mirror of
https://github.com/x1xhlol/system-prompts-and-models-of-ai-tools.git
synced 2026-06-18 15:29:36 +00:00
Phase 1-6 implementation for Dealix AI Revenue OS: - AI Arabic Engine: NLP (arabic_nlp.py), lead scoring (lead_scoring.py) - PDPL Compliance: consent manager, data rights handler, consent model - Sequence Engine: multi-channel sequences with WhatsApp/Email/SMS - CPQ System: quote engine, AI proposal generator - Security Gate: pre-release checks, PDPL message validation - Tool Verification: agent action audit trail - Project Operating Files: AGENTS.md, CLAUDE.md - Project Memory: architecture, ADRs, provider routing, PDPL checklist - Design System: IBM Plex Sans Arabic tokens, RTL-safe components - Sequence/Consent models for database https://claude.ai/code/session_01LsnvBa7HwF5hs99VZbgLGj
3.6 KiB
3.6 KiB
AGENTS.md — Dealix AI Revenue OS
Project Identity
- Name: Dealix (ديلكس)
- Type: AI-Powered CRM SaaS for Saudi Arabia
- Stack: FastAPI + Next.js 15 + PostgreSQL + Redis + Celery
- Market: Saudi SMBs (real estate, healthcare, retail, contracting, education)
- Language: Arabic-first, bilingual (AR/EN)
Architecture Boundaries
Backend (salesflow-saas/backend/)
- FastAPI 0.115.6 on Python 3.12
- SQLAlchemy 2.0 async with PostgreSQL 16
- Celery 5.x with Redis broker
- JWT authentication (python-jose)
- Multi-tenant data isolation via
tenant_id
Frontend (salesflow-saas/frontend/)
- Next.js 15 with App Router
- TypeScript 5.7, Tailwind CSS 3.4
- RTL-first layout (dir="rtl")
- Fonts: IBM Plex Sans Arabic (primary), Tajawal (secondary)
AI Layer (backend/app/services/ai/)
- LLM Provider: Groq (primary) → OpenAI (fallback)
- Arabic NLP with Saudi dialect support
- Model routing via
services/model_router.py
Agent System (backend/app/services/agents/)
- Manus-style orchestrator with 8 specialized roles
- Event-to-agent routing via
router.py - Executor with retry logic and escalation
Coding Conventions
- Python: async/await, type hints, Pydantic models, 4-space indent
- TypeScript: strict mode, functional components, Tailwind classes
- Database: all queries through SQLAlchemy ORM, never raw SQL
- API: RESTful, versioned (/api/v1/), proper HTTP status codes
- Naming: snake_case (Python), camelCase (TypeScript)
- Arabic: all user-facing strings must have Arabic versions
- Currency: SAR default, Numeric type for money fields
- Timezone: Asia/Riyadh (UTC+3)
Forbidden Actions
- Never hardcode API keys or secrets
- Never bypass tenant isolation
- Never send messages without PDPL consent check
- Never delete data without soft-delete first
- Never push directly to main branch
- Never skip security review for auth/payment changes
- Never use synchronous DB calls in async endpoints
- Never store PII in logs
Policy Classes
Class A — Auto-allowed
- Code reading and inspection
- Test generation and execution
- Documentation updates
- Memory/knowledge base updates
- Linting and formatting
- Architecture analysis
Class B — Approval Required
- Database migrations
- Customer-facing message sends
- Payment/billing changes
- Permission model changes
- External API integrations
- Production deployments
- PDPL consent configuration changes
Class C — Forbidden
- Secret exfiltration
- Bypassing branch protections
- Silent destructive changes
- Disabling security gates
- Cross-tenant data access
- Ungoverned bulk messaging
How to Install
cd salesflow-saas
cp .env.example .env # Configure your environment
docker-compose up -d
make migrate
make seed
How to Test
cd salesflow-saas/backend
pytest -v
# Or with coverage
pytest --cov=app --cov-report=html
How to Run
docker-compose up # All services
# Or individually:
cd backend && uvicorn app.main:app --reload --port 8000
cd frontend && npm run dev
Provider Preferences
- Fast classification: Groq (llama-3.1-70b)
- Arabic NLP: Groq with Arabic context prompts
- Sales copy/proposals: Claude (via model_router)
- Research/analysis: Gemini (via model_router)
- Coding tasks: DeepSeek (via model_router)
- Fallback: OpenAI GPT-4o-mini
Release Process
- Feature branch → PR → Code review
- Run tests + security scan
- Deploy to staging
- Smoke test (Arabic + English)
- Deploy to production with canary (10%)
- Monitor 30 min → full rollout
- Rollback plan documented per release