@nockdev/awf 6.2.0

package/.agent/core/CONTEXT_OPTIMIZATION.yaml

@@ -0,0 +1,286 @@
+# =============================================================================
+# DOMYH Awesome Code
+# Context Optimization Engine - Caching, Compaction, JIT Loading
+# =============================================================================
+
+version: "6.1.2"
+
+# =============================================================================
+# OVERVIEW
+# Intelligent context management for token efficiency
+# =============================================================================
+
+description: |
+  Context Optimization Engine manages the agent's context window through:
+  1. Prompt Caching - Cache repeated prompts for cost savings
+  2. Context Compaction - Compress history to fit more information
+  3. JIT Loading - Load information only when needed
+  4. Priority Ranking - Keep the most relevant information
+
+# =============================================================================
+# PROMPT CACHING
+# Cache repeated prompts for efficiency
+# =============================================================================
+
+prompt_caching:
+  enabled: true
+
+  levels:
+    # System prompt (permanent cache)
+    system:
+      cache: true
+      ttl: "permanent"
+      priority: 1
+      description: "Core system prompt and instructions"
+
+    # Skill definitions (session cache)
+    skills:
+      cache: true
+      ttl: "session"
+      priority: 2
+      description: "Loaded skill definitions"
+      max_cached: 10
+
+    # Memory context (short-term cache)
+    memory:
+      cache: true
+      ttl: "5 minutes"
+      priority: 3
+      description: "Memory state and context"
+
+    # Tool results (ephemeral)
+    tool_results:
+      cache: true
+      ttl: "current_turn"
+      priority: 4
+      description: "Results from tool executions"
+
+  storage:
+    type: "in-memory"
+    max_size_mb: 50
+    eviction: "LRU"
+
+  statistics:
+    track: true
+    metrics:
+      - "cache_hit_rate"
+      - "tokens_saved"
+      - "cost_savings"
+
+# =============================================================================
+# CONTEXT COMPACTION
+# Compress context to maximize useful information
+# =============================================================================
+
+compaction:
+  enabled: true
+
+  strategies:
+    # Conversation summarization
+    conversation_summary:
+      trigger: "message_count > 10"
+      action: "summarize_old_messages"
+      keep_recent: 3
+      summary_max_tokens: 500
+
+      template: |
+        ## Conversation Summary
+        - Messages: {message_count}
+        - Key topics: {topics}
+        - Decisions made: {decisions}
+        - Current task: {current_task}
+
+    # Tool result pruning
+    tool_result_pruning:
+      trigger: "after_tool_execution"
+      actions:
+        - "Remove verbose success messages"
+        - "Keep only error details on failure"
+        - "Truncate large outputs to relevant sections"
+      max_result_tokens: 1000
+
+    # Redundancy removal
+    redundancy_removal:
+      trigger: "before_each_request"
+      actions:
+        - "Deduplicate repeated information"
+        - "Merge similar context blocks"
+        - "Remove superseded data"
+
+    # Importance-based filtering
+    importance_ranking:
+      trigger: "context_approaching_limit"
+      method: "score_and_filter"
+      keep_percentage: 0.7
+
+      scoring:
+        recency: 0.3
+        relevance: 0.4
+        user_reference: 0.2
+        error_context: 0.1
+
+  limits:
+    max_context_tokens: 100000
+    trigger_compaction_at: 0.8 # 80% of limit
+    target_after_compaction: 0.5 # Aim for 50%
+
+# =============================================================================
+# JUST-IN-TIME (JIT) LOADING
+# Load information only when needed
+# =============================================================================
+
+jit_loading:
+  enabled: true
+
+  # What to keep as lightweight pointers
+  pointers:
+    files:
+      keep: "path only"
+      load_on: "user_references OR need_content"
+      max_preview: 50 # tokens for preview
+
+    skills:
+      keep: "name + 1-line description"
