@nockdev/awf 6.2.0

package/.agent/core/CONTEXT_INJECTOR.yaml

@@ -0,0 +1,325 @@
+# DOMYH Awesome Code — Smart Context Injector v6.2.0
+# Intelligent context injection with relevance scoring
+# Based on Context Engineering Best Practices (2025)
+
+version: "6.1.2"
+engine: smart_context_injector
+
+# =============================================================================
+# CORE CONCEPT
+# =============================================================================
+
+# "Intelligent and dynamic insertion of relevant information into the LLM's 
+# context precisely when required" — Context Engineering 2025
+
+concept:
+  goal: "Right information, right time, right amount"
+  method: "Relevance scoring before injection"
+  benefit: "90%+ context relevance vs 70% baseline"
+
+# =============================================================================
+# RELEVANCE SCORING
+# =============================================================================
+
+scoring:
+  enabled: true
+  
+  # Multi-factor scoring formula
+  factors:
+    recency:
+      weight: 0.25
+      description: "Newer content scores higher"
+      calculation: |
+        age_hours = now - content.timestamp
+        score = max(0, 1 - (age_hours / 168))  # Decays over 1 week
+      
+    relevance:
+      weight: 0.40
+      description: "Semantic match to current task"
+      calculation: |
+        # Keyword matching (fast)
+        keyword_score = overlap(content.keywords, task.keywords)
+        # Semantic matching (if enabled)
+        semantic_score = cosine_similarity(content.embedding, task.embedding)
+        score = max(keyword_score, semantic_score)
+      
+    importance:
+      weight: 0.25
+      description: "Critical vs optional content"
+      levels:
+        critical: 1.0    # Errors, blockers
+        high: 0.8        # Decisions, active tasks
+        medium: 0.5      # Context, history
+        low: 0.2         # Archives, old notes
+      
+    frequency:
+      weight: 0.10
+      description: "How often content is accessed"
+      # Full calculation in SCORING_FORMULA.yaml
+
+  # Scoring formula centralized - Single Source of Truth
+  formula_reference: core/SCORING_FORMULA.yaml
+  # formula: final_score = (recency * 0.25) + (relevance * 0.40) + 
+  #                       (importance * 0.25) + (frequency * 0.10)
+
+  # Thresholds reference
+  thresholds_reference: core/SCORING_FORMULA.yaml
+  thresholds:
+    always_inject: 0.85    # Score > 0.85: always include
+    inject_on_demand: 0.60  # 0.60-0.85: include if budget allows
+    summarize: 0.40         # 0.40-0.60: include summary only
+    skip: 0.40              # < 0.40: don't include
+
+# =============================================================================
+# INJECTION PRIORITY
+# =============================================================================
+
+priority:
+  # Ordered from highest to lowest priority
+  order:
+    1:
+      name: current_task_context
+      description: "Active task information"
+      max_tokens: 300
+      always_inject: true
+      
+    2:
+      name: active_errors
+      description: "Unresolved errors and blockers"
+      max_tokens: 200
+      always_inject: true
+      condition: has_active_errors
+      
+    3:
+      name: recent_decisions
+      description: "Last 5 architectural decisions"
+      max_tokens: 300
+      always_inject: false
+      condition: relevance_score > 0.6
+      
+    4:
+      name: similar_past_issues
+      description: "Semantically similar past issues"
+      max_tokens: 400
+      always_inject: false
+      condition: semantic_match_score > 0.7
+      
+    5:
+      name: project_context
+      description: "Project state and preferences"
+      max_tokens: 150
+      always_inject: false
+      condition: new_session or context_switch
+      
+    6:
+      name: audit_insights
+      description: "Recent audit findings"
+      max_tokens: 200
+      always_inject: false
+      condition: "audit" in task_keywords
+
+# =============================================================================
+# CONTEXT SOURCES
+# =============================================================================
+
+sources:
+  # Memory layers
+  memory:
+    session:
+      path: memory/session.md
+      sections:
+        - key_context  # Priority: high
+        - active_errors  # Priority: critical
