Files
html/wiki/doctrine-135-nl-audit-intent.md
Opus 42f7ab9e8d
Some checks failed
WEVAL NonReg / nonreg (push) Has been cancelled
feat(v185-phase1-nl-audit-intent): NL audit 6 handlers reels zero LLM
PHASE 1 - Combler manque autonomie (PHASE 0 avait 0/6 questions NL resolues).

Ajouts:
- /var/www/html/api/nl-audit-handler.php (247L, 6 sous-handlers)
  * handler_baseline: NR+L99+commits+tags+doctrines
  * handler_kpi_source: reutilise wtp-kpi-global-v2.php (source unique)
  * handler_dashboards_overlap: scan 7 hubs fetch count
  * handler_orphan_pages: 330 scanned / 208 orphelines
  * handler_autowire_capabilities: gap analysis
  * handler_tips_library: corpus 400 .md, 10 categories
- chat-v2-direct.php: dispatch nl_audit avant master-api fallback
- doctrine 135 (3 emplacements)

Verification:
- 6/6 questions PHASE 0 resolues (test via curl direct)
- Tous handlers < 100ms
- NR 153/153 invariant
- L99 340/340 invariant
- Zero regression intents existants (hi/nonreg/ethica/intents_pool verifies)

GOLD: vault-gold/opus/phase05-20260423-010548/wevia-chat-v2-direct-pre-phase1.php
Gap identifies PHASE 3: Claude.ai/Gemini/Qwen tips manquants, execute_tip intent absent
2026-04-23 01:14:56 +02:00

1.7 KiB

Doctrine 135 - NL Audit Intent (PHASE 1)

Date: 2026-04-23 Trigger: PHASE 0 audit autonomie WEVIA Master - 0/6 questions complexes resolues (LLM fallback timeout 120s)

Principe

WEVIA Master doit savoir repondre aux questions introspectives (audit de soi, KPI sources, orphelins, doublons, autonomie, tips) SANS recours a un LLM externe. Source de verite locale uniquement.

Architecture nl_audit

  • Endpoint: /var/www/html/api/nl-audit-handler.php (247L, PHP 8.5)
  • Dispatcher: nl_audit_route($msg) - regex keyword routing
  • 6 sous-handlers EXEC:
    1. handler_baseline() - NR + L99 + commits + tags + doctrines recentes
    2. handler_kpi_source() - Reutilise /api/wtp-kpi-global-v2.php (source unique deja construite 21avr)
    3. handler_dashboards_overlap() - Scan 7 hubs fetch count + api refs
    4. handler_orphan_pages() - 330 .html scannees, 208 orphelines detectees
    5. handler_autowire_capabilities() - Gap analysis + TODO PHASE 3
    6. handler_tips_library() - Corpus 400 .md scanne, 10 categories (6 OK 4 MISS)

Integration

chat-v2-direct.php: nl_audit dispatch avant fallback master-api. Match regex large mais post tous intents precablees (pas de regression hi/nonreg/ethica).

Performance

  • Tous handlers < 100ms (preg_match_all sur corpus static = rapide)
  • 6/6 questions PHASE 0 resolues live (vs 0/6 avant)
  • NR invariant 153/153
  • L99 340/340

Gaps identifies pour PHASE 3

  • execute_tip($category) intent manquant (Selenium Chrome via Blade)
  • auto_wire($trigger, $desc, $cmd) existe deja (user-wired-intents.json)
  • clone_github_tool($url) intent manquant (clone + adapt OSS tool)
  • Memoire 7 chatbots internes non unifiee (ethica-chatbot, claw-chat, etc.)