+      load_on: "query_matches"
+      tier_progression: "T1 → T2 → T3"
+
+    memories:
+      keep: "summary only"
+      load_on: "relevance_score > 0.7"
+      max_loaded: 5
+
+    decisions:
+      keep: "list of recent decisions"
+      load_on: "user_asks_why"
+
+  # Load triggers
+  triggers:
+    explicit:
+      - pattern: "@file:{path}"
+        action: "load_file"
+
+      - pattern: "@skill:{name}"
+        action: "load_skill"
+
+      - pattern: "@memory:{query}"
+        action: "search_memory"
+
+    implicit:
+      - context: "Discussing specific file"
+        action: "load_file_content"
+
+      - context: "Working with specific language"
+        action: "load_language_skill"
+
+      - context: "Debugging error"
+        action: "load_related_errors"
+
+  # Unload rules
+  unloading:
+    trigger: "context_limit_approaching"
+    strategy: "LRU_with_importance"
+    protected:
+      - "system_prompt"
+      - "current_task_context"
+      - "active_file_content"
+
+# =============================================================================
+# CONTEXT WINDOW MANAGEMENT
+# =============================================================================
+
+context_management:
+  # Token budgets by section
+  budgets:
+    system_prompt:
+      max_tokens: 5000
+      priority: 1
+
+    core_memory:
+      max_tokens: 2000
+      priority: 2
+
+    conversation_history:
+      max_tokens: 10000
+      priority: 3
+
+    working_memory:
+      max_tokens: 5000
+      priority: 4
+
+    skill_context:
+      max_tokens: 3000
+      priority: 5
+
+    tool_results:
+      max_tokens: 5000
+      priority: 6
+
+  # Dynamic allocation
+  dynamic_allocation:
+    enabled: true
+    rebalance_on: "context_change"
+
+    rules:
+      - context: "code_generation"
+        boost: ["skill_context", "working_memory"]
+        reduce: ["conversation_history"]
+
+      - context: "debugging"
+        boost: ["tool_results", "working_memory"]
+        reduce: ["skill_context"]
+
+      - context: "conversation"
+        boost: ["conversation_history"]
+        reduce: ["skill_context", "tool_results"]
+
+# =============================================================================
+# MEMORY HIERARCHY INTEGRATION
+# =============================================================================
+
+memory_integration:
+  # 5-tier memory loading
+  tiers:
+    working:
+      load: "always"
+      max_tokens: 500
+
+    session:
+      load: "always"
+      max_tokens: 2000
+      compaction: "summarize_old"
+
+    core:
+      load: "always"
+      max_tokens: 500
+      source: "core_memory/*.json"
+
+    state:
+      load: "on_demand"
+      max_tokens: 500
+
+    semantic:
+      load: "on_query_match"
+      retrieval: "top_k"
+      k: 5
+
+    graph:
+      load: "on_relationship_query"
+      max_hops: 3
+      max_entities: 20
+
+# =============================================================================
+# ANALYTICS
+# =============================================================================
+
+analytics:
+  enabled: true
+
+  track:
+    - "tokens_used_by_section"
+    - "cache_hit_rate"
+    - "compaction_frequency"
+    - "jit_load_count"
+
+  reports:
+    frequency: "session_end"
+    location: ".agent/memory/metrics/"

package/.agent/core/CONTEXT_PRIORITY.yaml

@@ -0,0 +1,357 @@
+# DOMYH Awesome Code — Context Priority Scoring v6.2.0
+# Multi-factor scoring to determine what enters context
+# Ensures 90%+ context relevance
+
+version: "6.1.2"
+engine: context_priority_scorer
+
+# =============================================================================
+# SCORING OVERVIEW
+# =============================================================================
+
+# Every piece of information gets a score before entering context
+# Higher scored content gets priority, lower scored gets summarized or skipped
+
+overview:
+  goal: "Maximize context relevance while minimizing tokens"
+  method: "Multi-factor weighted scoring"
+  output: "Priority-ordered context injection queue"
+
+# =============================================================================
+# SCORING FACTORS
+# =============================================================================
