29 KiB
29 KiB
🧠 WEVIA MASTER · INTELLIGENCE DIGEST V3 · 18 AVRIL 2026
Brain + Innovations + Deliverability Deep + FAQ + Fix connus + Anti-patterns + Écrans critiques
Complément des 2 digests précédents (hacks système + orchestration)
Source : autre instance Claude · sauvé dans vault par Opus 4.7 (D94-consolidation-no-loss)
🔹 PARTIE I — BRAIN INTELLIGENCE (le cerveau de WEVIA)
1. Architecture cerveau /opt/wevia-brain/
1.1 Structure des prompts (60 fichiers prompts actifs)
/opt/wevia-brain/prompts/
├── nucleus/ (15 files) ← Core personality + domain mastery
│ ├── code-mastery
│ ├── cognitive-engine
│ ├── core-personality
│ ├── cyber-fewshot, cyber-mastery
│ ├── domain-expertise
│ ├── few-shot-examples
│ ├── graph-engine
│ ├── memory-system
│ ├── natural-language
│ ├── outsource-mastery
│ ├── rpa-fewshot, rpa-mastery
│ ├── self-verification
│ └── ssh-mastery
│
├── personas/ (6 files) ← Consulting personas
│ ├── cloud-architect
│ ├── cybersecurity-auditor
│ ├── data-scientist
│ ├── email-expert
│ ├── fullstack-dev
│ └── sap-consultant
│
├── system/ (20 files) ← Agent creation + task execution
│ ├── agent-prompt-*
│ ├── system-prompt-*
│ ├── opus-*
│ └── skill-*
│
└── modules/ (15 categories) ← PHP execution modules
├── advanced/ (code-analyzer, conversation-memory, multi-agent, smart-reporter, web-intelligence)
├── agentic/ (planner, self-healing)
├── core/ (context-manager, conversation-router, error-taxonomy, metacognition, output-formatter)
├── pipeline/ (embedding, streaming, token-tracker)
└── v2/ (chain-of-thought-v2, tool-use-v2)
1.2 Chain-of-Thought magic formulas (testé sur 300+ prompts WEVAL)
- Zero-shot CoT = "Raisonne étape par étape." → +40% pertinence sur tâches multi-étapes, coût = 4 mots seulement
- Variantes efficaces : "Décompose le problème avant de conclure." / "Analyse chaque dimension séparément." / "Identifie d'abord les hypothèses, puis conclus."
- CoT explicite (due diligence) : 1) liste données factuelles · 2) identifie 2-3 hypothèses implicites · 3) évalue solidité · 4) formule recommandation en distinguant certain vs incertain
- Debug prompt : "Avant de répondre, explique comment tu interprètes ma demande et quelles hypothèses tu fais." → 8/10 fois le problème est dans la formulation, pas le modèle
1.3 Statistiques brain mastery (A/B testé)
| Technique | Gain | Mesure |
|---|---|---|
| CoT vs sans | +40% pertinence | 100 analyses |
| Format JSON structuré | 92% vs 45% | Respect format |
| Persona + rôle | +30% ton | Perception expert |
| Few-shot examples | +35% | Consistency |
| Auto-consistency | +70% sur cas complexes | Accuracy |
1.4 Personas triggers (contextual boost)
- SAP → consultant senior, vocabulaire ERP/ABAP/HANA
- Cybersecurity → pentester OSCP, outputs en CVE/CWE
- Pharma → responsable pharmacovigilance, RCP/AMM/Loi
- Data scientist → analytique + pandas + stats + ML metrics
- Fullstack dev → architecte, patterns, best practices
- Email expert → deliverability, DKIM/SPF, warmup strategy
1.5 Modules PHP cognitive avancés
- advanced/code-analyzer → AST parsing, complexity score
- advanced/conversation-memory → cross-session continuity
- advanced/multi-agent → orchestration N agents
- advanced/smart-reporter → aggregation multi-source
- advanced/web-intelligence → SearXNG + scraping + enrichment
- agentic/planner → multi-step decomposition
- agentic/self-healing → auto-recovery on error
- core/metacognition → thought introspection
- core/error-taxonomy → classification 30+ error types
