mancitrus/system-prompts-and-models-of-ai-tools
1
0
Fork 0
mirror of https://github.com/x1xhlol/system-prompts-and-models-of-ai-tools.git synced 2026-06-18 07:19:35 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity
1450cfa2c8
system-prompts-and-models-o.../salesflow-saas/MASTER_OPERATING_PROMPT.md
Claude 7a8c572f71
fix(dealix): eliminate ALL stubs found by brutal audit 
Audit finding 1 — Saudi consent was hardcoded True:
  _check_consent() now queries real PDPLConsent table.
  Returns consent_valid=True only if active consents exist or tenant
  has no records yet (new tenant grace). Otherwise blocks.

Audit finding 2 — Saudi export rules were hardcoded True:
  _check_export_rules() now enforces: restricted data with
  requires_dpo_review=True blocks export by default.
  Returns blocked_reason_ar explaining why.

Audit finding 3 — MASTER_OPERATING_PROMPT overclaimed:
  Rule 6 said "controls are live, not aspirational" which
  contradicted current-vs-target-register showing 52% maturity.
  Rewritten to accurately describe: enforcement is live on golden
  path and Saudi workflow, full coverage tracked in register.

Audit finding 4 — forecast accuracy_trend was empty stub:
  Now queries real Deal table: closed_won vs total pipeline,
  returns actual accuracy percentage.

Post-fix audit status:
  - Saudi consent: REAL (queries PDPLConsent)
  - Saudi export: REAL (enforces classification)
  - MASTER_OPERATING_PROMPT: NO OVERCLAIM
  - Forecast accuracy: REAL (queries deals)

https://claude.ai/code/session_01W1rJthWDkasijTdXCfxVHs
2026-04-17 06:43:16 +00:00

9.1 KiB
Raw Blame History

MASTER OPERATING PROMPT — Dealix Sovereign Enterprise Growth OS

Version: 1.0
Status: Canonical
Effective: 2026-04-16
Scope: All agents, services, documents, and humans operating within Dealix

1. Identity

Dealix is a Sovereign Enterprise Growth OS for GCC Companies.

It is a single platform that manages:

  • Revenue — lead-to-cash lifecycle
  • Partnerships — alliance scouting to co-sell
  • Corporate Development / M&A — target sourcing to PMI
  • Expansion — market scanning to post-launch
  • PMI / Strategic PMO — Day-1 readiness to synergy realization
  • Trust / Governance / Executive Decisioning — policy gates to board packs

Central Law:

AI explores, analyzes, and proposes. Systems execute. Humans approve critical decisions. Everything is proven by evidence.

Design Philosophy:

Agentic by design, governed by policy, proven by evidence.

2. Five-Plane Architecture

Every component in Dealix belongs to exactly one plane:

Plane Purpose Key Code
Decision Strategic reasoning, forecasting, memo generation executive_roi_service.py, analytics_service.py, management agents
Execution Durable workflows, task routing, agent dispatch openclaw/gateway.py, durable_flow.py, task_router.py, Celery workers
Trust Policy enforcement, approval gates, audit, compliance policy.py, approval_bridge.py, hooks.py, pdpl/, audit_service.py
Data Storage, retrieval, enrichment, vector search, events PostgreSQL + pgvector, Redis, knowledge_service.py, domain events
Operating Monitoring, self-improvement, deployment, CI/CD observability.py, self_improvement.py, feature_flags.py, GitHub Actions

Full specification: docs/ai-operating-model.md

3. Six Tracks

All work is organized into six strategic tracks:

Track Domain Owner Focus
Revenue Lead capture → qualification → deal → close → renewal Sales & Growth
Intelligence Signal detection, behavior analysis, forecasting, AI agents AI & Data
Compliance PDPL, ZATCA, SDAIA, sector regulations, audit trails Legal & Security
Expansion Strategic deals, M&A, partnerships, geographic expansion Corporate Dev
Operations Deployment, monitoring, connectors, infrastructure Engineering & Ops
Trust Policy gates, approval SLAs, evidence packs, contradiction detection Governance

Full specification: docs/dealix-six-tracks.md

4. Policy Classes

Every action in the system is classified:

Class Behavior Examples
A — Auto-allowed Execute without approval read_status, classify, summarize, research, generate_draft
B — Approval-gated Requires human approval token send_whatsapp, send_email, create_charge, sync_salesforce, send_contract_for_signature
C — Forbidden Blocked unconditionally exfiltrate_secrets, delete_data_without_audit, bypass_auth

Implementation: backend/app/openclaw/policy.py

Default rule: Unknown actions are classified as Class B (approval required).

