# Loading Config for Token Optimization & Perf

## Core Config (Always Load)
```yaml
Core:
  Always: [CLAUDE.md, RULES.md, PERSONAS.md, MCP.md]
  Priority: Critical behavioral rules, personas & MCP patterns
  Size: ~4600 tokens
  Reason: Essential for all Claude Code behavior, personas globally available
  
Global Availability:
  PERSONAS.md: All 9 cognitive archetypes available via /persona:
  MCP.md: All MCP patterns available automatically
  
Commands:
  Trigger: /
  Path: .claude/commands/
  Size: ~50 tokens per command
  Cache: Most recent 5 commands
  Index: command names & risk levels only
  
SharedResources:
  LoadWith: Associated commands
  Path: .claude/commands/shared/
  Size: ~150 tokens per YAML
  Examples:
    - cleanup-patterns.yml→loads w/ /cleanup
    - git-workflow.yml→loads w/ git ops
    - planning-mode.yml→loads w/ risky commands
```

## Advanced Loading Optimization
```yaml
Smart Loading Strategies:
  Predictive: Anticipate likely-needed resources based on command patterns
  Contextual: Load resources based on project type and user behavior
  Lazy: Defer loading non-critical resources until explicitly needed
  Incremental: Load minimal first, expand as complexity increases
  
Intelligent Caching:
  Command Frequency: Cache most-used commands permanently
  Workflow Patterns: Preload resources for common command sequences
  User Preferences: Remember and preload user's preferred tools
  Session Context: Keep relevant context across related operations
  
Token Efficiency:
  Base load: 4600 tokens (CLAUDE.md + RULES.md + PERSONAS.md + MCP.md)
  Optimized commands: 4650-4700 tokens (~50 tokens per command)
  Smart shared resources: Load only when needed, avg 150-300 tokens
  Performance gain: ~20-30% reduction through intelligent loading
  Trade-off: Higher base load for consistent global functionality
  
Context Compression:
  Auto UltraCompressed: Enable when context approaches limits
  Selective Detail: Keep summaries, load detail on demand
  Result Caching: Store and reuse expensive analysis results
  Pattern Recognition: Learn and optimize based on usage patterns
```

## Performance Monitoring Integration
```yaml
Loading Metrics:
  Time to Load: Track component loading speed
  Cache Hit Rate: Measure effectiveness of caching strategies
  Memory Usage: Monitor total context size and optimization opportunities
  User Satisfaction: Track command completion success rates
  
Adaptive Optimization:
  Slow Loading: Automatically switch to lighter alternatives
  High Memory: Trigger context compression and cleanup
  Cache Misses: Adjust caching strategy based on usage patterns
  Performance Degradation: Fall back to minimal loading mode
```

## Loading Strategies

```yaml
Loading_Strategies:
  Core_Loading:
    Essential_Files:
      Always_Load: ["CLAUDE.md", "RULES.md", "PERSONAS.md", "MCP.md"]
      Token_Budget: "~4600 tokens base cost"
      Load_Order: "CLAUDE.md → RULES.md → PERSONAS.md → MCP.md"
      Validation: "Ensure all files exist and are valid before proceeding"
      
    Command_Loading:
      Discovery_Method: "Scan .claude/commands/ directory for .md files"
      Index_Creation: "Build lightweight index with command names and flags"
      Lazy_Loading: "Load full command content only when invoked"
      Cache_Strategy: "Keep last 5 used commands in memory"
      
  Progressive_Loading:
    Minimal_Start:
      Initial_Load: "Core files + command index only"
      Token_Cost: "~4650 tokens (base + command index)"
      Expansion_Triggers: "Command invocation, flag usage, complexity detection"
      
    Context_Expansion:
      Shared_Resources: "Load .yml files from shared/ when referenced"
      Pattern_Files: "Load specific pattern files based on command needs"
      MCP_Resources: "Load MCP-specific patterns when MCP flags used"
      Research_Resources: "Load research patterns when external libs detected"
      
  Intelligent_Caching:
    Usage_Patterns:
      Frequency_Based: "Cache most frequently used commands permanently"
      Sequence_Based: "Preload resources for common command chains"
      Context_Based: "Cache resources relevant to current project type"
      Time_Based: "Cache recently used resources for quick access"
      
    Cache_Management:
      Size_Limits: "Maximum cache size to prevent memory bloat"
      Eviction_Policy: "LRU (Least Recently Used) for cache cleanup"
      Preemptive_Loading: "Load likely-needed resources during idle time"
      Cache_Warming: "Pre-populate cache with common patterns on startup"
      
  Context_Optimization:
    Token_Management:
      Budget_Tracking: "Monitor total context token usage continuously"
      Threshold_Alerts: "Warning at 70%, critical at 85% capacity"
      Auto_Compression: "Enable UltraCompressed mode when approaching limits"
      Cleanup_Triggers: "Remove unused resources when space needed"
      
    Smart_Compression:
      Selective_Detail: "Keep summaries, load details on demand"
      Result_Caching: "Store expensive analysis results for reuse"
      Pattern_Recognition: "Learn user patterns to optimize loading"
      Adaptive_Strategies: "Adjust loading based on session characteristics"
      
  Performance_Optimization:
    Loading_Speed:
      Parallel_Loading: "Load independent resources concurrently"
      Streaming_Parse: "Process large files incrementally"
      Incremental_Updates: "Update cache without full reload"
      Background_Refresh: "Update cached resources during idle periods"
      
    Memory_Efficiency:
      Lazy_Initialization: "Create objects only when needed"
      Resource_Pooling: "Reuse common objects across commands"
      Garbage_Collection: "Clean up unused resources periodically"
      Memory_Profiling: "Track memory usage patterns for optimization"
      
  Failure_Handling:
    Resource_Unavailable:
      Missing_Files: "Continue with degraded functionality, warn user"
      Parse_Errors: "Skip corrupted files, use fallback patterns"
      Permission_Issues: "Graceful degradation with user notification"
      Network_Failures: "Use cached versions when available"
      
    Fallback_Strategies:
      Minimal_Mode: "Operate with core files only when resources unavailable"
      Alternative_Sources: "Use backup locations or embedded defaults"
      Graceful_Degradation: "Reduce functionality rather than complete failure"
      Recovery_Attempts: "Retry failed loads with exponential backoff"
```