context-fundamentals

Context Engineering Fundamentals

Context is the complete state available to a language model at inference time: system instructions, tool definitions, retrieved documents, message history, and tool outputs. Understanding context fundamentals is prerequisite to all other context engineering skills.

When to Use / Not Use

Use when:

• Designing new agent systems or modifying existing architectures
• Onboarding to context engineering concepts
• Reviewing context-related design decisions
• Debugging unexpected agent behavior that may relate to context structure

Do NOT use when:

• Diagnosing context failures or degradation -> use context-degradation
• Compressing or summarizing context -> use context-compression
• KV-cache optimization or context partitioning -> use context-optimization
• File-based context patterns or scratch pads -> use filesystem-context

Decision Tree

What do you need to understand about context?
├── What goes INTO context
│   ├── System instructions / identity -> System Prompts (§Anatomy)
│   ├── Available actions -> Tool Definitions (§Anatomy)
│   ├── External knowledge -> Retrieved Documents (§Anatomy)
│   ├── Conversation so far -> Message History (§Anatomy)
│   └── Action results -> Tool Outputs (§Anatomy)
├── How context is CONSUMED
│   ├── Attention mechanics / budget -> Attention Budget (§Attention)
│   ├── Position encoding limits -> Position Encoding (§Attention)
│   └── Quality vs quantity trade-offs -> Quality vs Quantity (§Quality)
├── How to MANAGE context
│   ├── Load only when needed -> Progressive Disclosure (§Disclosure)
│   ├── Allocate token budget -> Context Budgeting (§Budgeting)
│   └── Mix pre-load + JIT -> Hybrid Strategies (§Hybrid)
└── Not about fundamentals? -> See related skills

The Anatomy of Context

Five components, each with different characteristics:

Component	Persistence	Typical Token Share	Key Risk
System Prompts	Session-long	Low	Wrong altitude: too brittle or too vague
Tool Definitions	Session-long	Medium	Poor descriptions force agent guessing
Retrieved Documents	Dynamic	Medium	Pre-loading creates distraction
Message History	Growing	Medium-High	Dominates context in long sessions
Tool Outputs	Growing	83.9% of total	Verbose outputs consume budget