This commit improves keyword generation by providing hierarchical context for each element and fixing the LLM response format parsing.
Changes:
1. lib/MissingKeywords.js:
- Add buildHierarchicalContext() to generate compact contextual info for each element
- Display hierarchy in prompt (e.g., "H2 existants: 'Titre1', 'Titre2'")
- For Txt elements: show associated MC keyword + parent title
- For FAQ elements: count existing FAQs
- Fix LLM response format by providing 3 concrete examples from actual list
- Add explicit warning to use exact tag names [Titre_H2_3], [Txt_H2_6]
- Improve getElementContext() to better retrieve hierarchical elements
2. lib/selective-enhancement/SelectiveUtils.js:
- Fix createTypedPrompt() to use specific keyword from resolvedContent
- Remove fallback to csvData.mc0 (log error if no specific keyword)
3. lib/pipeline/PipelineExecutor.js:
- Integrate generateMissingSheetVariables() as "Étape 0" before extraction
Prompt format now:
1. [Titre_H2_3] (titre) — H2 existants: "Titre1", "Titre2"
2. [Txt_H2_6] (texte) — MC: "Plaque dibond" | Parent: "Guide dibond"
3. [Faq_q_1] (question) — 3 FAQ existantes
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
🐛 Problème identifié:
- Le système utilisait toujours csvData.mc0 (mot-clé général) pour tous les éléments
- Exemple: "plaque numero de maison" au lieu de "Plaque en PVC : économique et léger"
✅ Solution:
- Ajout paramètre specificKeyword à createTypedPrompt()
- Extraction de item.originalElement.resolvedContent avant génération
- Pas de fallback sur csvData.mc0 (log d'erreur si manquant)
- Détection des variables non résolues "[Mc+1_5 non résolu]"
📍 Fichier modifié:
- lib/selective-enhancement/SelectiveUtils.js:603-615 (createTypedPrompt)
- lib/selective-enhancement/SelectiveUtils.js:1083-1091 (extraction + appel)
🎯 Impact:
- Chaque élément utilise maintenant son mot-clé spécifique
- Prompts LLM plus précis et contextualisés
- Meilleure qualité de contenu généré
🤖 Generated with Claude Code (https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Ajouts majeurs:
- Détection intelligente des contraintes de longueur dans les instructions (évite conflits entre instruction et type)
- Contexte de personnalité enrichi: colonnes B, D, E, F, H, I, J du Google Sheet
- Randomisation des éléments de personnalité (max 2 par catégorie pour variabilité)
- Injection du titre associé dans le prompt des textes/intros pour cohérence
- Exclusion de personnalité pour questions FAQ (seulement réponses)
Corrections:
- Fix détection type: vérification préfixes (Intro_, Titre_, Txt_) avant suffixes (_title, _text)
- Fix parsing vocabulairePref pour gérer arrays et strings
- Logs détaillés sur détection de contraintes
Impact:
- Meilleure cohérence titre→texte
- Plus de variabilité anti-détection
- Respect strict des contraintes de longueur spécifiées
- Personnalités mieux contextualisées dans les prompts
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Plan détaillé pour réintégrer la logique de génération par couples
et contraintes de longueur du système legacy.