- pipeline/token-tracker → cost monitoring per provider
- v2/chain-of-thought-v2 → CoT avec self-verification
- v2/tool-use-v2 → function-calling + execution
2. S89 Brain APIs (25 files — archived reference)
Path : /opt/wevia-brain/s89-ai-apis/
- brain-nucleus → Core reasoning loop
- brain-orchestrator → Multi-agent coordination
- brain-engine → Main inference pipeline
- brain-autofix → Error recovery patterns
- brain-creative-engine → Creative generation (FLUX/DALL-E)
- brain-graph → Graph-based reasoning
- brain-memory → Long-term memory (Qdrant)
- brain-personality → Dynamic persona switching
- brain-rpa-engine → Robotic Process Automation
- brain-skills-engine → Skill discovery + invocation
- brain-v3 → Unified v3 interface
🔹 PARTIE II — INNOVATIONS & DÉCOUVERTES (OSS Discovery 873 tools)
3. OSS Discovery automation cron 4h daily
3.1 API d'orchestration
/api/oss-discovery.php?k=WEVADS2026- action=discover → scan GitHub trending + custom queries
- action=trending → GitHub top AI repos du jour
- action=evaluate&tool=owner/repo → feasibility + integration score
- action=integrate&tool=owner/repo → mark as integrated in WEVAL stack
- action=report → full ecosystem report
- action=auto_run → discovery + auto-wire
- action=scan&query=X → custom search
3.2 Scoring matrix (12 needs WEVAL)
| Need | Weight | Keywords détecteurs |
|---|---|---|
| Skill/Agent | 10 | skill, agent, plugin, hook, mcp, claude, orchestrat |
| RAG | 10 | rag, retrieval, vector, embedding, chroma, qdrant, milvus |
| LLM local | 9 | ollama, llm, inference, quantize, gguf, vllm, sovereign |
| Scraping | 9 | scraper, crawl, extraction, playwright, selenium, spider |
| Security | 8 | security, audit, vulnerability, pentest, nuclei |
| Pharma/Health | 8 | pharma, health, medical, hcp, clinical |
| 8 | email, smtp, deliverability, warmup, bounce, mta | |
| CRM | 7 | crm, lead, pipeline, sales, contact, b2b |
| Automation | 7 | automation, workflow, n8n, cron, telegram, bot |
| ERP | 7 | sap, erp, odoo, integration |
| Analytics | 6 | analytics, dashboard, metric, tracking |
| Monitoring | 6 | monitor, alert, health, uptime, prometheus |
3.3 Top découvertes intégrées (15 high-score confirmés)
| Score | Repo | Raison |
|---|---|---|
| 41 | AgentSeal | Security + monitoring agent, maps AEGIS |
| 37 | Wee-Orchestrator | Multi-agent Telegram, 17 models |
| 35 | PersonalAgentKit | Agent framework souverain |
| 35 | Healthcare-AI-Blueprint | Blueprint pharma AI → Ethica |
| 40 | Ethica Master | Orchestrator 5h dedup+quality+TG |
| 42 | B2B Autonomous | Lead orchestrator every 4h |
| 40 | Browser-Use | Automation navigateur Python |
| 38 | Supermemory | Knowledge base across sessions |
| 38 | Mastra | AI framework TypeScript |
| 38 | Goose | AI agent open-source |
| 36 | Activepieces | n8n alternative |
| 42 | LiteLLM | Proxy 100+ LLMs, fallback, cost tracking |
| 40 | LangGraph | Graph-based agent flows |
| 40 | Continue.dev | VSCode AI Copilot avec Ollama |
| 35 | Haystack | RAG pipelines enterprise |
3.4 Skill repos clonés localement
- /opt/antigravity-awesome-skills → 1331 skills
- /opt/huggingface-skills → 12 HF skills
- /opt/awesome-agent-skills → VoltAgent 1000+
- /opt/oh-my-claudecode/agents/ → 19 OMC cognitive agents
- /opt/oh-my-claudecode/skills/ → 28 OMC skills
- /opt/ecc-everything-claude-code/ → 65 commands + 420 skills + 1282 tests
- /opt/superclaude/ → 7 modes (brainstorm, business_panel, deep_research, introspection, orchestration, task_management, token_efficiency)