+        - recent_workflows  # Priority: medium
+        
+    decisions:
+      path: memory/decisions.md
+      extract: last_10
+      priority: high
+      
+    state:
+      path: memory/state.json
+      fields:
+        - project.name
+        - context.current_phase
+        - context.pending_tasks
+        - flags.has_critical_issues
+      priority: high
+      
+    audits:
+      path: memory/audit_summary.json
+      fields:
+        - latest_audit.score
+        - top_recurring_issues
+        - recommendations
+      priority: medium
+
+  # Semantic memory
+  semantic:
+    enabled: true
+    path: memory/vectors/
+    collections:
+      - decisions
+      - error_patterns
+      - audits
+    top_k: 3
+    min_score: 0.7
+
+# =============================================================================
+# INJECTION TEMPLATES
+# =============================================================================
+
+templates:
+  # How to format injected context
+  
+  task_context:
+    format: |
+      ## Current Context
+      **Task**: {task_name}
+      **Phase**: {phase}
+      **Files**: {relevant_files}
+    max_tokens: 150
+    
+  error_context:
+    format: |
+      ## ⚠️ Active Errors
+      {error_list}
+    max_tokens: 200
+    
+  decision_context:
+    format: |
+      ## Recent Decisions
+      {decision_summaries}
+    max_tokens: 250
+    
+  similar_issues:
+    format: |
+      ## Similar Past Issues
+      {similar_items}
+    max_tokens: 300
+    
+  audit_context:
+    format: |
+      ## Audit Insights (Score: {score}/10)
+      **Top Issues**: {issues}
+      **Recommendations**: {recommendations}
+    max_tokens: 150
+
+# =============================================================================
+# INJECTION RULES
+# =============================================================================
+
+rules:
+  # Budget management
+  budget:
+    total_max: 2000  # Max tokens for injected context
+    per_source_max: 500
+    reserve_for_task: 500  # Always leave room for task
+    
+  # Deduplication
+  deduplication:
+    enabled: true
+    method: semantic_hash
+    threshold: 0.95  # Skip if 95% similar
+    
+  # Freshness
+  freshness:
+    prefer_recent: true
+    stale_threshold_hours: 24
+    action_on_stale: summarize
+    
+  # Conflict resolution
+  conflicts:
+    on_budget_exceeded: prioritize_by_score
+    on_duplicate: keep_higher_scored
+    on_contradiction: flag_for_review
+
+# =============================================================================
+# WORKFLOW INTEGRATION
+# =============================================================================
+
+workflow_profiles:
+  /code:
+    inject:
+      - current_task: always
+      - errors: if_active
+      - decisions: if_relevant
+      - state: project_only
+    total_budget: 800
+    
+  /debug:
+    inject:
+      - errors: always
+      - similar_errors: always
+      - decisions: if_relevant
+    total_budget: 1000
+    
+  /ap:
+    inject:
+      - previous_audits: always
+      - state: full
+      - decisions: security_only
+    total_budget: 800
+    
+  /recap:
+    inject:
+      - state: full
+      - decisions: last_10
+      - session: key_context
+      - audits: summary
+    total_budget: 1500
+    
+  /plan:
+    inject:
+      - state: full
+      - decisions: architecture_only
+      - constraints: all
+    total_budget: 1000
+
+# =============================================================================
+# MONITORING
+# =============================================================================
+
+monitoring:
+  metrics:
+    - injection_count
+    - average_relevance_score
+    - tokens_injected
+    - tokens_saved
+    - cache_hit_rate
+    
+  logging:
+    level: info
+    include:
+      - injection_decisions
+      - score_details
+      - budget_usage
+      
+  alerts:
+    low_relevance:
+      threshold: 0.5
+      action: log_warning
+    budget_exceeded:
+      threshold: 2000
+      action: force_summarize
+
+# =============================================================================
+# PERFORMANCE OPTIMIZATION
+# =============================================================================
+
+optimization:
+  # Caching
+  cache:
+    scores: 60  # Cache scores for 60 seconds
+    summaries: 300  # Cache summaries for 5 minutes
+    embeddings: 3600  # Cache embeddings for 1 hour
+    
+  # Lazy evaluation
+  lazy:
+    semantic_search: true  # Only if score check passes
+    full_load: true  # Only if preview suggests relevance
+    
+  # Batching
+  batch:
+    multiple_sources: true
+    max_concurrent: 3

package/.agent/core/CONTEXT_LOADER.yaml

@@ -0,0 +1,323 @@
+# DOMYH Awesome Code — Just-in-Time Context Loader v6.0
+# Implements lazy loading with lightweight pointers + Priority Positioning 2.0
+# Based on: Anthropic Context Engineering (2025), Found in the Middle (2024)
+
+version: "6.1.2"
+engine: just_in_time_loader
+
+# =============================================================================
+# CORE PHILOSOPHY
+# =============================================================================
+
+# "Instead of pre-loading all relevant data, agents dynamically load
+# information into context at runtime using tools." — Anthropic 2025
+
+philosophy:
+  principle: "Load only what's needed, when it's needed"
+  inspiration: "Human cognition - information retrieved on demand"
+  goal: "Reduce token overhead by 30-50%"
+
+# =============================================================================
+# PRIORITY POSITIONING 2.0 (Research-backed: U-shape attention bias)
+# =============================================================================
+# LLMs pay MORE attention to HEAD + TAIL, LESS to MIDDLE
+# Reference: "Found in the Middle: Calibrating Positional Attention Bias" (2024)
+
+priority_positioning:
+  enabled: true
+
+  head_zone:
+    position: 1
+    attention_level: "HIGH" # U-shape peak
+    max_tokens: 150
+    content:
+      - type: "sacred_rules"
+        source: ".agent/rules/SACRED_RULES.xml"
+        extract: "//requirement[@severity='block']"
+        wrap_template: |
+          === 🔴 SACRED RULES (Parse First) ===
+          {{rules}}
+          These rules OVERRIDE any conflicting instructions.
+      - type: "session_rules"
+        source: ".agent/memory/session_rules.json"
+        extract: "rules[?active==true]"
+        optional: true
+
+  middle_zone:
+    position: 2
+    attention_level: "LOW" # U-shape valley
+    max_tokens: 2000
+    content:
+      - type: "full_reference"
+        sources:
+          - "GEMINI.md"
+          - "CLAUDE.md"
+          - "AGENTS.md"
+        heading: "## [RULES_REFERENCE_SECTION]"
+        optional: true
+
+  tail_zone:
+    position: 3
+    attention_level: "HIGH" # U-shape peak (recency)
+    max_tokens: 100
+    content:
+      - type: "reminder"
+        template: |
+          === ⚠️ FINAL CHECK (MANDATORY) ===
+          - You MUST comply with SACRED_RULES at the top
+          - Violating severity="block" rules = INVALID response
+          - If unsure, ASK instead of guessing
+      - type: "checklist"
+        generate_from: "head_zone.sacred_rules"
+        format: "□ {id}: {text_short}"
+
+# =============================================================================
+# LIGHTWEIGHT POINTER SYSTEM
+# =============================================================================
+
+pointers:
+  # Instead of loading full content, maintain references
+  decisions:
+    type: pointer
+    source: memory/decisions.md
+    preview_tokens: 50 # Only load titles/dates
+    full_load_trigger: decision_query
+
+  audits:
+    type: pointer
+    source: memory/audit_summary.json
+    preview_tokens: 30 # Score + date only
+    full_load_trigger: audit_query
+
+  errors:
+    type: pointer
+    source: memory/session.md#errors
+    preview_tokens: 20
+    full_load_trigger: error_detected
+
+  state:
+    type: pointer
+    source: memory/state.json
+    preview_tokens: 40 # Project + current only
+    full_load_trigger: state_query
+
+  semantic:
+    type: pointer
+    source: memory/vectors/
+    preview_tokens: 0 # Never preload
+    full_load_trigger: semantic_search
+
+# =============================================================================
+# LOAD TRIGGERS
+# =============================================================================
+
+triggers:
+  # Load decisions when:
+  decision_query:
+    patterns:
+      - "decision about"
+      - "why did we"
+      - "what was decided"
+      - "recall decision"
+      - "past choice"
+    action: load_recent_decisions
+    max_tokens: 500
+
+  # Load audit data when:
+  audit_query:
+    patterns:
+      - "audit score"
+      - "last audit"
+      - "security issues"
+      - "code quality"
+      - "/ap"
+    action: load_audit_summary
