context-budget

# Context Budget Management

Strategies for managing limited context windows efficiently using progressive disclosure and tiered loading.

## When to Use

- Conversation approaching context limits
- Working with large codebases or documentation
- Multi-turn interactions with accumulating context
- Need to balance breadth vs. depth of information
- Optimizing API costs (tokens = money)

## When NOT to Use

- Short, single-turn interactions
- When full context fits comfortably
- Simple Q&A without state accumulation

---

## Quick Start

1. **Assess** current context usage and remaining budget
2. **Tier** information by relevance (core → domain → task → optional)
3. **Summarize** or evict low-priority context
4. **Load** new information only when needed
5. **Monitor** and rebalance as conversation evolves

---

## Core Principle: Progressive Disclosure

**Load information only when needed, in order of relevance.**

```
┌─────────────────────────────────────────────────────────┐
│  ALWAYS LOADED (~500 tokens)                            │
│  • Core identity/constraints                            │
│  • Critical safety rules                                │
│  • Navigation hints to deeper content                   │
└─────────────────────────────────────────────────────────┘
         ↓ Load on demand
┌─────────────────────────────────────────────────────────┐
│  DOMAIN CONTEXT (~2K tokens)                            │
│  • Project-specific rules                               │
│  • Architecture patterns                                │
│  • Key file locations                                   │
└─────────────────────────────────────────────────────────┘
         ↓ Load when task requires
┌─────────────────────────────────────────────────────────┐
│  TASK CONTEXT (~4K tokens)                              │
│  • Specific file contents                               │
│  • API documentation                                    │
│  • Implementation details