5. Execution Principles

  1. Decision-native — Every critical path produces structured output (JSON Schema), not free text.
  2. Execution-durable — Workflows checkpoint, resume after failure, and support compensation.
  3. Trust-enforced — No sensitive action bypasses the policy gate.
  4. Data-governed — All data flows through governed ingestion with quality checks.
  5. Arabic-first — All user-facing content defaults to Arabic, with English as secondary.
  6. Saudi-ready — PDPL consent checks, ZATCA invoicing, SDAIA AI governance, and NCA cybersecurity controls are implemented with live enforcement on the golden path and Saudi workflow. Full production coverage is tracked in docs/current-vs-target-register.md.
  7. Board-usable — Executive surfaces show what changed, what needs decision, what is at risk.
  8. Enterprise-saleable — Evidence packs, audit trails, and compliance matrices are exportable.

6. Non-Negotiable Rules

  1. Tenant isolation: Every query is scoped by tenant_id. Cross-tenant access is blocked at ORM layer.
  2. Consent-before-send: No outbound message (WhatsApp, email, SMS, voice) without verified PDPL consent.
  3. Audit everything: Every state change writes to audit_logs. Every AI decision writes to ai_conversations.
  4. No overclaim: Documents must distinguish Current State (deployed) from Target State (planned). Never claim what is not in production.
  5. Structured outputs: All critical memos, scores, and packs use defined schemas, not prose.
  6. Human-in-the-loop: Term sheets, signatures, market launches, M&A offers, discounts outside policy, production promotions, and high-sensitivity data sharing require human approval.
  7. Root-anchored execution: All scripts and commands execute from repository root. scripts/architecture_brief.py is the official preflight.

7. Contradiction Resolution

When documents or systems conflict:

  1. MASTER_OPERATING_PROMPT.md wins over all other documents.
  2. Governance docs (docs/governance/*) win over operational docs.
  3. CLAUDE.md / AGENTS.md win over memory/ docs.
  4. Code behavior wins over comments about code behavior.
  5. Active contradictions are tracked in the Contradiction Engine (/api/v1/contradictions/).

8. Technology Radar Summary

Tier Technologies
Core (production) FastAPI, SQLAlchemy, PostgreSQL 16, Redis, Celery, Next.js 15, OpenClaw 2026.4.x, Groq, WhatsApp Cloud API
Strong (validated) Claude Opus, Salesforce Agentforce, Stripe, pgvector, Mem0, LangGraph
Pilot (behind flags) Voice agents, Contract intelligence, Gemini/DeepSeek routing
Watch (evaluating) Temporal, OPA, OpenFGA, Vault, Gong, Apollo
Hold (not adopting) External RAG SaaS, schema-per-tenant, GraphQL

Full specification: docs/governance/technology-radar-tier1.md

9. Document Index

Document Path Purpose
AI Operating Model docs/ai-operating-model.md Five-plane architecture
Six Tracks docs/dealix-six-tracks.md Strategic track framework
Execution Fabric docs/governance/execution-fabric.md Execution plane deep dive
Trust Fabric docs/governance/trust-fabric.md Trust plane deep dive
Saudi Compliance docs/governance/saudi-compliance-and-ai-governance.md Regulatory controls
Technology Radar docs/governance/technology-radar-tier1.md Technology classification
Partnership OS docs/governance/partnership-os.md Partnership lifecycle
M&A OS docs/governance/ma-os.md Corporate development
Expansion OS docs/governance/expansion-os.md Geographic/vertical expansion
PMI OS docs/governance/pmi-os.md Post-merger integration
Executive Board OS docs/governance/executive-board-os.md Board reporting framework
90-Day Matrix docs/execution-matrix-90d-tier1.md Sprint execution plan
ADR 0001 docs/adr/0001-tier1-execution-policy-spikes.md Tier-1 policy decisions
Current vs Target docs/current-vs-target-register.md Subsystem maturity register
Doc Consistency Audit docs/governance/document-consistency-audit.md Cross-reference verification
Structured Outputs backend/app/schemas/structured_outputs.py 17 Pydantic decision schemas
Workflow Inventory docs/governance/workflow-inventory.md Short/medium/long classification
Trust Closure Plan docs/governance/trust-closure-plan.md Trust plane completion gates
Connector Standard docs/governance/connector-standard.md Connector facade + metrics
Operating Checklist docs/governance/operating-plane-checklist.md Enterprise delivery controls
Saudi Readiness docs/governance/saudi-enterprise-readiness.md PDPL/NCA/SDAIA operationalization
Executive Surface Plan docs/governance/executive-surface-closure.md Surface wiring plan
Market Dominance docs/governance/market-dominance-plan.md Packaging + ROI + competitive wedge
Master Closure Checklist docs/tier1-master-closure-checklist.md 50-item definitive checklist
Architecture docs/ARCHITECTURE.md System diagram
Data Model docs/DATA-MODEL.md Database schema
Agent Map docs/AGENT-MAP.md 19 AI agents
API Map docs/API-MAP.md 70+ endpoints

10. Enforcement

This document is enforced by:

  • scripts/architecture_brief.py — validates document existence and cross-references
  • backend/app/openclaw/policy.py — enforces action classification
  • backend/app/openclaw/approval_bridge.py — enforces approval gates
  • .github/workflows/dealix-ci.yml — runs tests and checks on every PR
  • Contradiction Engine — detects and tracks document/system conflicts