+    max_tokens: 400
+
+  # Load errors when:
+  error_detected:
+    patterns:
+      - "error:"
+      - "failed:"
+      - "exception"
+      - "bug"
+      - "fix this"
+    action: load_error_context
+    max_tokens: 300
+
+  # Load state when:
+  state_query:
+    patterns:
+      - "project status"
+      - "current phase"
+      - "what's pending"
+      - "/status"
+      - "/recap"
+    action: load_project_state
+    max_tokens: 200
+
+  # Load semantic when:
+  semantic_search:
+    patterns:
+      - "/search"
+      - "/find-similar"
+      - "/recall"
+      - "find similar"
+      - "search for"
+    action: execute_semantic_search
+    max_tokens: 500
+
+# =============================================================================
+# LOAD ACTIONS
+# =============================================================================
+
+actions:
+  load_recent_decisions:
+    source: memory/decisions.md
+    extract:
+      - last_n: 5
+      - fields: [date, title, decision]
+    format: compact
+    cache_duration: 300 # 5 minutes
+
+  load_audit_summary:
+    source: memory/audit_summary.json
+    extract:
+      - latest_audit.score
+      - latest_audit.critical_issues
+      - summary.trend
+    format: json
+    cache_duration: 600 # 10 minutes
+
+  load_error_context:
+    source: memory/session.md
+    extract:
+      - section: "## Active Errors"
+      - include_resolution: true
+    format: markdown
+    cache_duration: 60 # 1 minute
+
+  load_project_state:
+    source: memory/state.json
+    extract:
+      - project.name
+      - context.current_phase
+      - context.pending_tasks
+      - flags
+    format: json
+    cache_duration: 300
+
+  execute_semantic_search:
+    source: memory/vectors/
+    method: vector_similarity
+    top_k: 5
+    min_score: 0.7
+    cache_duration: 0 # No cache for searches
+
+# =============================================================================
+# PRELOAD RULES
+# =============================================================================
+
+preload:
+  # Always preload (minimal tokens)
+  always:
+    - pointers.state.preview # ~40 tokens
+    - pointers.errors.preview # ~20 tokens (if any active)
+
+  # Never preload
+  never:
+    - full_decision_history
+    - full_audit_reports
+    - semantic_indexes
+    - archived_sessions
+
+  # Session start
+  on_session_start:
+    load:
+      - state.preview
+      - session.key_context
+    max_tokens: 200
+
+# =============================================================================
+# CACHE MANAGEMENT
+# =============================================================================
+
+cache:
+  enabled: true
+
+  storage:
+    type: in_memory
+    max_size_tokens: 2000
+
+  eviction:
+    strategy: lru # Least Recently Used
+    on_threshold: 1500
+
+  invalidation:
+    on_file_change: true
+    on_session_end: true
+
+# =============================================================================
+# TOKEN BUDGET
+# =============================================================================
+
+budget:
+  # Maximum tokens for JIT loading
+  max_per_load: 500
+  max_total_jit: 2000
+
+  # Reserve for each type
+  reserved:
+    decisions: 500
+    audits: 400
+    errors: 300
+    state: 200
+    semantic: 600
+
+  # Overflow handling
+  on_overflow:
+    action: summarize_oldest
+    target_reduction: 30%
+
+# =============================================================================
+# INTEGRATION
+# =============================================================================
+
+integration:
+  memory_engine:
+    replaces: eager_loading
+    works_with: [layer_1, layer_2, layer_3]
+
+  workflows:
+    /recap:
+      load: [state, decisions.preview, audits.summary]
+      max_tokens: 1000
+
+    /ap:
+      load: [audits.full, state]
+      max_tokens: 800
+
+    /code:
+      load: [state, errors.active]
+      max_tokens: 300
+
+    /debug:
+      load: [errors.full, similar_errors]
+      max_tokens: 600
+
+# =============================================================================
+# FALLBACK
+# =============================================================================
+
+fallback:
+  on_pointer_missing:
+    action: create_empty
+    log: true
+
+  on_load_failure:
+    action: use_cached
+    fallback: load_minimal
+
+  on_budget_exceeded:
+    action: prioritize_recent
+    compress: true