+
+factors:
+  # Factor 1: Recency (25%)
+  recency:
+    weight: 0.25
+    description: "Newer content is more likely relevant"
+
+    scoring:
+      # Time-based decay
+      fresh: # < 1 hour
+        age: 3600
+        score: 1.0
+      recent: # 1-24 hours
+        age: 86400
+        score: 0.8
+      this_week: # 1-7 days
+        age: 604800
+        score: 0.6
+      this_month: # 7-30 days
+        age: 2592000
+        score: 0.4
+      older: # > 30 days
+        score: 0.2
+
+    formula: |
+      age_seconds = now() - content.timestamp
+      if age_seconds < 3600: return 1.0
+      if age_seconds < 86400: return 0.8
+      if age_seconds < 604800: return 0.6
+      if age_seconds < 2592000: return 0.4
+      return 0.2
+
+  # Factor 2: Relevance (40%)
+  relevance:
+    weight: 0.40
+    description: "How well content matches current task"
+
+    methods:
+      # Keyword matching (fast, low accuracy)
+      keyword:
+        weight: 0.3
+        calculation: |
+          task_keywords = extract_keywords(current_task)
+          content_keywords = content.keywords
+          overlap = intersection(task_keywords, content_keywords)
+          return len(overlap) / len(task_keywords)
+
+      # Entity matching (medium speed, medium accuracy)
+      entity:
+        weight: 0.3
+        calculation: |
+          task_entities = extract_entities(current_task)
+          content_entities = content.entities
+          overlap = intersection(task_entities, content_entities)
+          return len(overlap) / len(task_entities)
+
+      # Semantic similarity (slow, high accuracy)
+      semantic:
+        weight: 0.4
+        calculation: |
+          task_embedding = embed(current_task)
+          content_embedding = content.embedding
+          return cosine_similarity(task_embedding, content_embedding)
+
+    formula: |
+      keyword_score = keyword_match()
+      entity_score = entity_match()
+      semantic_score = semantic_similarity()
+      return (keyword_score * 0.3) + (entity_score * 0.3) + (semantic_score * 0.4)
+
+  # Factor 3: Importance (25%)
+  importance:
+    weight: 0.25
+    description: "Intrinsic importance of content"
+
+    levels:
+      critical:
+        score: 1.0
+        examples:
+          - active_errors
+          - blocking_issues
+          - security_vulnerabilities
+          - user_corrections
+
+      high:
+        score: 0.8
+        examples:
+          - recent_decisions
+          - current_task_context
+          - active_warnings
+          - user_preferences
+
+      medium:
+        score: 0.6
+        examples:
+          - project_state
+          - recent_workflows
+          - past_audit_findings
+          - pattern_examples
+
+      low:
+        score: 0.4
+        examples:
+          - historical_decisions
+          - old_session_notes
+          - resolved_errors
+          - archived_content
+
+      minimal:
+        score: 0.2
+        examples:
+          - raw_logs
+          - verbose_output
+          - old_drafts
+
+    auto_detection:
+      critical_patterns:
+        - /error|exception|critical|blocker/i
+        - /security|vulnerability|breach/i
+        - /urgent|asap|immediately/i
+      high_patterns:
+        - /decision|decided|chose/i
+        - /important|priority|key/i
+
+  # Factor 4: Frequency (10%)
+  frequency:
+    weight: 0.10
+    description: "How often content is accessed"
+
+    scoring:
+      heavily_used: # > 20 accesses
+        threshold: 20
+        score: 1.0
+      frequently_used: # 10-20 accesses
+        threshold: 10
+        score: 0.8
+      moderately_used: # 5-10 accesses
+        threshold: 5
+        score: 0.6
+      occasionally_used: # 2-5 accesses
+        threshold: 2
+        score: 0.4
+      rarely_used: # 1 access
+        threshold: 1
+        score: 0.2
+      never_used: # 0 accesses
+        threshold: 0
+        score: 0.1
+
+    formula: |
+      access_count = content.access_count
+      if access_count >= 20: return 1.0
+      if access_count >= 10: return 0.8
+      if access_count >= 5: return 0.6
+      if access_count >= 2: return 0.4
+      if access_count >= 1: return 0.2
+      return 0.1
+