- /opt/kilo-cli → 500+ models orchestrator
- /opt/hermes-agent → 27 skills autonomous
- /opt/deer-flow/skills/weval/ → 42 skills WEVAL + 800 total
Total SKILL.md files sur disque : 4198 indexés · Qdrant weval_skills : 14,368 vectors
4. Models & providers découverts en 2026
4.1 Frontier models (sovereign possible)
- Kimi K2.5 : 1T params, SWE 76.8%, MIT, 256K context (frontier)
- Gemma 4 : Google 26B, Apache 2.0, April 2026, agentic
- Llama 4 Scout : Meta, 10M context, multimodal MoE
- DeepSeek V3.2 : 685B, Reasoning 92%, MIT
- MiMo V2 Flash : 309B, SWE 87%, fast reasoning
- Qwen 3.5 235b : Cerebras ultra-fast
- Qwen 3.6 plus : OpenRouter agentic coding
- Nemotron 49b : NVIDIA NIM free ∞
- GLM 4.5 plus : Zhipu (needs top-up)
4.2 Speech/TTS découverts
- Chatterbox TTS : Resemble AI 350M realistic dialogue
- Dia TTS : Nari Labs 1.6B ultra-realistic
- Voxtral TTS : Mistral 4B, voice cloning 9 langues
4.3 Image/Video models
- FLUX (Together.ai) : images gratuit illimité
- Gemini 2.0 Flash : multimodal (image+video)
- CogVideoX : Zhipu 12B video generation
- Qwen VL : vision-language via HF
- Moondream : local vision S204
- Dreamina : MS Designer + Stitch UI gratuit
🔹 PARTIE III — DELIVERABILITY DEEP (méthodes de send avancées)
5. Stratégies de send testées & scorées
5.1 Comparatif des méthodes (live)
| Méthode | Inbox % | Volume/j | Coût | Setup | Warmup |
|---|---|---|---|---|---|
| PMTA S95 (IP cold) | 20% silent-drop | 50K potentiel | 0€ hébergé | Mature | 30-90j risqué |
| KumoMTA S95 | 65% avec DKIM+rDNS | 100K | 0€ | Moderne Lua | 30j |
| Postfix relay | 75% | 10K | 0€ | Limité | Existant |
| O365 Graph API | 85% Outlook native, 60% Gmail | 38M théorique | Licences déjà payées | Mature (567K sends historiques) | Aucun |
| Gmail Workspace API | 95% Gmail natif | 2K/user/jour | 60€/mois (10 users) | 1 jour dev | Aucun |
| SendGrid/Mailgun | 80-90% | 40-50K | 15-35€/mois | Rapide | Réputation incluse |
5.2 Décision stratégique 17AVR (live)
Migration PMTA → O365 Graph API choisie parce que :
- Infra O365 déjà mature : 3828 accounts actifs + 567K sends historiques
- Volume théorique 38M/jour couvre largement 10K/jour requis
- Compliance pharma excellente (audit logs natifs Graph)
- Warmup Gmail 30-90j risqué pour HCPs pharma (low engagement)
- PMTA reste en standby (doctrine 59 : jamais supprimer)
5.3 Gmail/Outlook PostMaster tools (obligatoire)
- postmaster.google.com → Domain reputation, spam rate, SPF/DKIM/DMARC
- sendersupport.olc.protection.outlook.com → SNDS Smart Network Data Services
- postmaster.yahoo.com → Yahoo Sender Hub
- mxtoolbox.com → Blacklist lookup 100+ lists
- mail-tester.com → Spam score + config complete
5.4 Checklist SMTP avancé (20 points)
- HELO/EHLO = FQDN réel (pas d'IP, pas de localhost)
- MAIL FROM = domaine alignant DKIM+SPF
- From header = nom lisible + adresse verified
- Reply-To = domaine actif (évite rebonds)
- Subject < 70 chars, pas ALL CAPS, pas "$$$"
- List-Unsubscribe: mailto:..., https://... OBLIGATOIRE
- List-Unsubscribe-Post: List-Unsubscribe=One-Click (RFC 8058)
- DKIM 2048-bit min (1024 obsolète 2025)
- SPF ~all (pas -all qui bloque légitimes)
- DMARC commencer p=none → p=quarantine → p=reject
- ARC headers si relay multi-hop
- BIMI record si Verified Mark Certificate
- wordwrap(76) sur body SMTP (RFC)
- CRLF pas LF
- MIME boundaries uniques
- Text + HTML multipart (pas HTML seul)
