Context Management Patterns

<h2>The Layered Context Pattern</h2> <p>Structure your context in layers, from general to specific:</p> <ol> <li><strong>Global layer</strong> - Universal preferences (communication style, formatting)</li> <li><strong>Domain layer</strong> - Industry or field-specific knowledge</li> <li><strong>Project layer</strong> - Project-specific requirements and constraints</li> <li><strong>Task layer</strong> - Immediate task context</li> </ol> <h2>The Progressive Disclosure Pattern</h2> <p>Don't overwhelm the AI with context. Start minimal and add more as needed:</p> <pre><code>Initial prompt: Basic task + essential constraints Follow-up: Add specific requirements as the conversation develops Deep dive: Include detailed specifications when working on complex features</code></pre> <h2>The Reference Pattern</h2> <p>Instead of repeating context, reference established patterns:</p> <pre><code>"Follow our standard error handling pattern as defined in the project memories"</code></pre> <h2>Anti-Patterns to Avoid</h2> <ul> <li><strong>Context dumping</strong> - Including all possible context regardless of relevance</li> <li><strong>Stale context</strong> - Using outdated information that contradicts current reality</li> <li><strong>Conflicting instructions</strong> - Multiple memories that contradict each other</li> </ul>