use this Skill to diagnose whether tool outputs are dominating the context window and apply compaction triggers at 70-80% utilization. ## Quick Start Ask the agent to explain how context windows and attention budgets affect your agent design and recommend a context budgeting strategy.