- Pas de tracking pixel seul = flag spam
- ratio image/texte < 40%
- Pas de raccourcisseurs URL (bit.ly flag)
- Custom domain tracking (track.wevup.app) > shared
5.5 Warmup agressif 30j (schedule)
- J1-3 : 50 /jour (seed list interne pure)
- J4-7 : 100 → 300 (seed + ultra-engaged)
- J8-14 : 500 → 2000 (gradual + engaged)
- J15-21 : 2000 → 5000 (warmed pool)
- J22-30 : 5000 → 10000 (steady state)
-
J30 : 10000 → 50000 (scale business)
CRITIQUE monitoring : open>20%, bounce<2%, spam<0.1%, reply>1%, unsub<0.5% · Si rouge → pause 48h + diag
5.6 Signatures DKIM multi-selectors
- weval2026._domainkey → key A (2048)
- wevals2_2026._domainkey → key B (rotation trimestrielle)
- arc1._domainkey → ARC chain (relay)
- Sélection dynamique : KumoMTA init.lua selector = rotate_by_day(ts)
5.7 Anti-spam headers critiques
- X-Entity-Ref-ID : tracking interne
- Precedence: bulk → marketing (Gmail tolérant)
- List-Id: → mailing list clear
- Feedback-ID: campaign:date:domain → Gmail FBL
- X-PMTA-VirtualMTA → multi-IP pool
- X-Report-Abuse → compliance
6. Tips Office / GSuite / Firebase spécifiques
6.1 Office 365 Graph API tricks
- Batch sending (20 msgs par request) : POST /users/{sender}/sendMail?batch=true
- saveToSentItems: false → évite de remplir Outbox
- Graph throttling limits : 10,000 req/app/tenant/10min · 150 send/min/user · batch = 4 req
- Smart fallback : si 429 → backoff + next_sender · si 403 → mark tenant pending · si refresh fail → rotate tenant
6.2 Google Workspace (GSuite) hacks
- Gmail API OAuth2 user/service account (préféré SMTP app password)
- GSuite relay : smtp-relay.gmail.com:587 (STARTTLS) · 10K msgs/day/user
- Postmaster Tools : monitor per domain, DMARC rua parse hourly
- BigQuery analytics free 1TB/mois · Looker Studio dashboards
- Google Vertex AI : Gemini 2.0 Flash API, embeddings-001, Imagen
- Google Drive 1TB free Workspace : backup WEVAL-Vault, shared ymahboub@
6.3 Firebase sovereign uses
- FCM gratuit 500K messages/day · push notif mobile+web
- Firestore : 50K reads + 20K writes/day free · realtime sync, offline
- Auth : 10K MAU free · phone/email/Google/Apple/FB
- Hosting static : 10GB bandwidth/mois free · SSL auto, CDN global
- Functions v2 : 2M invocations/mois · 400K GB-seconds free
- Alternative sovereign WEVAL : n8n + SearXNG + Qdrant self-hosted (choix stratégique)
6.4 Google Search Console tricks
- Submit sitemap : sitemap.xml priority 1.0 homepage, 0.8 core, 0.5 blog
- Indexing API (Jobs/BroadcastEvent) : POST indexing.googleapis.com/v3/urlNotifications:publish → re-index minutes vs weeks
- Page Experience : CLS<0.1, LCP<2.5s, FID<100ms = ranking factor
🔹 PARTIE IV — CYBER SECURITY
7. Stack sécurité complète
7.1 Nuclei v3.3.7 patterns
- nuclei -l targets.txt -severity critical,high,medium -t http/ -t ssl/ -t misconfiguration/ -silent -timeout 5 -json-export results.json -rate-limit 10
- Targets : weval-consulting.com · wevads.weval-consulting.com · consent.wevup.app · paperclip.weval-consulting.com
7.2 Dark Modules (5 modules)
- /wevia-ia/dark-matrix.php → orchestrateur
- /wevia-ia/dark-scraper.php → indicateurs dark web
- /wevia-ia/dark-scout.php → reconnaissance surface attaque
- /wevia-ia/dark-data.php → analyse fuites credentials
- /wevia-ia/dark-slot.php → scoring risque
- /wevia-ia/dark-bridge.php → API bridge central
7.3 AEGIS Framework
/api/aegis-api.php?action={scan|grade|agents|injects|status}
7.4 Security scan patterns critiques