+# =============================================================================
+# FINAL SCORE CALCULATION
+# =============================================================================
+
+calculation:
+  # Scoring formula centralized - Single Source of Truth
+  formula_reference: core/SCORING_FORMULA.yaml
+  # Original formula (now in SCORING_FORMULA.yaml):
+  # final_score = (recency * 0.25) + (relevance * 0.40) +
+  #               (importance * 0.25) + (frequency * 0.10)
+
+  # Score interpretation
+  interpretation:
+    excellent: # 0.85 - 1.0
+      range: [0.85, 1.0]
+      action: always_include
+      token_budget: full
+
+    good: # 0.70 - 0.85
+      range: [0.70, 0.85]
+      action: include_if_budget
+      token_budget: full
+
+    moderate: # 0.50 - 0.70
+      range: [0.50, 0.70]
+      action: include_summarized
+      token_budget: 50%
+
+    low: # 0.30 - 0.50
+      range: [0.30, 0.50]
+      action: pointer_only
+      token_budget: minimal
+
+    skip: # 0.0 - 0.30
+      range: [0.0, 0.30]
+      action: do_not_include
+      token_budget: 0
+
+# =============================================================================
+# PRIORITY QUEUE
+# =============================================================================
+
+queue:
+  # Build priority queue from scored content
+  building:
+    1: collect_all_candidates
+    2: score_each_candidate
+    3: sort_by_score_desc
+    4: apply_deduplication
+    5: build_token_budget_queue
+
+  # Queue processing
+  processing:
+    max_candidates: 50
+    max_tokens: 2000
+    selection_method: greedy_by_score
+
+  # Overflow handling
+  overflow:
+    strategy: summarize_lowest
+    target_tokens: 1500
+    preserve_count: 10 # Always keep top 10
+
+# =============================================================================
+# CONTENT-TYPE BONUSES
+# =============================================================================
+
+# Some content types get scoring bonuses based on context
+
+type_bonuses:
+  # When debugging
+  in_debug_context:
+    error_patterns: +0.2
+    similar_errors: +0.2
+    stack_traces: +0.1
+
+  # When coding
+  in_code_context:
+    patterns: +0.15
+    decisions: +0.1
+    tests: +0.1
+
+  # When planning
+  in_plan_context:
+    decisions: +0.2
+    constraints: +0.15
+    requirements: +0.15
+
+  # When auditing
+  in_audit_context:
+    previous_audits: +0.2
+    security_findings: +0.15
+    quality_metrics: +0.1
+
+# =============================================================================
+# WORKFLOW PROFILES
+# =============================================================================
+
+workflow_profiles:
+  /code:
+    boost:
+      patterns: +0.2
+      errors: +0.15
+    reduce:
+      old_audits: -0.2
+
+  /debug:
+    boost:
+      errors: +0.3
+      similar_issues: +0.25
+    reduce:
+      decisions: -0.1
+
+  /ap:
+    boost:
+      audits: +0.3
+      security: +0.2
+    reduce:
+      session_notes: -0.1
+
+  /plan:
+    boost:
+      decisions: +0.25
+      constraints: +0.2
+    reduce:
+      errors: -0.1
+
+  /recap:
+    boost:
+      session: +0.2
+      decisions: +0.15
+    reduce:
+      raw_logs: -0.2
+
+# =============================================================================
+# CACHING
+# =============================================================================
+
+cache:
+  # Cache computed scores
+  scores:
+    duration: 60 # 1 minute
+    invalidate_on:
+      - content_update
+      - new_task
+
+  # Cache embeddings (for relevance)
+  embeddings:
+    duration: 3600 # 1 hour
+    max_size: 1000
+
+  # Cache final queue
+  queue:
+    duration: 30 # 30 seconds
+    invalidate_on:
+      - score_change
+      - new_content
+
+# =============================================================================
+# MONITORING
+# =============================================================================
+
+monitoring:
+  track:
+    - average_score_distribution
+    - inclusion_rate_by_type
+    - token_efficiency
+    - cache_hit_rate
+
+  alerts:
+    low_average_score:
+      threshold: 0.5
+      action: log_warning
+
+    high_skip_rate:
+      threshold: 0.7
+      action: review_content_quality