Context Engineering & Memory Systems

Purpose

Optimize AI agent context management and build persistent memory systems.

Context Engineering

Principles

1. Relevance: Include only contextually relevant information
2. Hierarchy: Organize context by importance
3. Freshness: Prioritize recent information
4. Compression: Summarize long contexts

Techniques

Context Window Management

• Sliding window for conversations
• Summarization for long documents
• Chunking for large codebases
• Priority scoring for relevance

Context Injection

• System prompts
• Few-shot examples
• Retrieved documents
• Conversation history

Memory Systems

Short-Term Memory

• Current conversation context
• Active task information
• Recent actions and results

Long-Term Memory

• Persistent knowledge storage
• Learned patterns and solutions
• Historical decisions

Implementation

File-Based Memory

.opencode/memory/
├── cmz-knowledge.md      # Solutions and patterns
├── cmz-issues.md         # Known issues and resolutions
└── cmz-evolution.md      # Change history

Vector Memory (Future)

• Embeddings for semantic search
• Similarity matching
• Automatic retrieval

Context Engineering Patterns

Pattern 1: Progressive Disclosure

Start with minimal context, expand as needed.

Pattern 2: Selective Retrieval

Retrieve only relevant information from memory.

Pattern 3: Hierarchical Summarization

Maintain summaries at different levels of detail.

Pattern 4: Temporal Weighting

Weight recent information more heavily.

Best Practices

1. Regular Cleanup: Remove outdated context
2. Explicit Storage: Store important learnings explicitly
3. Retrieval Testing: Test memory retrieval accuracy
4. Context Monitoring: Monitor context size and relevance

Integration

Works with CMZ self-learning protocol for continuous improvement.

Usage

Apply when:

• Managing large codebases
• Long-running conversations
• Building agent memory
• Optimizing context usage