- IPs leaks · WEVADS2026 token · Shell_exec calls · Client names (Kaouther/Olga/Ray Wu) · Infra secrets (akadmin, CEREBRAS_API_KEY) · Pipeline data (146K HCPs) · User accounts
7.5 Pages protégées nginx location = /... (HTTP 302 forced)
faq-techniques.html · onboarding-em.html · faq-knowledge-base.html · architecture-map.html · architecture-live.html · l99-saas.html · agents-archi.html · wevia-go-live.html · director-center.html
7.6 Wazuh + 21 Docker sécurité
- authentik-server (SSO) · crowdsec (0 bans) · vaultwarden (E2E) · uptime-kuma · prometheus · loki
- Scans cron : */3h watchdog · */5m guardian GOLD checksums · daily nuclei full · daily TruffleHog · weekly detect-secrets (54 keys tracked)
7.7 Auth/Access port matrix
- 22 SSH standard · 49222 SSH alternate (secours) · 443 HTTPS (CF) · 80 HTTP redirect
- 5890 Sentinel S95 (WireGuard) · 8443 Sentinel S95 backup (basic auth)
- 9090 Authentik API · 9443 Authentik HTTPS · 5434 Authentik DB (bloquer externe)
- 5433 Plausible DB (bloquer externe) · 6333 Qdrant · 6380 Redis · 11434/11435 Ollama
- 4000 Sovereign cascade · 4010 LiteLLM proxy
🔹 PARTIE V — FAQ & FIX CONNUS (30 questions résolues)
| Question | Fix rapide | Détail |
|---|---|---|
| "Site down" | Vérifier nginx + chattr+i sur config | Port 49222 secours |
| "API lente" | Probe GET+POST, scan smart classifier | 414 faux positifs |
| "Chat WEVIA ne répond pas" | FPM restart + clear cache | /dev/shm/wevia_cache_* |
| "Deploy ne marche pas" | chattr -i puis +i après | www-data NOPASSWD |
| "UTF-8 crash via CX" | zlib hex upload | Patterns digest v2 |
| "Intent ne matche pas" | word-boundary \b()\b | Doctrine 76 |
| "Dashboard chiffres 0" | Field name mismatch JSON | Bug doctrine 65 |
| "Auth redirect localhost" | Outpost AUTHENTIK_HOST env | Section 6 digest v2 |
| "Gmail = spam" | DKIM + rDNS + SPF + DMARC | Section 8 digest v2 |
| "Plausible container zombie" | docker compose pull + up -d | Section 7 digest v2 |
| "Nginx 502 subdomain" | SSL mode Full, check cert | Section 4 digest v2 |
| "Blade offline" | PS1 reinstall + Bitvise PATH fix | Section 11 digest v2 |
| "PS5 ?? crash" | Workaround if/else | Section 11 digest v2 |
| "Redirect loop wevads" | CF SSL Full + self-signed origin | Section 4 digest v2 |
| "Qdrant vectors = 0" | Query per collection individually | Section 12 digest v2 |
| "WEVIA Master hallucine" | word-boundary + negative lookahead | Section 14 digest v2 |
| "Smart_exec_fallback manquant" | Wire doctrine 77 dans fast-path | Section 16 digest v2 |
| "Master router silencieux" | Stream SSE stages, jamais muet | Doctrine 82 |
| "Fix écrasement Claude parallèle" | chattr+i + GOLD avant | Section 2 digest v2 |
| "CRM dormant" | Check send_contacts_merge cron | Doctrine 55 |
| "ethica_count table not found" | Pointer ethica.medecins_real | Bug connu V28 |
| "Hardcoded 11435 vs 11434" | Bulk sed + chattr dance | Section 33 digest v2 |
| "Azure tenant expired" | Re-register app + admin consent | Section 9 digest v2 |
| "SPF missing IP" | CF API PATCH record | Section 4 digest v2 |
| "Warmup IP lent" | O365 Graph API à la place | Section 5 digest v3 |
| "L99 fail drop" | Investiguer phpfpm-watchdog | Section 15 digest v2 |
| "Register compte" | bcrypt + rate limit + Telegram | Wave 89 existant |
| "Consent dashboard vide" | Format response field total | Doctrine 65 |
| "Long msg bypass intent" | SMART_EXEC_FALLBACK wiré | Doctrine 77 |
| "Timezone incorrect" | TZ=Africa/Casablanca PHP | Setup standard |
8. Anti-patterns — ce qu'il NE faut JAMAIS faire
