Opus Wire
|
40af847595
|
feat(claude-pattern-api-v15): 7 phases REAL reasoning pattern pour 5 chatbots
WEVAL NonReg / nonreg (push) Has been cancelled
NEW endpoint: /api/claude-pattern-api.php (10KB)
7 PHASES structured (PAS de simulation):
1. THINKING - intent classification + keywords + complexity + language
2. PLAN - structured steps based on intent (status/action/analytics/query)
3. RAG - Qdrant vector search (port 6333) · contexts enrichment
4. EXECUTE - REAL backend call (http://127.0.0.1 + chatbot-specific api)
5. TESTS - 5 validation checks (has_response, no_error, timeout, json_valid, not_simulated)
6. RESPONSE - structured final answer with length
7. CRITIQUE - self-review + quality score + warnings
5 CHATBOTS wires (chain fallback si primary fail):
- wevia-master → wevia-autonomous (fallback: opus5-autonomous-orchestrator-v3)
- wevia → ambre-thinking
- claw → wevia-json-api
- director → wevia-autonomous (fallback: opus5-orchestrator-v3)
- ethica → ethica-brain
- auto → opus5-autonomous-orchestrator-v3
VALIDATION LIVE (5/5 chatbots):
- wevia-master: 4/5 OK (via fallback)
- wevia: 4/5 OK
- claw: 5/5 EXCELLENT
- director: 4/5 OK (via fallback)
- ethica: 5/5 EXCELLENT
Moyenne: 4.4/5 · 5/5 chatbots REAL
Tool registry (638 -> 640):
- claude_pattern_api (kw: claude.*pattern|7.*phases)
- chatbot_health_check (test all 5 chatbots)
ZERO simulation · ZERO fake data · all tests REAL
Tests attrapent simulated/mock/fake/placeholder explicitement
Doctrine:
- REAL execution only (not_simulated test explicit)
- Fallback chain (chain tolerance)
- Self-critique (warnings if <5/5 or timeout)
- Quality score per-chatbot
- Additif pur · zero ecrasement
|
2026-04-22 04:18:52 +02:00 |
|