Analyse COT complète avec:
- Comparaison plan initial vs legacy code
- Identification erreurs de conception
- 2 options d'implémentation (Option A recommandée)
- Risques et mitigations
- Critères de succès
- Stratégie de rollback
Status: EN ATTENTE VALIDATION UTILISATEUR
Action requise: Analyser buildSmartHierarchy() output
🤖 Generated with Claude Code
Co-Authored-By: Claude <noreply@anthropic.com>
- Add documentation for new advanced systems (DynamicPromptEngine, TrendManager, WorkflowEngine, BatchProcessing)
- Document 11 web interfaces and complete RESTful API
- Compact from 726 to 354 lines (-51%) while maintaining clarity
- Restructure sections for better readability and navigation
- Add detailed workflow sequences and module explanations
- Update Key Components with enhanced modules architecture
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Corrections majeures:
- Digital Ocean: Récupération réelle XML depuis /wp-content/XML/ (86k chars au lieu de mock 1k)
- Nettoyage tags: Suppression <strong> dans extractElements() pour éviter parsing errors
- Doublons résilients: Tolérance doublons XML avec validation tags uniques
- Hiérarchie complète: StepExecutor génère 36 éléments depuis hierarchy.originalElement.name
- Format Google Sheets: Adaptation colonnes selon useVersionedSheet (17 legacy vs 21 versioned)
- Range Google Sheets: Force A1 avec INSERT_ROWS pour éviter décalage U:AO
- xmlTemplate optimisé: Exclusion du JSON metadata pour limite 50k chars
Résultat: 2151 mots, 36 éléments, sauvegarde correcte A-Q
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
- Add complete modular demo interface (public/modular-pipeline-demo.html)
- Add standalone demo server (simple-server.js) on port 3333
- Integrate real SelectiveCore, AdversarialCore, HumanSimulation, PatternBreaking modules
- Add configurable pipeline with step-by-step content transformation display
- Add new trend management and workflow configuration modules
- Add comprehensive test suite for full pipeline validation
- Update core modules for better modular integration and demo compatibility
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
- Remove Gemini configuration from LLMManager (geo-blocked)
- Fix elements.forEach error in ContentAssembly.js
- Update provider count from 6 to 5 in tests
- Add array validation in injectGeneratedContent
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
- Fix BatchProcessor constructor to avoid server blocking during startup
- Add comprehensive integration tests for all modular combinations
- Enhance CLAUDE.md documentation with new test commands
- Update SelectiveLayers configuration for better LLM allocation
- Add AutoReporter system for test automation
- Include production workflow validation tests
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
• AntiDetection_Level: Now shows actual adversarial mode instead of hardcoded "MVP"
• Parent_ID offset: Fixed calculation to match Google Sheets row numbers (removed -1)
• Pass adversarialMode parameter in all saveGeneratedArticleOrganic calls for proper tracking
Fixes:
- AntiDetection_Level column now displays: none, light, standard, heavy, adaptive
- Parent_ID now correctly corresponds to Google Sheets line numbers
- Consistent metadata across all versioning stages (v1.0 through v2.0)
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
• Created QueueProcessor base class for shared queue management, retry logic, and persistence
• Refactored BatchProcessor to extend QueueProcessor (385→142 lines, 63% reduction)
• Created BatchController with comprehensive API endpoints for batch operations
• Added Digital Ocean templates integration with caching
• Integrated batch endpoints into ManualServer with proper routing
• Fixed infinite recursion bug in queue status calculations
• Eliminated ~400 lines of duplicate code across processors
• Maintained backward compatibility with existing test interfaces
Architecture benefits:
- Single source of truth for queue processing logic
- Simplified maintenance and bug fixes
- Clear separation between AutoProcessor (production) and BatchProcessor (R&D)
- Extensible design for future processor types
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
- APIController.js: Full RESTful API with articles, projects, templates endpoints
- Real HTTP integration tests with live server validation
- Unit tests with proper mocking and error handling
- API documentation with examples and usage patterns
- Enhanced audit tool supporting HTML, npm scripts, dynamic imports
- Cleaned 28 dead files identified by enhanced audit analysis
- Google Sheets integration fully validated in test environment
- Fixed OpenAI API key hardcoding in BrainConfig.js causing 401 errors
- Implemented proper Digital Ocean Spaces integration with AWS SDK
- Added deployArticle() function for HTML upload
- Fixed fetchXMLFromDigitalOcean() to retrieve from correct path
- Direct access to root bucket instead of wp-content/XML/
- Added aws-sdk dependency for S3-compatible operations
- Created comprehensive integration test suite
- Validated complete workflow: Google Sheets → DO XML → Processing
All external systems now operational:
✅ Google Sheets auth and data retrieval
✅ Digital Ocean Spaces upload/download
✅ XML template processing
✅ LLM API authentication
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
Combines two parallel developments:
1. Remote: API setup base (326a6a5) with new /api/generate-simple endpoint
2. Local: Main.js/StepExecutor alignment for production consistency
Key technical improvements:
- Main.js: Migrated from generateSimple() to StepExecutor.executeInitialGeneration()
- StepExecutor: Fixed to respect HTML options (selectiveStack, adversarialMode, etc.)
* executeSelective() now uses applyPredefinedStack() with specific stack
* executeAdversarial() uses proper mode mapping (light→lightDefense, etc.)
* executeHumanSimulation() uses applyPredefinedSimulation() with mode
* executePatternBreaking() uses applyPatternBreakingStack() with mode
- Updated process_real.js and DigitalOceanWorkflow.js to use handleModularWorkflow()
- Removed legacy ManualTrigger.js (Apps Script compatibility)
This ensures 100% consistency between step-by-step debugging interface and
automated Main.js workflow, while preserving the new API functionality.
Both development paths now use identical, tested and validated functions.
🤖 Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <noreply@anthropic.com>