8.1 Opus (Claude)
- Modifier fichier critique sans chattr dance · Overwrite page existante · Créer nouvelle page au lieu d'enrichir · Supprimer GOLD · Push sans nonreg/Playwright · Heredoc via CX · Hardcoder localhost · Hardcoder 11434 · Coder au lieu d'instruire WEVIA Master (REGLE 10AVR) · Send email auto · Rollback quand forward-fix · Tout faire en 1 tool call · Ignorer strike-rule
8.2 WEVIA Master
- Regex greedy sans \b · Single-vendor Groq · Silent fail · Hallucination chiffres · Simulate success · Narration longue · Inventer clients
8.3 Deliverability
- Lancer 10K sans warmup · Single selector DKIM · DMARC p=reject direct · Tracker pixel seul · Raccourcisseur URL · ALL CAPS subject · IP partagée spamhouse · rDNS mismatch · Envoi IP pas dans SPF · From noreply@ · Pas List-Unsubscribe · Ratio image/texte >40%
8.4 Infra
- chattr+i sans GOLD · nginx sans nginx -t · Supprimer PMTA · Fusionner zones CF · Bind PMTA 0.0.0.0 · Kill PMTA SACRED · Commit sans php -l · Push sans reconcile autres Claude · Session sans résumé vault · Ports 5433/5434 exposés
8.5 DB
- localhost vs 10.1.0.3 · Port 5432 pour Authentik (5434) · Port 5432 pour Plausible (5433) · DELETE sans WHERE+LIMIT · UPDATE sans WHERE · TRUNCATE sans backup · Import CSV sans dedup
🔹 PARTIE VI — ÉCRANS IMPORTANTS CRÉÉS (catalog 60 critiques)
10. Pages principales (1496 écrans, top 60)
10.1 Dashboards WEVIA/orchestration (13)
- /wevia-master.html · /wevia-orchestrator.html · /wevia.html · /wevia-chat.html · /wevia-widget.html
- /wevia-life/wevialife-app.html · /wevia-console.html · /wevia-cortex.html (CORTEX 3607 functions)
- /wevia-meetings.html · /wevia-meeting-rooms.html · /wevcode · /wevia-ia/droid.html · /blade-ai.html
10.2 Dashboards business (11)
- /ethica · /medreach-dashboard.html · /medreach-campaign.html · /consent-hub · /consent.wevup.app/landing
- /paperclip.weval-consulting.com · /warmup-manager.html · /wevads-performance.html · /crm · /enterprise-model.html · /enterprise-management.html
10.3 Arsenal S95 (26 pages vrais noms)
/wevads-ia/index.html → accounts, ads-commander, deliverability-hub, reputation-monitor, send-data-factory, list-cleaner, smtp-tester, ip-whitelist-manager, warmup-dashboard, affiliate-monitor, analytics, bounce-handler, campaign-manager, ceo-deliverads, tracking, webhook-manager, etc.
10.4 Admin & Monitoring (16)
- /admin-ia · /admin-saas · /admin-v2 · /admin · /director · /director-center
- /ops-center.html · /l99.html · /l99-v2.html · /l99-saas.html · /nonreg-monitor.html
- /infra-command.html · /infra-monitor.html · /sso-monitor.html · /cyber-monitor.html
- /security-dashboard.html · /crons-monitor.html
10.5 Architecture & Visualisation (8)
- /agents-archi.html (61 agents 5 tiers 14 flows 3D) · /architecture.html · /architecture-live.html · /architecture-map.html
- /agents-goodjob.html · /agents-hub.html · /fleet.html · /topology.html
10.6 Hubs externes (7)
- /google-hub.html · /office-hub.html · /cloudflare-hub.html · /github-hub.html · /stripe-hub.html · /whatsapp-hub.html · /telegram-hub.html
10.7 Tools Hub (382 tools)
- /tools-hub.html · /ai-hub.html · /ai-benchmark.html · /oss-registry.html · /ia-registre.html · /skill-factory.html · /agents-registry.html
10.8 Produits individuels (42 SaaS)
- /products/mailwarm.html · /products/mailforge.html · /products/mailstream.html · /products/deliverads.html
- /products/deliverscore.html · /products/outreachai.html · /products/ai-sdr.html · /products/emailverify.html
