continuous-learning-construction

建設オートメーションの継続的学習

このスキルは、建設オートメーションのセッションから価値あるパターン、解決策、ベストプラクティスを自動的に抽出し、組織のナレッジを構築します。

使用するタイミング

次の場合にこのスキルを有効にします。

• 複雑な見積もりセッションの終了時
• 自明でないデータ処理の問題を解決した後
• 新しい統合パターンを発見したとき
• ドキュメント処理を正常に完了した後
• 新しい自動化ワークフローを開発したとき

パターン抽出フレームワーク

1. セッション分析

class ConstructionSessionAnalyzer:
    """Extract learnings from automation sessions"""

    # Categories of learnable patterns
    PATTERN_CATEGORIES = [
        'data_processing',      # Data transformation patterns
        'estimation',           # Cost estimation techniques
        'scheduling',           # Schedule optimization patterns
        'integration',          # API/system integration patterns
        'document_processing',  # Document handling patterns
        'quality_assurance',    # Validation and QA patterns
        'error_handling',       # Error resolution patterns
        'optimization'          # Performance optimization patterns
    ]

    def analyze_session(self, session_log: list) -> dict:
        """Extract patterns from session history"""

        patterns = {
            'successful_solutions': [],
            'error_resolutions': [],
            'optimization_discoveries': [],
            'integration_patterns': [],
            'reusable_code': [],
            'decision_rationales': []
        }

        for entry in session_log:
            if self._is_solution(entry):
                patterns['successful_solutions'].append(
                    self._extract_solution_pattern(entry)
                )

            if self._is_error_resolution(entry):
                patterns['error_resolutions'].append(
                    self._extract_error_pattern(entry)
                )

            if self._is_optimization(entry):
                patterns['optimization_discoveries'].append(
                    self._extract_optimization(entry)
                )

        return patterns

2. 建設に関するナレッジカテゴリ

2.1 コスト見積もりパターン

# Example learned pattern
pattern:
  name: "electrical_cost_adjustment_pattern"
  category: "estimation"
  context: "When estimating electrical work for high-rise buildings"
  problem: "Standard rates don't account for vertical transportation costs"
  solution: |
    Apply height factor multiplier:
    - Floors 1-5: 1.0x base rate
    - Floors 6-15: 1.15x base rate
    - Floors 16-30: 1.25x base rate
    - Floors 30+: 1.35x base rate
  confidence: 0.85
  source_sessions: ["session_2026_01_15", "session_2026_01_20"]
  validations: 3

2.2 BIM データ処理パターン

pattern:
  name: "revit_level_extraction"
  category: "data_processing"
  context: "Extracting elements by level from Revit exports"
  problem: "Elements sometimes missing level association"
  solution: |
    1. First check 'Level' parameter
    2. If missing, check 'Reference Level' parameter
    3. If still missing, derive from bounding box Z coordinate
    4. Map Z ranges to known level elevations
  code_snippet: |
    def get_element_level(element: dict, levels: list) -> str:
        # Direct level parameter
        if level := element.get('Level'):
            return level

        # Reference level fallback
        if ref_level := element.get('Reference Level'):
            return ref_level

        # Derive from geometry
        z_coord = element['BoundingBox']['Min']['Z']
        return find_nearest_level(z_coord, levels)
  confidence: 0.92

2.3 統合パターン

pattern:
  name: "procore_rate_limit_handling"
  category: "integration"
  context: "Syncing data with Procore API"
  problem: "API returns 429 Too Many Requests during bulk operations"
  solution: |
    Implement exponential backoff with jitter:
    1. Initial delay: 1 second
    2. Multiply by 2 on each retry
    3. Add random jitter (0-500ms)
    4. Max retries: 5
    5. Max delay: 32 seconds
  code_snippet: |
    async def procore_request_with_retry(url, data):
        delay = 1
        for attempt in range(5):
            try:
                response = await procore_api.post(url, data)
                return response
            except RateLimitError:
                jitter = random.uniform(0, 0.5)
                await asyncio.sleep(delay + jitter)
                delay *= 2
        raise MaxRetriesExceeded()
  confidence: 0.95

2.4 エラー解決パターン

pattern:
  name: "cwicr_no_match_resolution"
  category: "error_handling"
  context: "CWICR semantic search returns no relevant matches"
  problem: "Query too specific or uses non-standard terminology"
  solution: |
    Resolution steps:
    1. Simplify query to core concepts
    2. Remove brand names and specifications
    3. Try alternative terminology (US vs UK terms)
    4. Expand search to parent category
    5. If still no match, flag for manual mapping
  examples:
    - original: "Kohler K-4519 wall-mounted water closet"
      simplified: "wall mounted toilet"
    - original: "Lutron Caseta wireless dimmer switch"
      simplified: "dimmer switch"
  confidence: 0.88