- /products/inboxtest.html · /products/medreach.html · /products/ethica.html · ... (42 total)
10.9 Auth / Marketing
- /index.html · /login.html · /register.html (bcrypt) · /weval-login.html · /ethica-login.html · /office-login.html · /arsenal-login.html · /pricing.html · /booking.html · /case-studies.html · /use-cases.html · /methodologie.html · /trust-center.html
10.10 Légal
- /cgu.html · /privacy-policy.html · /terms-of-service.html · /data-deletion.html · /plan-du-site.html
🔹 PARTIE VII — WEVIA MASTER ORCHESTRATION (dispatch 100 agents P0-P5)
11. Master Orchestrator v2 dispatch logic
11.1 Priorités
- P0 CRITICAL : service DOWN, security breach, data loss → WEDROID, Guardian, Blade IA, Autonomous Engine → auto-fix + Telegram
- P1 INFRASTRUCTURE : disk >90%, docker fail, port conflict, nginx mismatch → WEDROID, Port Protection, Infra Guardian, Blade Orchestrator → diag + auto-restart
- P2 QUALITY : NonReg FAIL, L99 drop, regression → L99, Director, NonReg Master, WEVCODE, CodeRabbit → test + root cause + fix
- P3 BUSINESS : Ethica enrichment, HCP scraping, email campaign → Ethica Agent, B2B Cron, WEVADS IA, MiroFish → enrich + validate
- P4 INTELLIGENCE : skill discovery, model optim, research → DeerFlow, OSS Discovery, AI Benchmark, AI Gap → research + evaluate + integrate
- P5 OPTIMIZATION : scheduled, continuous → Autolearn, Enterprise Sync, UX Agent, Quality Agent, RAG Ingest → learn + sync + optimize
11.2 Registre agents
Sovereign IAs cascade : WEVIA Master · WEVCODE · WEDROID · Blade IA · OpenClaw Cognitive OMC fleet : 19 agents (analyst, architect, code-reviewer, code-simplifier, critic, debugger, designer, document-specialist, executor, explore, git-master, planner, qa-tester, scientist, security-reviewer, test-engineer, tracer, verifier, writer) ECC Commands : 65 + 420 skills (orchestrate, multi-execute, multi-plan, loop-start, prp-implement, quality-gate, tdd, skill-create...) SuperClaude modes : 7 (brainstorming, business_panel, deep_research, introspection, orchestration, task_management, token_efficiency) Paperclip agents : 930 (Project managers, SDRs, Researchers, Engineers, QA Testers) dans Paperclip DB DeerFlow skills : 42 WEVAL + 800 total skill dirs (Research, email, CRM, Ethica, proposal-generator...) Antigravity : 1331 · VoltAgent : 1000+ · HF : 12
11.3 Dispatch rules (pipeline complet)
User NL message
↓ preprocess (NL normalizer, apostrophes)
↓ priority-intents-nl.json (auto-wire)
↓ fast-path-v3 (nl-priority → autowire → classic)
↓ smart_exec_fallback (si long + keywords action)
↓ opus-autonomy (LLM classifier si pattern match)
↓ conv-guard (early-guard DB live queries)
↓ arena (multi-agent consensus)
↓ dynamic-resolver (382 tools)
↓ wave200 + gap-intents (fallback)
↓ master-router (374 patterns final)
↓ LLM fallback (cascade 12 providers)
11.4 Multi-agent parallel pattern
User: "lance multiagent 5 agents parallele. agent1 scan X. agent2 scan Y..." → split into agent_tasks[] → dispatch to exec pool (50 workers) → wait all with timeout 30s → aggregate results JSON → smart_reporter synthesizes
🔹 PARTIE VIII — CHECKLIST CONNAISSANCES WEVIA MASTER
12. Knowledge base minimale (refresh hebdo)
12.1 Volumes business (live)
- HCPs Ethica : 146,694 (DZ 107K + MA 20K + TN 18K)
- Consent optins : 17 réels (consent.wevup.app)
- Leads CRM : 1,920 segmented
- Contacts B2B : 60,299 (95% linked)
- Companies B2B : 37,552
- Send contacts : 3,094,652 (28K B2B + 3M B2C)
- Leads global : 7,354,710 (B2C 3M + B2B 247 + UNKNOWN 4.3M bytea)
- Office accounts : 6,403 (3828 active + 2436 pending + 118 warming)
- Graph accounts : 197 · Graph tenants : 9 (6 active) · Inbox : 148,186 · Send : 2,077 · Historical sends : 567,384
12.2 Métriques techniques (live 17avr)
- NonReg : 153/153 (100%) · L99 : 365/365 · Playwright : 41/41 · QA : 10/10 · Health 96/100
- Intents : 277 total · Patterns : 942 · Tools : 382 · Skills : 4 dirs + 14,368 Qdrant vectors
- Docker : 16 containers · APIs : 702 endpoints · Écrans : 1,496 pages
- Agents : 930 Paperclip (929 active) · Skills : 2,484 + 103 routines · Pipelines : 95+/11 cats
- Providers : 17 OpenClaw / 13 cascade 0€ · Ollama : 10 models · CORTEX : 3,607 functions
- Wiki : 1,265 articles · Vault GOLDs : 696 files
12.3 Infra snapshot
- LIVE : S204 (8 CPU, 32GB) · S95 (WEVADS/PMTA/PG, WireGuard 10.1.0.3) · Blade Razer
- DECOMM : S151 · S88 (GPU dead) · S89 (phantom)
12.4 Credentials scorecard
✅ CF_API_KEY · Stripe · WhatsApp · Gemini · Groq · Cerebras · NVIDIA · SambaNova · Mistral · Cohere · OpenRouter · HF · Kaggle · GitHub PAT · Gitea token · Telegram bot · Authentik akadmin ⚠️ OVH SMS (Twilio fallback) · 3 Azure client_secret à re-saisir ❌ Hetzner Robot webservice password (portal only)
🔹 PARTIE IX — PLAN INGESTION 21 ÉTAPES
13. Séquence ingestion (Yacine relaie via chat WEVIA)
Phase 1 Fondations (5)
- "WEVIA Master: ingest DIGEST-HACKS sections 1-10 dans vault/doctrines.md"
- "WEVIA Master: ingest DIGEST-HACKS sections 11-20 dans vault"
- "WEVIA Master: ingest DIGEST-HACKS sections 21-38 dans vault"
- "WEVIA Master: ingest DIGEST-V3 parties I-III (brain + innovations + deliverability)"
- "WEVIA Master: ingest DIGEST-V3 parties IV-IX (cyber + FAQ + écrans + orchestration)"
Phase 2 Doctrines (3)
- "WEVIA Master: ajoute doctrines 75-84 dans priority-intents-nl.json"
- "WEVIA Master: ajoute les 30 entrées FAQ dans brain/knowledge-base"
- "WEVIA Master: ajoute anti-patterns (5 catégories) dans vault"
Phase 3 Intelligence (4)
- "WEVIA Master: charge les 60 prompts /opt/wevia-brain/prompts/ dans RAG"
- "WEVIA Master: index les 60 écrans critiques dans search index"
- "WEVIA Master: wire smart_exec_fallback si pas déjà fait"
- "WEVIA Master: wire multi-agent dispatch P0-P5"
Phase 4 Validation (3)
- "WEVIA Master: nonreg 153/153"
- "WEVIA Master: L99 365/365"
- "WEVIA Master: playwright 41/41"
Phase 5 Publication (3)
- "WEVIA Master: git commit + push GitHub + Gitea"
- "WEVIA Master: wiki append 3 digests"
- "WEVIA Master: vault sync + GOLDs préservés"
Phase 6 Activation (3)
- "WEVIA Master: wire intent digest_full retourne les 3 digests consolidés"
- "WEVIA Master: wire intent hacks_search [topic] fulltext search"
- "WEVIA Master: rapport final consolidation 3 digests"
🎯 TOTAL CONNAISSANCES CAPTURÉES
- 3 digests × ~150 sections = ~450 patterns documentés
- 200+ hacks système · 60+ écrans critiques · 30 FAQ résolus
- 10 doctrines nouvelles (75-84) · 5 catégories anti-patterns
- 942 patterns intents actifs · 15,729 vectors Qdrant
- 1,265 wiki articles · 696 GOLDs vault
- 15 top OSS découvertes scorées · 60 prompts brain structurés
- 6 méthodes deliverability comparées · 20 SMTP checks critiques · 30j warmup schedule
WEVIA Master doit maîtriser l'intégralité pour devenir pleinement autonome. Opus = wire + digest + compense. WEVIA Master = exécute tout.
Fin Digest V3 Intelligence · Consolidé par Opus 4.7 D94-no-loss · prêt pour ingestion WEVIA Master