3. 学習パイプライン

class ConstructionLearningPipeline:
    """Continuous learning pipeline for construction automation"""

    def __init__(self, knowledge_base_path: str):
        self.kb_path = knowledge_base_path
        self.patterns = self._load_patterns()

    def learn_from_session(self, session: dict) -> list:
        """Extract and store learnings from session"""

        # Analyze session
        analyzer = ConstructionSessionAnalyzer()
        new_patterns = analyzer.analyze_session(session['log'])

        # Validate patterns
        validated = []
        for pattern in new_patterns['successful_solutions']:
            if self._validate_pattern(pattern):
                # Check if similar pattern exists
                existing = self._find_similar_pattern(pattern)
                if existing:
                    # Reinforce existing pattern
                    self._reinforce_pattern(existing, pattern)
                else:
                    # Add new pattern
                    self._add_pattern(pattern)
                validated.append(pattern)

        # Persist to knowledge base
        self._save_patterns()

        return validated

    def apply_learnings(self, context: dict) -> list:
        """Retrieve relevant patterns for current context"""

        relevant_patterns = []

        for pattern in self.patterns:
            similarity = self._calculate_similarity(pattern['context'], context)
            if similarity > 0.7:
                relevant_patterns.append({
                    'pattern': pattern,
                    'relevance': similarity
                })

        return sorted(relevant_patterns, key=lambda x: x['relevance'], reverse=True)

    def _validate_pattern(self, pattern: dict) -> bool:
        """Validate pattern before adding to knowledge base"""

        # Check minimum confidence
        if pattern.get('confidence', 0) < 0.6:
            return False

        # Check for code quality (if code snippet)
        if code := pattern.get('code_snippet'):
            if not self._is_valid_code(code):
                return False

        # Check for completeness
        required_fields = ['name', 'category', 'context', 'solution']
        if not all(f in pattern for f in required_fields):
            return False

        return True

4. ナレッジベースの構造

knowledge_base/
├── patterns/
│   ├── estimation/
│   │   ├── height_factors.yaml
│   │   ├── material_adjustments.yaml
│   │   └── labor_productivity.yaml
│   ├── data_processing/
│   │   ├── revit_extraction.yaml
│   │   ├── ifc_parsing.yaml
│   │   └── excel_transformations.yaml
│   ├── integration/
│   │   ├── procore_patterns.yaml
│   │   ├── plangrid_patterns.yaml
│   │   └── webhook_handlers.yaml
│   └── error_handling/
│       ├── cwicr_resolutions.yaml
│       ├── api_errors.yaml
│       └── data_validation.yaml
├── code_snippets/
│   ├── python/
│   ├── javascript/
│   └── sql/
├── decision_trees/
│   ├── estimate_type_selection.yaml
│   ├── schedule_method_selection.yaml
│   └── integration_approach.yaml
└── metrics/
    ├── pattern_usage.json
    └── success_rates.json

5. セッション終了時の学習プロンプト

各建設オートメーションセッションの終了時に：

## Session Learning Review

### What Worked Well
- [Successful approaches discovered]
- [Efficient patterns used]
- [Integrations that worked smoothly]

### Challenges Overcome
- [Errors encountered and how resolved]
- [Workarounds developed]
- [Edge cases handled]

### New Patterns Discovered
- [Novel approaches to problems]
- [Optimization techniques found]
- [Reusable code created]

### Knowledge to Preserve
- [Key learnings to remember]
- [Context-specific solutions]
- [Client/project-specific adaptations]

### Recommendations for Future
- [Improvements to suggest]
- [Patterns to apply elsewhere]
- [Automation opportunities identified]

6. パターンの適用

新しい建設タスクを開始するとき：

def suggest_approaches(task_context: dict) -> list:
    """Suggest learned approaches for new tasks"""

    pipeline = ConstructionLearningPipeline('knowledge_base/')
    relevant = pipeline.apply_learnings(task_context)

    suggestions = []
    for item in relevant[:5]:  # Top 5 suggestions
        pattern = item['pattern']
        suggestions.append({
            'name': pattern['name'],
            'relevance': f"{item['relevance']*100:.0f}%",
            'summary': pattern['solution'][:200],
            'confidence': pattern['confidence'],
            'previous_uses': pattern.get('usage_count', 0)
        })

    return suggestions

他のスキルとの統合

このスキルは、次のスキルと連携します。

• verification-loop-construction：検証の失敗から学習します
• security-review-construction：セキュリティパターンを記録します
• estimation skills：見積もりのナレッジベースを構築します
• integration skills：API パターンを記録します

使用コマンド

# Extract learnings from current session
/learn

# View patterns for current context
/suggest-patterns

# Add manual pattern
/add-pattern --category estimation --name "my_pattern"

# Export knowledge base
/export-kb --format yaml

---

すべてのセッションが学習の機会です。ナレッジを記録し、時間をかけて専門性を高めましょう。